This GigaOm Research Reprint Expires November 16, 2027
November 11, 2026

GigaOm Radar for Service Mesh v6

Ivan McPhee

1.
Market Context

1. Market Context

A service mesh moves service-to-service communication out of application code and into infrastructure. A control plane issues workload identities and policy. A data plane of sidecar proxies, shared node-level proxies, or kernel-level programs encrypts, routes, and observes traffic. Some products offer L7 functions only through optional components.

Most products in this Radar build on a small number of graduated upstream Cloud Native Computing Foundation (CNCF) projects: Istio, Linkerd, and Cilium. What separates the field is where mesh functions run, which security defaults apply out of the box, how far the mesh extends beyond Kubernetes, and who does the work of keeping it current. Buyers are choosing an architecture and an operating model as much as a feature set. Five themes capture where the field separates.

Upstream Projects Set the Terms of Competition

Most commercial products here are distributions or enterprise editions of a foundation-hosted open-source project, and the projects themselves compete alongside them. A managed cloud service and a few independently developed platforms make up the rest. The most widely adopted project anchors much of the field, and the managed cloud service uses its APIs for configuration as well.

Where products share a base, commercial value shifts from mesh features to the release line itself: long-term CVE patching, FIPS builds, tested platform matrices, multicluster tooling, and a vendor accountable for incidents. Lineage also sets the pace a buyer inherits. One distribution holds an upstream base several minor releases behind the current line, while others ship in step with upstream or within one release of it.

What this means for buyers: Decide on the upstream project before the vendor. Once you choose the base, compare distributions on what they add: the support window, patch commitments, FIPS and platform coverage, and how far the release line trails upstream. A distribution that stays on an older upstream release trades new capability for predictability, so make that tradeoff explicit in the contract.

The Data Plane Splits Three Ways

A proxy alongside each workload remains the production norm across most of the field, whether a full Envoy sidecar, a purpose-built Rust microproxy, or proxyless gRPC libraries for some clients. A few products run ambient and sidecar data planes side by side at general availability: node-level proxies carry L4 identity and encryption, waypoint proxies add L7 processing where needed, and namespaces move between modes reversibly.

A few more are sidecarless by design. Most of these handle traffic with eBPF programs in the node kernel and a shared per-node proxy, and one routes traffic through site-level nodes that the vendor manages as a service. Some sidecar products list ambient or eBPF modes at alpha or experimental status, and another has removed its experimental eBPF path.

What this means for buyers: Pick the data plane model first: it decides resource cost, upgrade mechanics, and which products remain options. If ambient is the target, ask whether sidecar and ambient workloads coexist at general availability and whether migration is reversible per namespace. If the mesh runs inside the network layer, confirm which security features are generally available there.

Secure by Default Is Still a Configuration Choice

Every product encrypts service traffic, but defaults differ sharply. Some products install with strict mTLS and refuse plaintext unless an operator declares an exception. Others encrypt meshed traffic automatically but accept plaintext from unmeshed sources until an operator tightens policy. The most widely adopted project and some of its distributions ship permissive inbound defaults, which makes strict enforcement an operator task.

In the kernel-based products, mesh-managed mTLS and SPIFFE mutual authentication remain in beta, and transport encryption comes from WireGuard or IPsec instead. Cryptography is moving quickly. A small group negotiates hybrid post-quantum key exchange by default, and others offer it once configured or in a release candidate. FIPS paths range from validated builds to cipher restrictions on hosts running in FIPS mode.

What this means for buyers: Audit defaults, not datasheets. Ask what a fresh installation enforces before anyone touches it, how the mesh handles plaintext from unmeshed workloads, and whether workload certificates chain to your enterprise CA. Regulated buyers should confirm which validated cryptographic module covers the data plane and whether post-quantum key exchange works in FIPS mode.

The Mesh Reaches Past Kubernetes, Unevenly

Several products enroll VMs, bare-metal hosts, and Amazon ECS tasks as first-class mesh members, with the same discovery, identity, and policy as Kubernetes services. These products also join zones, clusters, and data centers under a global control plane. Others add VMs to the mesh through mesh expansion features while keeping Kubernetes as their primary platform. At the other end, one product runs only on its parent Kubernetes platform, and the kernel-based products stay within Kubernetes clusters.

Multicluster designs vary just as much. Some run independent control planes per cluster that exchange scoped service information, some distribute policy from one management plane to per-cluster instances, some rely on a shared trust anchor, and some copy policy objects to each cluster.

What this means for buyers: Inventory your estate before you build a shortlist. Count the workloads that run outside Kubernetes today and will still be there in three years, and test the mesh against them, not only against pods. For multicluster designs, ask what happens to cross-cluster traffic, identity, and new configuration when one cluster or one control plane fails.

AI Traffic Arrives Mostly Through Companion Products

Agent gateways, AI gateways, and agent routers carry most of the governance of AI agent and model traffic. Vendors sell them as companion products rather than inside the mesh SKU. They add token metering, model-aware routing, and prompt filtering, and in some cases run inside the mesh as a waypoint.

The upstream reference project adds model-aware routing for inference traffic through a Gateway API extension. One specialist platform registers MCP agents as mesh services with workload identities, allow-lists the tools each agent may call, inspects prompts for injection and PII, and sets token budgets per team. Another product includes AI service roles only in a release candidate, not yet in a production release. The capability is real but early, and the packaging is unsettled.

What this means for buyers: Separate the mesh decision from the AI gateway decision unless one product truly delivers both. Ask which AI controls ship in the mesh SKU, which require a companion product and license, and whether agent identity reuses the mesh's workload identity. Treat release-candidate and preview AI features as roadmap items until they reach a production release.

2.
Radar Graphic

2. Radar Graphic

Inline Image

Figure 1. GigaOm Radar for Service Mesh. Inner ring denotes Leaders and outer ring denotes Challengers; dot color carries market velocity.

Notable Patterns

The paragraphs below describe how this market populates each quadrant. Section 3, Tech Buying Triad, gives the house definitions of the quadrants.

Foundation Builders. These products treat the mesh as dependable infrastructure for estates that cannot absorb disruption. Products here run sidecar data planes in production and sell continuity: long-lived release lines with LTS windows, validated platform matrices, staged CA rotation, and governance layered over many clusters and runtimes, including VMs and container services outside Kubernetes. Ambient operation is absent or still early, and some products in this quadrant hold a deliberately conservative upstream base. Change arrives as hardening and lifecycle work rather than new architecture.

Agile Optimizers. The largest group improves a workflow buyers already run, release by release. The improvements are operational: lighter per-pod proxies, post-quantum key exchange switched on by default, managed control planes with gradual proxy upgrades, ambient operation delivered inside an existing platform subscription, and mesh features built into the cluster's networking layer. Products here ship often and ask the buyer to keep pace, but none asks for a wholesale rearchitecture to get the next gain.

Disruptive Pioneers. Products here change where mesh functions live. One enforces identity-aware policy in kernel eBPF programs and hands L7 traffic to a shared per-node proxy. Another delivers the mesh as a site-level fabric managed from a SaaS console, and a third extends an accredited zero-trust platform to AI agent and MCP traffic. Each trades some conformance with mesh standards, such as Gateway API routing for east-west traffic or generally available mesh mTLS, for a model the sidecar meshes do not offer.

Enduring Innovators. The smallest group treats ambient operation as the long-horizon target and makes the path there safe. Ambient and sidecar workloads coexist in one mesh at general availability, and services move between modes incrementally and reversibly. Multicluster ambient matures release by release toward a single mesh across clusters and clouds. The value compounds as more of the estate moves onto the new data plane, while vendors continue to support the sidecar model for as long as buyers need it.

3.
Tech Buying Triad

3. Tech Buying Triad

The quadrant descriptions and decision framework help buyers read this Radar from an enterprise procurement perspective. The Radar plots solutions on two axes: Operational Focus versus Strategic Focus, and Stability-Centered versus Evolution-Centered. Placement therefore signals both the outcome a product aims to deliver and how it approaches change.

The Tech Buying Triad maps decision responsibility: the CIO owns Brown Money tradeoffs, the CTO owns Green Money tradeoffs, and the CISO owns GRCS tradeoffs. The Triad treats Emerging Technology inputs as consulted advice. In service mesh, all three roles hold a real stake, and misalignment among them is a common reason deployments stall at the pilot stage.

Role

Bucket

What the bucket means

Responsibilities in the procurement

CIO

Brown Money

Operational and infrastructure value: capabilities tied to operational reliability, cost containment, and table-stakes function.

Resource Efficiency and Data Plane Overhead, Sidecarless / Ambient / eBPF Architecture Support, Sidecar-to-Ambient Migration and Coexistence, Control Plane Scalability, Control Plane Resilience and Blast Radius, Operational Lifecycle and Upgrades, Platform and Distribution Support, Day-2 Tooling and Stack Integration.

CTO

Green Money

Innovation and growth value: capabilities tied to architectural durability, interoperability, and competitive differentiation.

Multi-Cluster and Hybrid Cloud Management, Non-Kubernetes and Multi-Runtime Support, Multiprotocol Support, Progressive Delivery and Advanced Traffic Control, Resilience and Circuit Protection Depth, Extensibility and Programmability, Gateway API / GAMMA Conformance, Load Balancing and Locality Routing.

CISO

GRCS

Governance, risk, compliance, and security: the controls required to operate the technology safely and within policy.

mTLS Enforcement Posture, Workload Identity and Federation, Enterprise PKI and CA Integration, L7 Authorization Granularity, FIPS and Regulated-Environment Cryptography, Audit Logging and Compliance Reporting, Supply Chain and Patch Integrity, Policy Rollout Safety.

All three (consulted)

Emerging Technology

Forward-looking capabilities that signal where the market is moving; not owned by any single role.

Input on AI and Agentic Traffic Governance, eBPF Data Plane Maturity, AI-Assisted Policy and Operations, WebAssembly Extensibility Maturity, Ambient Multicluster Maturity, Post-Quantum Cryptography, and Runtime Security and Observability Convergence, to inform tradeoffs across Brown Money, Green Money, and GRCS.

Quadrants

Foundation Builders

The buyer is typically a CIO or a platform engineering lead running large, established Kubernetes and VM estates where a control plane outage or a disruptive upgrade is expensive. The purchase is operational confidence: mTLS and workload identity that do not require rearchitecting the data plane every eighteen months, traffic management that behaves consistently across Kubernetes versions, long support windows, and conservative, documented upgrade paths. Change is the risk in this quadrant, so buyers judge vendors here on continuity before anything else.

Agile Optimizers

The buyer here is a platform engineering team or SRE group that treats the mesh as a living system and is comfortable adopting releases on a regular cadence. The intent is still operational: lower resource overhead, simpler day-2 operations, faster multicluster failover, and better telemetry on the workflow already in production. This buyer expects measurable improvement from each release and funds the work from platform budgets rather than a transformation program. Vendors land here when each release removes friction from running the mesh, without asking the buyer to adopt a new architecture to get it.

Enduring Innovators

The buyer is a CTO or platform architect who expects the mesh to become the foundation of a zero-trust architecture or a multicloud application platform. Over several years, the buyer requires a predictable roadmap, sustained investment, and enterprise support commitments around an architecture whose value compounds as more of the estate moves onto it. These buyers accept measured change in exchange for depth: federated control, enterprise PKI and SIEM integration, FIPS-validated builds, and a supported migration path to the target data plane.

Disruptive Pioneers

The buyer profile for this quadrant is a CTO or CISO with strong platform engineering capability who wants architectural advantage now and will accept a faster-moving product surface to get it. The bet is to move mesh functions somewhere new: the node kernel, a managed network fabric, or a zero-trust platform that governs AI agents as well as services. These buyers tolerate gaps in standards conformance and maturity, and they have the expertise to work around them.

Axes

The X-axis: Operational Focus versus Strategic Focus describes the value a product aims to deliver, with Operational placements on the left of the chart and Strategic placements on the right. Buyers choose solutions toward the Operational side to secure and stabilize the services an organization already runs, minimizing operational risk and the effort of running the mesh day to day. Buyers choose solutions toward the Strategic side to change how the organization builds and connects distributed systems, with value that compounds through architectural commitments over several years.

The Y-axis: Stability-Centered versus Evolution-Centered describes change posture, with Stability placements in the upper half of the chart and Evolution placements in the lower half. Stability-centered placements favor long support windows, conservative deprecation, and controlled upgrade cycles. For them, change is the risk to minimize. Evolution-centered placements treat change as the benefit buyers are paying for: frequent releases that extend what the mesh can do, or an architecture that redefines what a mesh is.

Ring System (Leader Versus Challenger)

The Radar's concentric rings communicate overall capability depth and maturity independent of quadrant. This Radar uses two rings: Leader, the inner ring, and Challenger, the outer ring. Six products hold the inner ring, and nine hold the outer ring.

Leader denotes products that combine broad capability across the evaluation lenses, production-grade security and lifecycle engineering, and enough documentation and ecosystem integration to be referenceable at scale.

Challenger denotes solid capability with important areas of growth: useful in many deployments, but likely to need more integration effort, operational validation, or workarounds for specific enterprise requirements. In this field, Challenger placements often reflect a narrower data plane model, support for fewer runtimes, or security features that have not yet reached general availability.

Use ring position as a procurement heuristic. Inner-ring placements reduce integration and validation effort when the buyer's requirements align with the product's strengths. Outer-ring placements call for more architectural validation, clearer commitments during procurement, and often a pilot. Ring position does not change which quadrant a product targets.

Archetype Variants

Different buyer archetypes shift priorities across the Tech Buying Triad's three roles. Compliance-bound buyers in defense, government, and financial services tilt toward GRCS. For them, strict mTLS by default, FIPS-validated cryptography, enterprise CA chaining, audit records tied to workload identity, and patch integrity dominate. The CISO's accountabilities lead the procurement.

Platform-led buyers building multicluster and multicloud estates shift toward Green Money and CTO leadership, weighting multicluster management, non-Kubernetes reach, Gateway API conformance, and extensibility. Cost-pressured buyers elevate Brown Money and the CIO, focusing first on data plane overhead, the sidecarless option, and upgrade mechanics.

Succeeding with any selection requires explicit Triad alignment up front. Agree which bucket has the final say when priorities conflict. Then define the acceptance tests before the mesh carries production traffic: a control plane outage under load, a canary upgrade of the data plane, and a strict-mTLS cutover for a namespace with unmeshed callers.

Decision Matrix — Which Quadrant Fits Your Organization?

Use the per-quadrant profile below to self-identify the quadrant whose buyer most resembles your organization. Read the row that sounds like your team. That quadrant's vendor capsules deserve your closest reading.

Quadrant

You are likely this buyer if…

Start your shortlist with

Foundation Builders

You are a CIO or platform engineering lead running a large, established Kubernetes and VM estate. You need mTLS, identity, and traffic management that stay predictable across long support windows, with conservative upgrades and governance over many clusters and runtimes.

F5 Aspen Mesh, HashiCorp (IBM), Kong, Tetrate

Agile Optimizers

You lead a platform engineering or SRE team that adopts releases on a regular cadence. You want each release to cut proxy overhead, simplify day-2 operations, or improve multicluster behavior on the mesh you already run, without a rearchitecture.

Buoyant, Cilium (CNCF), Google Cloud, Kuma (CNCF), Linkerd (CNCF), Red Hat

Enduring Innovators

You are a CTO or platform architect making a multiyear bet on the mesh as the foundation of zero-trust and multicloud architecture, and you want a supported, reversible path to ambient operation across clusters.

Istio (CNCF), Solo.io

Disruptive Pioneers

You are a CTO or CISO with deep platform skills who wants mesh functions in a new place, such as the kernel, a managed fabric, or an accredited zero-trust platform that also governs AI agents, and you accept gaps in standards conformance to get there.

F5 Distributed Cloud Mesh, Greymatter.io, Isovalent

4.
Vendor Capsules by Quadrant

4. Vendor Capsules by Quadrant

Vendor Capsules by Quadrant

Enduring Innovators

These vendors suit a CTO or platform architect who is making a multiyear bet on the mesh as the base for zero trust or a multicloud application platform. Buyers trade fast change for depth: a predictable roadmap, steady investment, enterprise-grade security integrations, and a supported route to their target architecture.

Brown Money
Vendor
Control Plane Resilience and Blast Radius
Control Plane Scalability
Day-2 Tooling and Stack Integration
Operational Lifecycle and Upgrades
Platform and Distribution Support
Resource Efficiency and Data Plane Overhead
Sidecar-to-Ambient Migration and Coexistence
Sidecarless / Ambient / eBPF Architecture Support
Average
Istio (CNCF)2.54.54.53.55.05.04.54.54.3
Solo.io3.05.04.53.55.05.05.05.04.5
Green Money
Vendor
Extensibility and Programmability
Gateway API / GAMMA Conformance
Load Balancing and Locality Routing
Multi-Cluster and Hybrid Cloud Management
Multiprotocol Support
Non-Kubernetes and Multi-Runtime Support
Progressive Delivery and Advanced Traffic Control
Resilience and Circuit Protection Depth
Average
Istio (CNCF)4.05.04.54.04.03.54.04.54.2
Solo.io4.05.04.55.04.05.04.05.04.6
GRCS
Vendor
Audit Logging and Compliance Reporting
Enterprise PKI and CA Integration
FIPS and Regulated-Environment Cryptography
L7 Authorization Granularity
Policy Rollout Safety
Supply Chain and Patch Integrity
Workload Identity and Federation
mTLS Enforcement Posture
Average
Istio (CNCF)3.54.53.04.54.04.05.04.54.1
Solo.io3.04.03.55.04.03.53.54.03.8
Emerging Tech
Vendor
AI and Agentic Traffic Governance
AI-Assisted Policy and Operations
Ambient Multicluster Maturity
Post-Quantum Cryptography
Runtime Security and Observability Convergence
WebAssembly Extensibility Maturity
eBPF Data Plane Maturity
Average
Istio (CNCF)4.53.53.53.54.03.54.03.8
Solo.io5.04.55.04.04.53.54.04.4

Istio (CNCF)

Istio*

Leader | Enduring Innovators | Outpacing

Summary:

Istio is an open-source service mesh, licensed under Apache 2.0, that the CNCF hosts as a graduated project. It installs into the customer's own Kubernetes clusters on any major cloud or on premises, and virtual machines can join as mesh workloads. The buyer runs the infrastructure.

There is no license fee and no single commercial vendor. The community patches each minor release until six weeks after the second release that follows it, and longer-term support comes from third-party distributions. The project targets platform teams building on upstream Kubernetes.

Assessment:

Istio is the upstream project behind several commercial meshes, so what a buyer gets depends on whether it runs Istio directly or through a distribution. If the buyer runs Istio directly, it owns installation, upgrades, and incident response, and must keep pace with a roughly quarterly release cadence. Canary control plane revisions and upgrade prechecks make that cadence manageable.

Ambient mode, generally available (GA) since Istio 1.24, removes the per-pod proxy. A node-level ztunnel handles mutual TLS (mTLS) encryption and L4 policy, and waypoint proxies add L7 processing only where workloads need it. Sidecar and ambient workloads coexist in one mesh, and namespaces can move between the two modes and back. A buyer can therefore cut proxy overhead without a single, all-at-once migration.

Gateway API support for ingress and mesh routing is stable, including GRPCRoute and BackendTLSPolicy. The Gateway API Inference Extension adds model-aware routing for AI model-serving traffic.

Several capabilities are not yet generally available. Ambient multicluster is in beta, WebAssembly (Wasm) extensions and the TrafficExtension API are in alpha, and the post-quantum key exchange policy is experimental.

If the Istiod control plane becomes unreachable, proxies keep serving traffic on their last-known-good configuration, and workload certificates stay valid for a lifetime the operator sets. Beyond that, resilience depends on running several Istiod replicas spread across zones. Inbound proxies default to permissive mTLS, so the operator must apply a setting to enforce strict mTLS.

Platform teams with the Kubernetes expertise to operate a mesh themselves, or that buy a supported distribution, get the full upstream feature set, including ambient mode, as each release ships.

Solo.io

Solo Enterprise for Istio*

Leader | Enduring Innovators | Outpacing

Summary:

Solo Enterprise for Istio, formerly Gloo Mesh, is Solo.io's commercial Istio offering. It builds on Solo's hardened distribution of Istio and supports both ambient (sidecarless) and sidecar data planes. It installs into the customer's own Kubernetes and OpenShift clusters, including air-gapped environments, and extends the mesh to Amazon ECS, Amazon EC2, and VM workloads.

Licensing is tiered. Basic adds long-term CVE patching and FIPS builds, Premium adds faster vendor response, and Enterprise adds multicluster support, the Solo UI, and enterprise features. Pricing is by quote. Solo.io aims it at platform teams running Istio across many clusters.

Assessment:

Solo Enterprise for Istio runs ambient and sidecar data planes side by side, both generally available. Moving workloads between them is a supported operation rather than a rebuild. A migration tool assesses readiness, generates waypoint and policy configuration, and moves services incrementally and reversibly while traffic and policy remain in force. An existing sidecar estate can therefore gain ambient's lower per-pod cost without a cutover event.

Each cluster runs its own istiod control plane and shares scoped service information with its peers, an approach Solo.io has tested at large scale across many clusters and high pod counts. Global services provide automatic cross-cluster discovery, mTLS routing, and locality failover across ambient and sidecar meshes. The same identity and policy extend to ECS, VMs, and serverless functions. Only the Enterprise tier includes this breadth, so buyers on Basic or Premium get a hardened single-cluster Istio.

Strict mTLS across the mesh is a policy the operator applies, not a default. Certificate revocation enforcement is in alpha. Workload certificates chain to an enterprise intermediate CA through cert-manager or AWS Private CA.

If istiod becomes unavailable, proxies keep their last configuration. In ambient deployments that use Solo's SPIRE integration, ztunnel also continues to obtain workload identities from the local SPIRE agent. For AI traffic, the companion product Solo Enterprise for agentgateway adds token metering, routing across model providers, and prompt filtering.

This is the closest fit for organizations standardizing on Istio across many clusters and mixed runtimes that want a supported path from sidecars to ambient and a single vendor behind the distribution.

Disruptive Pioneers

These vendors suit a CTO or CISO with a strong platform engineering team who wants an architectural advantage now and will accept a faster-moving product to get it. They move mesh functions somewhere new, such as the node kernel, a managed network fabric, or a zero-trust platform that also governs AI agents. Their buyers have the skills to work around gaps in maturity and standards.

Brown Money
Vendor
Control Plane Resilience and Blast Radius
Control Plane Scalability
Day-2 Tooling and Stack Integration
Operational Lifecycle and Upgrades
Platform and Distribution Support
Resource Efficiency and Data Plane Overhead
Sidecar-to-Ambient Migration and Coexistence
Sidecarless / Ambient / eBPF Architecture Support
Average
F5 - F5 Distributed Cloud Mesh3.53.02.54.04.54.03.53.53.6
Greymatter.io4.05.05.04.55.03.53.54.04.3
Isovalent3.55.04.03.54.54.53.53.54.0
Green Money
Vendor
Extensibility and Programmability
Gateway API / GAMMA Conformance
Load Balancing and Locality Routing
Multi-Cluster and Hybrid Cloud Management
Multiprotocol Support
Non-Kubernetes and Multi-Runtime Support
Progressive Delivery and Advanced Traffic Control
Resilience and Circuit Protection Depth
Average
F5 - F5 Distributed Cloud Mesh2.51.03.04.52.54.03.05.03.2
Greymatter.io4.53.04.55.05.04.55.05.04.6
Isovalent3.55.03.03.54.03.03.54.53.8
GRCS
Vendor
Audit Logging and Compliance Reporting
Enterprise PKI and CA Integration
FIPS and Regulated-Environment Cryptography
L7 Authorization Granularity
Policy Rollout Safety
Supply Chain and Patch Integrity
Workload Identity and Federation
mTLS Enforcement Posture
Average
F5 - F5 Distributed Cloud Mesh3.52.53.03.53.03.02.52.52.9
Greymatter.io5.05.05.05.05.04.55.05.04.9
Isovalent3.5—3.03.55.04.01.0—2.5
Emerging Tech
Vendor
AI and Agentic Traffic Governance
AI-Assisted Policy and Operations
Ambient Multicluster Maturity
Post-Quantum Cryptography
Runtime Security and Observability Convergence
WebAssembly Extensibility Maturity
eBPF Data Plane Maturity
Average
F5 - F5 Distributed Cloud Mesh3.55.05.01.04.52.02.53.4
Greymatter.io5.04.53.55.05.04.03.54.4
Isovalent3.04.05.0—4.03.55.03.5

F5

F5 Distributed Cloud Mesh

Challenger | Disruptive Pioneers | Fast Following

Summary:

F5 Distributed Cloud Mesh is the service mesh and application networking service within F5 Distributed Cloud Services. F5 delivers it as SaaS, managed from the Distributed Cloud Console. Traffic runs through shared F5 Distributed Cloud Nodes that the customer deploys at its own cloud, data center, and edge sites, optionally supplemented by F5's regional points of presence.

F5 licenses Customer Edge mesh nodes by size (4, 8, or 16 vCPUs), either by subscription or pay-as-you-go through AWS Marketplace. Other pricing is by quote. F5 aims it at enterprises connecting and securing applications across multiple clouds and sites.

Assessment:

F5 Distributed Cloud Mesh routes service traffic through shared proxies on site-level nodes rather than through a proxy alongside each workload. It works more like a distributed application delivery fabric than a Kubernetes-native mesh.

Workloads join the mesh by label and carry no per-pod proxy. The same nodes provide multicluster discovery, routing, and unified policy from one console. Because there is a single data plane model, there is no sidecar-to-ambient migration to plan.

Origin pools enforce connection, request, and retry circuit limits, with endpoint ejection and active health checks. Endpoint health is shared globally, so traffic fails over to remote sites when local endpoints are unavailable. Disconnected sites keep routing traffic, issue certificates locally, and run on cached secrets, with three-node high availability. For a buyer spread across clouds and edge locations, failover is built into the fabric rather than configured cluster by cluster.

Operators configure F5's own model of virtual hosts, routes, clusters, and endpoints in the console, not Gateway API resources. Request customization runs through a V8 JavaScript engine rather than Wasm or Envoy filters.

By default, traffic between sites and to Regional Edges runs over tunnels authenticated with F5-issued identities. By contrast, operators enable mTLS on the local hop from proxy to workload per load balancer or origin pool rather than mesh-wide. Operators import enterprise CA certificates or pull them from external secret stores such as HashiCorp Vault, rather than having the organization's own CA issue them directly.

F5 states control plane capacity as per-tenant object limits for load balancers, origin pools, and sites. Teams standardizing on upstream Kubernetes APIs, OpenTelemetry tracing, or strict-by-default mTLS to each workload will need to do that work themselves.

This suits organizations already invested in F5 that want one managed fabric for connecting, securing, and observing applications across clouds, data centers, and edge sites. These buyers also prefer a vendor-operated control plane to running mesh infrastructure themselves.

Greymatter.io

Greymatter

Leader | Disruptive Pioneers | Outpacing

Summary:

Greymatter is a zero-trust service connectivity platform from Greymatter.io. It manages Envoy proxies, gateways, and waypoints, along with mTLS, workload identity, and audit, across applications, APIs, and AI workloads. It deploys into the customer's environment on AWS, Azure, Google Cloud, Oracle Cloud, Kubernetes distributions, VMs, and edge sites, including air-gapped environments.

Pricing is by quote. Buyers can purchase it directly, through Carahsoft, or on public-sector contract vehicles such as NASA SEWP V, and a 30-day trial is available. Greymatter.io aims it at defense, national security, financial services, and telecommunications organizations.

Assessment:

Greymatter is built for regulated and disconnected operation. Mesh mTLS is strict by default, and the mesh refuses plaintext traffic unless an operator declares an exception. CRL and OCSP checks reject revoked certificates on every connection, and the cryptography carries FIPS 140-3 validation.

Hybrid post-quantum key exchange (X25519MLKEM768) is on by default from version 2.5. Installation, updates, and operation all work without network connectivity. For defense and classified programs, this removes compliance work a buyer would otherwise have to build around a general-purpose mesh.

Day-2 operations run through Git. Operators declare configuration in a repository or as Kubernetes custom resources. The platform reconciles the running state against that configuration and reports drift and health.

Every proxied transaction emits an event that records actor and workload identities. These events stream with certificate telemetry to SIEM, SOAR, and XDR tools. SPIRE issues SPIFFE identities across containers, VMs, bare metal, and Windows hosts, so systems outside Kubernetes follow the same policy model.

Version 2.4 applies the same model to AI agents. MCP agents register as services with workload identities, and operators allow-list the tools each agent may call. The platform inspects prompts and responses for injection attacks and personally identifiable information (PII), and operators set token budgets per team.

The data plane remains a proxy alongside each meshed service. A CNI plugin redirects traffic without per-application port configuration, and operators can set each workload's proxy CPU and memory. Workloads can also run without a proxy in the pod, behind a shared Greymatter edge, and move between the two models one workload or namespace at a time. However, workloads fronted only by the edge do not carry mesh mTLS.

Operators configure routing in Greymatter's own GSL language or in Kubernetes custom resources. Gateway API ingress is generally available on OpenShift, but east-west routing stays in GSL and custom resources rather than Gateway API.

Greymatter is a natural fit for public-sector, defense, and financial organizations that must run service connectivity in accredited or air-gapped environments and want enforcement, audit, and oversight of AI agents from one vendor. Teams that prioritize sidecarless efficiency or Kubernetes-native routing APIs for east-west traffic should weigh those needs against its compliance depth.

Isovalent

Isovalent Enterprise Platform*

Challenger | Disruptive Pioneers | Fast Following

Summary:

Isovalent Enterprise Platform is the commercial distribution of the open-source Cilium, Hubble, and Tetragon projects. It provides eBPF-based networking, service mesh, observability, and runtime security. Isovalent is part of Cisco.

The platform installs into the customer's own clusters on upstream Kubernetes, AWS, Azure, Google Cloud, OpenShift, and Rancher-based environments, and it extends to virtual machines running on Kubernetes. Isovalent sells it as an enterprise subscription spanning Networking for Kubernetes, Networking for Virtualization, Load Balancer, and Runtime Security, with pricing by quote. Isovalent aims it at platform and network teams standardizing on Cilium.

Assessment:

Isovalent approaches the mesh from the network layer. Cilium enforces identity-based L3 and L4 policy in kernel eBPF programs and hands traffic to a shared Envoy proxy on each node when traffic needs L7 handling, so no workload carries its own proxy. The same design supports sidecarless cross-cluster operation through Cluster Mesh and stable Gateway API mesh routing for HTTP and gRPC.

There is no sidecar mode. Instead, workloads move in place between the in-kernel L4 path and the per-node Envoy L7 path, one workload or namespace at a time. Node-by-node CNI migration runs the old and new data planes side by side, with a rollback path.

Mesh-managed mTLS is not generally available. Cilium's SPIFFE-based mutual authentication and the newer ztunnel encryption are both in beta, and mutual authentication is deprecated. As a result, certificate-based workload identity and chaining to an enterprise CA are not part of the GA product.

IPsec or WireGuard delivers transport encryption instead, including a FIPS path over IPsec. Buyers who require mesh mTLS and SPIFFE identity will need to plan around this gap.

Isovalent's enterprise tooling turns network policy into a managed workflow. Isopolicy simulates proposed policy against live cluster data before rollout, and Hubble Timescape generates editable policy from observed traffic flows. Hubble also exports dropped and denied traffic, attributed to each workload, to SIEM pipelines.

Traffic management covers weighted and header-based canary routing, request mirroring, and fault injection. Argo Rollouts and Flagger can drive canary releases through Gateway API route weights, although neither is a supported Isovalent integration.

Organizations that have standardized on Cilium and want vendor-backed L7 visibility, policy simulation, and multicluster connectivity without sidecars get the most from Isovalent, provided network-layer encryption, rather than mesh certificates, meets their requirements.

Foundation Builders

These vendors treat the mesh as critical infrastructure for CIOs and platform leads running large, established Kubernetes and VM estates, where an outage or a disruptive upgrade is costly. Buyers judge them first on continuity: stable security and traffic management, long support windows, and careful, documented upgrade paths.

Brown Money
Vendor
Control Plane Resilience and Blast Radius
Control Plane Scalability
Day-2 Tooling and Stack Integration
Operational Lifecycle and Upgrades
Platform and Distribution Support
Resource Efficiency and Data Plane Overhead
Sidecar-to-Ambient Migration and Coexistence
Sidecarless / Ambient / eBPF Architecture Support
Average
F5 - F5 Aspen Mesh2.53.54.03.05.03.53.02.53.4
HashiCorp (IBM)3.05.04.53.55.03.03.53.03.8
Kong3.04.04.03.53.53.53.03.03.4
Tetrate3.53.54.53.55.03.03.02.03.5
Green Money
Vendor
Extensibility and Programmability
Gateway API / GAMMA Conformance
Load Balancing and Locality Routing
Multi-Cluster and Hybrid Cloud Management
Multiprotocol Support
Non-Kubernetes and Multi-Runtime Support
Progressive Delivery and Advanced Traffic Control
Resilience and Circuit Protection Depth
Average
F5 - F5 Aspen Mesh4.03.04.53.54.03.54.04.53.9
HashiCorp (IBM)4.53.04.05.02.54.54.04.54.0
Kong4.53.54.04.04.54.04.04.54.1
Tetrate5.02.54.04.03.53.54.04.53.9
GRCS
Vendor
Audit Logging and Compliance Reporting
Enterprise PKI and CA Integration
FIPS and Regulated-Environment Cryptography
L7 Authorization Granularity
Policy Rollout Safety
Supply Chain and Patch Integrity
Workload Identity and Federation
mTLS Enforcement Posture
Average
F5 - F5 Aspen Mesh3.53.53.05.04.03.53.54.03.8
HashiCorp (IBM)4.05.04.05.03.53.55.04.54.3
Kong4.54.03.05.04.05.03.04.04.1
Tetrate4.04.04.04.04.04.03.54.54.0
Emerging Tech
Vendor
AI and Agentic Traffic Governance
AI-Assisted Policy and Operations
Ambient Multicluster Maturity
Post-Quantum Cryptography
Runtime Security and Observability Convergence
WebAssembly Extensibility Maturity
eBPF Data Plane Maturity
Average
F5 - F5 Aspen Mesh3.03.53.01.04.03.53.53.1
HashiCorp (IBM)3.53.53.52.04.03.53.53.4
Kong4.53.03.51.03.53.03.53.1
Tetrate3.04.03.03.53.55.03.03.6

F5

F5 Aspen Mesh*

Challenger | Foundation Builders | Slow Following

Summary:

F5 Aspen Mesh is a supported distribution of the CNCF Istio project, with F5 additions such as Packet Inspector traffic capture, a certificate orchestrator, and IPv4/IPv6 dual-stack networking. It installs with Helm into the customer's own Kubernetes and OpenShift clusters, including air-gapped ones, and the buyer supplies the infrastructure.

The current line is Aspen Mesh Platform 1.21, based on Istio 1.21 and sold with 24/7 support. Pricing is by quote. F5 aims it chiefly at service providers building 5G core network functions and at regulated enterprises running on-premises or hybrid environments.

Assessment:

Aspen Mesh stays on Istio 1.21 and adds F5 patches and backports rather than following upstream releases. Upstream Istio, meanwhile, has moved on to the 1.29 to 1.31 lines. The data plane is the classic per-pod Envoy sidecar. Ambient mode and node-level proxies are not in the GA product, so workloads move between proxy revisions rather than between data plane models.

Gateway API covers ingress and Service-attached HTTPRoute mesh routing, while GRPCRoute is still experimental in Istio 1.21. In exchange, the buyer gets a long-lived release line. F5 shipped releases on it into 2026, each validated against named Kubernetes and OpenShift versions.

Security and traffic policy come from Istio. The mesh installs with strict mTLS, issues rotating SPIFFE workload identities, and authorizes requests based on identity and JWT claims, with delegation to external authorization services.

Locality-aware failover, circuit-breaker thresholds, and Flagger-driven canary promotion cover resilience and delivery. The bundled istioctl tool analyzes the running mesh and tests proposed configuration before an operator applies it. The istio-csr component ties workload certificates to an enterprise issuer through cert-manager.

The operations team runs upgrades, and no first-party Kubernetes operator runs staged rollouts. A bundled prerequisite script and istioctl precheck validate the cluster first. Application deployments and configuration changes then pause while operators upgrade CRDs, Istiod, and gateways in place. Operators then restart sidecar-bearing pods namespace by namespace, while proxies not yet restarted keep serving traffic.

A canary Istiod revision, inherited from Istio, can also run in parallel, with Helm release history as the rollback path. If Istiod fails, the mesh relies on Envoy keeping its last configuration and on Istiod replicas. The operator must set up failover targets and certificate grace periods.

Aspen Mesh suits service providers and regulated operators who want a vendor-backed, Istio-compatible sidecar mesh on infrastructure they operate. These buyers value a stable release line over access to ambient mode and newer upstream features.

HashiCorp (IBM)

Consul Enterprise*

Leader | Foundation Builders | Fast Following

Summary:

Consul Enterprise is HashiCorp's self-managed service networking platform, combining service discovery with an Envoy-based service mesh. HashiCorp is an IBM company. Customers run the servers and proxies on Kubernetes (including OpenShift, Tanzu, and the AWS, Azure, and Google Cloud managed services), VMs, bare metal, Nomad, and Amazon ECS.

HashiCorp sells Enterprise licenses through its sales team in two tiers: Self-Managed Standard for service discovery and Self-Managed Premium for service mesh. The free Community edition limits the mesh to one cluster. The managed HCP Consul Dedicated service reached end of life in November 2025.

Assessment:

Consul is a sidecar mesh that extends well beyond Kubernetes. One model for identity and intentions (Consul's service-to-service access rules) covers VMs, bare metal, Nomad, and Amazon ECS. Admin partitions, cluster peering, and WAN federation connect data centers with replicated mesh configuration. For organizations where much of the traffic still runs outside Kubernetes, that breadth is the main reason to consider Consul.

Consul treats identity and PKI as core mesh functions. Strict mTLS is the default at installation, workloads carry automatically rotated SPIFFE identities, and peering establishes trust with clusters that have their own independent roots.

External CAs such as Vault issue intermediate certificates that Consul renews automatically. During root rotation, Consul cross-signs the new root so trust holds throughout. Authorization combines identity, JWT claims, and external authorizers, and changes apply in Envoy without proxy restarts.

Every workload runs its own Envoy proxy. Consul Dataplane removes client agents, not proxies, so there is no sidecarless or eBPF mode. Workloads move in place between proxy modes, and from client agents to Consul Dataplane, one at a time without downtime.

L7 handling requires operators to declare services as HTTP or gRPC. The Gateway API implementation covers ingress but not GAMMA mesh routing.

During a control plane outage, proxies serve cached configuration for a certificate lifetime the operator sets, backed by server redundancy and secondary data centers. Post-quantum key exchange already protects control plane connections. Its use in mesh mTLS, along with AI agent roles, appears in the 2.1 release candidate but not yet in a production release.

Consul fits organizations that need one mesh across Kubernetes and non-Kubernetes runtimes, chained to their own enterprise CA, and that accept per-workload proxy overhead as the cost of that coverage.

Kong

Kong Mesh

Challenger | Foundation Builders | Fast Following

Summary:

Kong Mesh is Kong's enterprise service mesh, built on the CNCF Kuma project and Envoy. It runs on Kubernetes and, in Universal mode, on VMs and bare metal. Multizone deployments join clusters, clouds, and data centers under a global control plane that the customer hosts or that Kong runs in its Konnect SaaS platform. Workloads stay on customer infrastructure.

Kong sells Kong Mesh through Konnect plans. Plus is pay-as-you-go, Enterprise carries custom annual pricing, and Kong prices Konnect's hosted control plane per zone. Kong aims it at platform operators, often alongside Kong Gateway.

Assessment:

Every meshed workload in Kong Mesh runs its own kuma-dp and Envoy sidecar, and no node-level or ambient data plane has been released. Operators can size proxy resources per workload, but each pod still carries a proxy. Buyers who expect shared node proxies should note that Kong Mesh is a sidecar-based design. The eBPF option remains experimental and only redirects traffic into the sidecar.

An OPA agent inside each sidecar evaluates JWT claims, methods, headers, and paths on every request. It combines mesh-operator and service-owner policy into one decision and refreshes without proxy restarts.

Certificates can come from Vault, cert-manager, or AWS Private CA, and staged CA rotation avoids downtime. Kong backs this with software bills of materials (SBOMs) and SLSA provenance on current releases, plus a 15-day remediation deadline for critical CVEs. That gives security and compliance teams concrete evidence to audit.

Universal mode treats VMs and Amazon ECS tasks as full mesh members, with identity and policy from the mesh's control plane. Kafka is a supported mesh protocol type with L7 metrics. Operators govern AI traffic by deploying Kong AI Gateway inside the mesh, which adds token budgets, model-aware routing, and prompt filtering. Kong AI Gateway is a separate product from Kong Mesh.

If a zone control plane goes down, existing traffic keeps flowing, but new proxies cannot join and existing proxies receive no configuration updates. mTLS traffic stops once certificates expire, after a lifetime that operators set on the mesh CA backend (24 hours by default). Each zone therefore benefits from redundant, multi-instance control planes.

Traffic mirroring and fault injection are built in. Flagger promotes or rolls back canary releases automatically based on mesh metrics through its Kuma and Gateway API providers, although Kong Mesh does not analyze canaries itself. In-proxy extensions run through MeshProxyPatch, which inserts Lua or Wasm filters into the Envoy filter chain, rather than through a dedicated Wasm extension resource.

Organizations consolidating on Kong for both API gateway and mesh gain the most, particularly where per-request authorization and supply-chain assurance across Kubernetes and VMs matter.

Tetrate

Tetrate Service Bridge (TSB)*

Challenger | Foundation Builders | Slow Following

Summary:

Tetrate Service Bridge (TSB) is an enterprise service mesh platform built on Istio and Envoy. Configuration flows from TSB's management plane, through Tetrate's XCP layer, to an Istio instance in each cluster. It covers Kubernetes (including EKS, AKS, Google Cloud, and OpenShift), VMs, and bare-metal hosts on infrastructure the customer runs. A FIPS build is available.

Tetrate sells TSB under annual contracts. The AWS Marketplace listing is a five-cluster pack with premium support included, and other pricing is by quote. Tetrate aims it at platform and security teams running multitenant meshes across many clusters.

Assessment:

TSB is a governance layer over Istio rather than a separate mesh. Each cluster runs its own Istio instance as an independent failure domain. A single management plane defines routing and security policy once and distributes it to every cluster.

That model suits organizations with many clusters and teams. Workspaces and hierarchical policy let platform owners set guardrails that application teams inherit. VMs and bare-metal hosts join the mesh with the same kind of workload identity as pods.

Tetrate governs data-path extensions from a central catalog. TSB fetches Proxy-Wasm modules from registries, attaches them to chosen request phases, and distributes them to proxies. Alongside them, external authorization calls OPA-compatible engines on each request.

For buyers who need custom logic in the data path, extensions become managed, auditable assets rather than per-cluster patches. Regulated buyers also get a FIPS build backed by an active CMVP certificate and release notes that list both fixed and outstanding CVEs.

The production data plane is the Istio sidecar. Ambient mode, which moves L4 handling into shared node proxies, is not yet generally available, and TSB 1.14 lists it as alpha. Lower per-pod proxy cost and the sidecar-to-ambient transition are therefore still ahead for production environments.

Gateway API routing is in alpha in a secondary configuration model. L7 handling covers HTTP/1.1, HTTP/2, and gRPC, which TSB detects automatically when ports are undeclared, plus MongoDB. Other protocols, such as Kafka, travel as plain TCP. Model-aware and agent-aware routing policy is part of Tetrate's separate Agent Router product, not TSB.

TSB suits large enterprises that standardize on Istio across clusters, clouds, and data centers and want central policy, multitenancy, and governed extensions with commercial backing. It fits best where a sidecar data plane is acceptable for the near term.

Agile Optimizers

These vendors suit platform engineering and SRE teams that run the mesh as a living system and are happy to take regular releases. Each release should make the mesh cheaper, simpler, or more resilient to run, without forcing a move to a new architecture.

Brown Money
Vendor
Control Plane Resilience and Blast Radius
Control Plane Scalability
Day-2 Tooling and Stack Integration
Operational Lifecycle and Upgrades
Platform and Distribution Support
Resource Efficiency and Data Plane Overhead
Sidecar-to-Ambient Migration and Coexistence
Sidecarless / Ambient / eBPF Architecture Support
Average
Buoyant3.53.54.54.55.03.04.03.53.9
Cilium (CNCF)3.05.03.53.54.54.53.03.53.8
Google Cloud5.03.55.04.05.03.54.04.04.3
Kuma (CNCF)3.03.54.03.03.53.03.03.03.3
Linkerd (CNCF)3.03.54.03.53.53.04.03.53.5
Red Hat3.54.55.04.01.54.54.54.54.0
Green Money
Vendor
Extensibility and Programmability
Gateway API / GAMMA Conformance
Load Balancing and Locality Routing
Multi-Cluster and Hybrid Cloud Management
Multiprotocol Support
Non-Kubernetes and Multi-Runtime Support
Progressive Delivery and Advanced Traffic Control
Resilience and Circuit Protection Depth
Average
Buoyant2.04.55.03.53.54.04.05.03.9
Cilium (CNCF)4.05.03.03.04.03.04.04.03.8
Google Cloud4.04.04.54.02.53.54.04.53.9
Kuma (CNCF)3.54.04.54.04.54.04.04.54.1
Linkerd (CNCF)2.04.55.03.53.53.54.05.03.9
Red Hat4.55.04.04.03.53.04.04.54.1
GRCS
Vendor
Audit Logging and Compliance Reporting
Enterprise PKI and CA Integration
FIPS and Regulated-Environment Cryptography
L7 Authorization Granularity
Policy Rollout Safety
Supply Chain and Patch Integrity
Workload Identity and Federation
mTLS Enforcement Posture
Average
Buoyant3.54.05.03.05.04.53.03.53.9
Cilium (CNCF)4.0—2.55.02.04.01.0—2.3
Google Cloud4.54.03.04.54.03.03.54.03.8
Kuma (CNCF)3.02.03.02.53.53.53.54.03.1
Linkerd (CNCF)3.04.01.03.53.52.53.03.53.0
Red Hat4.53.54.04.54.04.53.04.54.1
Emerging Tech
Vendor
AI and Agentic Traffic Governance
AI-Assisted Policy and Operations
Ambient Multicluster Maturity
Post-Quantum Cryptography
Runtime Security and Observability Convergence
WebAssembly Extensibility Maturity
eBPF Data Plane Maturity
Average
Buoyant3.54.03.55.04.0—3.53.4
Cilium (CNCF)3.03.05.0—4.03.55.03.4
Google Cloud3.53.53.53.04.52.03.53.4
Kuma (CNCF)3.03.03.02.04.03.53.03.1
Linkerd (CNCF)3.54.03.55.04.5—3.53.4
Red Hat4.54.03.54.54.55.04.04.3

Buoyant

Buoyant Enterprise for Linkerd*

Challenger | Agile Optimizers | Outpacing

Summary:

Buoyant Enterprise for Linkerd (BEL) is the commercial distribution of the CNCF Linkerd service mesh from the company that created it. It runs a Rust microproxy alongside each workload in the customer's own Kubernetes clusters and is tested on EKS, GKE, AKS, Oracle OKE, Rancher RKE2, and OpenShift. It also extends to Linux VMs and bare-metal hosts, with Windows VM support in alpha.

Stable releases come with the Premium and Strategic plans. Strategic adds hotpatches and 24/7 enterprise support, and FIPS and Buoyant Cloud are available as add-ons. Pricing is by quote, and companies with fewer than 50 employees can run it for free.

Assessment:

BEL keeps a per-pod proxy and invests in making that proxy light and predictable rather than replacing it. Kubernetes-native sidecars are the GA default. Buoyant still supports classic sidecars, which operators can select per namespace or workload. There is no ambient, node-level, or eBPF data plane. The fixed-function proxy accepts no WebAssembly or custom filter code, so teams customize behavior through configuration and policy objects.

Zone-aware load balancing spreads traffic to other zones progressively and factors endpoint latency into federated services that span clusters. Hybrid ML-KEM-768 post-quantum key exchange is on by default for all meshed traffic. FIPS builds include an audit command that finds noncompliant proxies and a dashboard that shows each connection's cipher suite and CMVP certificate.

Workload certificates rotate every 24 hours and carry Linkerd service-account names inside Kubernetes. VMs and bare-metal hosts obtain SPIFFE identities from SPIRE after node and workload attestation. Multicluster deployments rely on a shared trust anchor rather than federating separate roots. The mesh accepts plaintext traffic from unmeshed sources until a policy disables it.

Control plane sizing guidance covers request rates and, based on a named production deployment, the number of meshed proxies per cluster. Two features ease day-2 work: policy that the mesh generates from live traffic and runs in audit mode, and an operator that upgrades in stages, control plane first and then the data plane.

For Kubernetes-centric teams that value low operational overhead, encryption on by default, and regulated-environment cryptography, BEL is a close match, provided they do not need sidecarless operation or in-proxy extensibility.

Cilium (CNCF)

Cilium Service Mesh

Challenger | Agile Optimizers | Fast Following

Summary:

Cilium Service Mesh is the mesh capability of Cilium, a graduated CNCF project that also serves as a Kubernetes CNI. It runs in the customer's own clusters on upstream Kubernetes, the major managed cloud services, and several enterprise distributions, and the buyer supplies the infrastructure.

The software is free under Apache 2.0, with its eBPF programs under GPL 2.0. Commercial support comes through enterprise distributions such as Cisco's Isovalent Enterprise Platform, sold in annual units per node. The project aims it at platform teams that want mesh functions from the networking layer they already run.

Assessment:

Cilium handles L4 mesh traffic with eBPF in each node's kernel and sends selected L7 traffic to a shared Envoy proxy on the node, so no workload pod carries a proxy. Workloads that need connectivity and identity-based L4 policy carry no per-pod proxy cost, and there is no sidecar mode to migrate from. Cluster Mesh extends the same sidecarless model across clusters with shared discovery, load balancing, and Hubble visibility, although each cluster receives its own copy of policy objects.

Mesh-managed mTLS is not part of the generally available product. The SPIFFE-based mutual authentication feature is in beta, and the project deprecated it in release 1.20. Its ztunnel successor is also in beta. Production workload encryption therefore relies on WireGuard or IPsec rather than per-workload certificates.

The same gap applies to enterprise CA chaining and post-quantum key exchange, which buyers with regulated encryption requirements would need to source elsewhere. For FIPS operation, the project removed the BoringCrypto-based FIPS build in 1.20. Operators can still restrict IPsec node-to-node encryption to FIPS-approved AES-GCM on FIPS-enabled hosts and limit Envoy's TLS cipher suites. Control plane TLS, however, stays on the default Go cipher suites.

Gateway API support extends into the mesh. HTTP and gRPC GAMMA routes attach directly to Services, and BackendTLSPolicy governs TLS from the proxy to the backend. DNS-aware policy and Hubble flow records that include policy-denial verdicts give security teams exportable logs attributed to each workload.

Traffic control covers weighted and header-based routing and request mirroring through the Gateway API, delay and abort fault injection through Cilium's Envoy integration, and automated canary promotion and rollback through Argo Rollouts. Locality routing keeps traffic within the caller's zone, then its cluster, and fails over automatically to other zones and remote clusters as local endpoints become unhealthy.

Cilium Service Mesh suits Kubernetes platform teams already running Cilium as their CNI who want L4 identity policy, flow observability, and Gateway API routing without sidecars. Organizations that need certificate-based workload identity or mesh-wide FIPS cryptography might need to consider a more fully featured mesh or run Istio alongside Cilium, an integration the project supports.

Google Cloud

Cloud Service Mesh*

Leader | Agile Optimizers | Fast Following

Summary:

Cloud Service Mesh is Google Cloud's managed service mesh, combining the former Anthos Service Mesh and Traffic Director. For GKE and Compute Engine, Google operates the mesh's control plane and its certificate authority, Mesh CA. Envoy sidecars or proxyless gRPC libraries form the data plane, configured through Istio APIs or Google Cloud service routing APIs.

An in-cluster control plane covers Google Distributed Cloud on premises. Google has deprecated GKE clusters on AWS and Azure, which lose support in March 2027.

Billing is about $0.50 per mesh client per month and includes telemetry, the control plane, and Mesh CA. A client is a pod, a proxyless gRPC instance, or a Cloud Run instance, and Cloud Run support is in Preview. Google aims it at Google Cloud customers.

Assessment:

Google operates Cloud Service Mesh as a managed service, which accounts for most of its operational appeal. The Traffic Director implementation carries a 99.9% SLA. The managed data plane upgrades proxies gradually after Google completes its control plane upgrade, and operators choose the scope and can track progress. The work of operating the mesh shifts from the platform team to Google, most completely on Google Cloud itself.

Envoy sidecars and proxyless gRPC coexist in one mesh with interoperable security. Individual gRPC clients can drop the sidecar for specific destinations, and services can move between the two models gradually. This gives gRPC-heavy environments a lighter data plane.

However, there is no node-level ambient mode. The Istio WasmPlugin API is not part of the managed product, so in-proxy extensions run through supported EnvoyFilter insertions and external authorization.

Day-2 operations include strong safeguards against misconfiguration. Diagnostics check the running mesh and pair detected faults with remediation guidance. Config Sync reconciles mesh policy from Git and reports drift.

Operators can dry-run authorization changes and analyze them against the live mesh before enforcing them. Traffic logs record SPIFFE principals and authorization denials, and Cloud Logging exports them to third-party SIEMs or Google Security Operations.

L7 handling covers HTTP/1.1, HTTP/2, and gRPC. The mesh does not support WebSocket at L7, and non-HTTP protocols pass through as opaque TCP streams. Google Cloud Deploy automates canary promotion and rollback based on mesh metrics and runs the analysis outside the mesh. Multicluster trust uses a shared root across the fleet.

The natural buyers are Google Cloud customers that want a mesh run as a service, with strong policy safeguards and SIEM-ready security logs. Teams that need broader L7 protocol support might need to supplement it.

Kuma (CNCF)

Kuma*

Challenger | Agile Optimizers | Fast Following

Summary:

Kuma is an open-source, Envoy-based service mesh created by Kong. It is now a CNCF Sandbox project under the Apache 2.0 license. It runs on Kubernetes and OpenShift and, in Universal mode, on virtual machines, bare-metal hosts, and Docker. The customer supplies and operates all infrastructure.

The project is free to download. Commercial support and an enterprise distribution come through Kong Mesh, which Kong sells separately. The project targets platform teams running mixed Kubernetes and VM estates across several zones or clouds.

Assessment:

Kuma treats Kubernetes and non-Kubernetes workloads as peers. Its kuma-dp data plane enrolls VMs, bare-metal hosts, and Docker containers with the same discovery, policy, and mesh identity as Kubernetes services. Kuma's global control plane synchronizes policy and service information down to each zone's control plane, and a global GUI provides visibility across the fleet. An environment spanning clusters, clouds, and older hosts can run as one mesh rather than several.

Locality routing shifts traffic to other zones progressively as local endpoints degrade and restores it as they recover. It works alongside circuit limits, active health checks, and per-service load-balancing algorithms. As a result, multizone deployments require less manual failover work.

The data plane is a per-workload Envoy sidecar. The project has removed an experimental eBPF interception path, so there is no ambient or node-level mode and no sidecar-to-ambient migration path. Operators tune proxy overhead per workload through resource patches.

Native authorization centers on authenticated workload identity (SPIFFE, service, and tag matching). Operators add rules on HTTP method, path, or headers by patching Envoy RBAC or external authorization filters into the proxy, rather than expressing them as native route conditions. JWT claims are not a native authorization condition.

The operator imports and rotates enterprise CA material, and mTLS applies once workloads have a mesh identity. From Kuma 3.0, inbound traffic then defaults to strict mTLS and undeclared plaintext destinations are refused, so relaxing enforcement is a deliberate operator choice.

Kuma suits platform teams that need one open mesh across Kubernetes, VMs, and several zones and are comfortable operating sidecars. Organizations that want sidecarless operation or native request-level authorization from the mesh might need to consider a more fully featured product. Kong Mesh is the route to commercial backing for Kuma itself.

Linkerd (CNCF)

Linkerd*

Challenger | Agile Optimizers | Fast Following

Summary:

Linkerd is a CNCF-graduated, open-source service mesh for Kubernetes, created by Buoyant. It is built on a purpose-built Rust proxy injected as a sidecar into each pod. It installs into the customer's own clusters, and mesh expansion brings VMs and physical hosts into the same mesh.

The project ships weekly edge releases under Apache 2.0. Since February 2024, stable releases have come from vendors, chiefly Buoyant Enterprise for Linkerd. Buoyant sells it in Premium and Strategic tiers by quote, and it is free in production for companies with fewer than 50 employees.

Assessment:

The evaluated product is the open-source CNCF project, and its release model matters as much as its features. Security fixes land in the next edge release. Supported stable lines with patch releases, along with FIPS builds, come from Buoyant's commercial distribution rather than the project. A buyer who wants long-lived versions and remediation commitments is in practice buying Buoyant Enterprise for Linkerd.

Linkerd's load balancing continuously weights each endpoint by latency and penalizes a degrading endpoint before removing it. Federated services merge local and remote endpoints into one balanced destination that survives the loss of a cluster. Cross-cluster resilience comes from the data plane rather than from failover runbooks.

Hybrid ML-KEM-768 with X25519 post-quantum key exchange is on by default for all meshed traffic, and proxy metrics show the negotiated groups. The linkerd check command ties detected failures to corrective commands.

The proxy is fixed-function by design. There is no WebAssembly host or API for customer-written filters, so custom request-path logic must run outside the mesh. Operators set in-proxy behavior through built-in Gateway API filters for headers and redirects, plus retries, timeouts, and rate limits.

Protocol awareness covers HTTP, gRPC, and WebSockets. The mesh routes and authorizes TLS egress traffic by hostname, while database and messaging traffic passes through as TCP.

Kubernetes workload identities follow Linkerd's own service-account naming rather than SPIFFE IDs, although mesh-expansion hosts can take SPIFFE identities from SPIRE. The default inbound policy accepts unauthenticated traffic until operators tighten it. Every mode keeps a per-pod proxy, with no sidecarless option.

A Kubernetes team that wants a small, opinionated proxy, latency-aware multicluster routing, and post-quantum encryption without extra configuration will find them here, provided it sources stable releases commercially. Organizations that need in-proxy extensibility, or FIPS from the open-source build, might need to consider a more extensible product or the Buoyant distribution.

Red Hat

Red Hat OpenShift Service Mesh

Leader | Agile Optimizers | Fast Following

Summary:

Red Hat OpenShift Service Mesh is Red Hat's supported distribution of Istio, Envoy, and Kiali for Red Hat OpenShift. It installs through an operator onto the customer's own OpenShift clusters, whether self-managed or on managed OpenShift services.

There is no separate purchase. The mesh comes with Red Hat OpenShift and OpenShift Platform Plus subscriptions, and Platform Plus adds the zero trust workload identity manager. Version 3.4 ships sidecar and ambient (sidecarless) data planes with the Kiali console and istioctl. Red Hat aims it at organizations standardized on OpenShift.

Assessment:

OpenShift Service Mesh runs on OpenShift and nowhere else. Upstream Kubernetes, other distributions, and non-Kubernetes runtimes are outside the supported footprint. VMs on OpenShift Virtualization can join the mesh with general availability. Meshing RHEL VMs outside the cluster is a Developer Preview, which Red Hat excludes from production use.

For an all-OpenShift environment, that boundary costs nothing. A mixed environment needs a second mesh for everything else.

Within that boundary, the product delivers the full Istio ambient model at general availability. Ztunnel handles L4 identity and mTLS at the node level, and waypoint proxies run L7 processing where workloads need it. Both coexist with sidecar workloads in the same mesh, and namespaces can move between modes one at a time and back again. Buyers can cut proxy overhead without a single cutover, although ambient multicluster remains a Technology Preview.

Security controls draw heavily on the wider Red Hat platform. Workload certificates can come from an external CA through cert-manager with automatic renewal, and a FIPS-validated cryptographic path is available. Hybrid ML-KEM post-quantum key exchange works in both data plane modes once configured. OpenShift platform services generate audit records for configuration changes, attributed to the person or system that made them.

WasmPlugin loads customer filters with digest verification, and Argo Rollouts, through OpenShift GitOps, drives metric-based canary promotion and rollback. Operators can set mesh-wide strict mTLS with a single policy, although permissive mode remains the default.

This is a natural fit for organizations committed to OpenShift that want a supported Istio, including sidecarless operation, inside a subscription they already hold. Teams running Kubernetes elsewhere, or meshing VMs outside OpenShift in production, might need to consider a product that meshes runtimes beyond OpenShift.

5.
Velocity

5. Velocity

Market-Level Velocity Story

A vendor's velocity is the pace at which it ships capability that moves the category, judged against the market's direction rather than raw feature count. Most of the field sits at Fast Following, with a minority Outpacing and a small tail at Slow Following.

The Fast Following majority ships steadily and on schedule. Its output includes quarterly or semiannual minor releases, long-term support lines, new identity and CA providers, native sidecars as the default, Gateway API updates, and ambient or post-quantum features that arrive soon after the upstream project ships them. Several are making substantial changes that affect their customers, through deprecated control plane modes, removed features, or mesh security paths in transition from one design to another. What they have not shipped is a generally available capability that reached the market first. For a category built on shared upstream code, that is a natural place for most commercial products to sit.

A vendor earns an Outpacing mark here by arriving first with a generally available capability or by originating a component that the upstream project later integrated. Five capabilities earned those marks. Three are conformance to the Gateway API inference extension, an agent gateway running as an ambient waypoint, and ambient multicluster held at general availability. The other two are hybrid post-quantum key exchange on by default for all meshed traffic, and post-quantum key exchange inside FIPS builds paired with in-proxy inspection of AI and MCP traffic. They come from upstream and commercial products alike, so speed does not depend on which upstream project a product comes from.

The Slow Following tail ships maintenance releases on an aging upstream base. These products build on the most widely adopted upstream project. One ships maintenance-only releases on a base several upstream minor releases old, with no new ambient, Gateway API, post-quantum, or MCP work while the market shipped all four. The other still carries ambient support and Gateway API routing at alpha on an upstream version whose community support has ended.

The mark describes pace, not reliability. It reflects the tradeoff of staying on an older upstream release, which Market Context describes. Expect the mark to stand until the vendor publishes a dated roadmap that changes it.

Read velocity as a sign of how a vendor approaches change, not as a quality ranking. In a market built on shared upstream projects, a Fast Following distribution is often exactly what a platform team wants: upstream capability, hardened and supported, a release or two later. Beyond the base-version questions Market Context raises, ask whether the vendor's release history shows steady, dated increments, and whether ambient operation, post-quantum cryptography, and AI traffic governance are moving toward general availability on a published schedule.

6.
Ecosystem

6. Ecosystem

Technical Support Quality and Responsiveness

Support depends on the commercial offering built around the code. Distributions sell it as the product, in tiers that run from around-the-clock coverage on a long-lived release line up to hotpatches, long-term CVE patching, faster response, and published remediation deadlines for critical vulnerabilities. Managed delivery adds an uptime SLA.

The foundation-hosted projects instead patch each release for a fixed period that the community sets, which is a release policy rather than a support relationship. Once you choose the distribution, what remains is contractual: get response targets and remediation deadlines in writing for the exact release line you will run, not for the vendor's newest one.

Professional Services Depth

Vendors deliver implementation help mostly as product features rather than as services engagements. Migration tooling does most of this work. It assesses readiness and generates waypoint and policy configuration for an incremental, reversible move to ambient. It also replaces the CNI node by node with both data planes live and a way back, and it moves workloads from client agents to a lighter data plane without downtime.

Operator-driven upgrades commonly run the control plane first and the data plane second. A few products hand operations to the vendor through a managed control plane or a SaaS console. Tooling covers the mechanics of a migration, not its judgment calls, so ask each vendor who performs the work and price it separately from the subscription.

Partner and Channel Ecosystem Breadth

A mesh that comes inside a container platform subscription, with no separate purchase, sets the procurement path before buyers compare any feature. A mesh that buyers often purchase alongside its vendor's API gateway does much the same. Cloud marketplaces carry several products on node-based or cluster-pack terms, under subscription, pay-as-you-go, or annual contracts. Public-sector buyers can reach a specialist platform through government resellers and contract vehicles.

Integration partnerships cluster around certificate authorities, with external secret stores, cert-manager, and cloud private CAs recurring as issuers, and around progressive delivery tools such as Argo Rollouts and Flagger. Map these relationships to the platform, gateway, and PKI estate you already run, because they decide whether the mesh fits into your existing stack or arrives as a parallel one.

Workforce Skills Availability

Envoy configuration and filter-chain debugging carry from one product to the next, because most of the field builds its data plane on Envoy, whether as a sidecar, a per-node proxy, or a waypoint. The exceptions narrow or redirect that skill. A purpose-built Rust microproxy that accepts no custom filter code asks less of operators. A proprietary node model with a JavaScript engine for request customization requires different skills.

Gateway API and upstream project resources are portable skills, while console-defined objects and vendor-specific languages tie expertise to one vendor. SPIFFE identity, cert-manager, and OpenTelemetry recur widely. Hire for Envoy, Kubernetes networking, and PKI fundamentals, and weight products whose configuration model matches skills you can recruit.

Training and Certification Programs

Documentation, release notes, and community material for a foundation-hosted project apply directly to every distribution built on it, so most enablement comes from upstream. Vendors add operational aids inside the product: diagnostics that pair a detected fault with remediation guidance or a corrective command, analysis of proposed configuration before operators apply it, in-cluster documentation, and policy generated from observed traffic for review. Treat the built-in diagnostics as part of the training program. Confirm which curricula and credentials cover the upstream project and which cover the distribution, and budget hands-on time for the data plane model you choose.

Community Health

Most of the open-source projects beneath service mesh hold graduated CNCF status, with one still at the Sandbox stage. That gives the category a stronger shared open-source base than most infrastructure markets have. Project health still varies.

The largest project ships roughly quarterly minor releases and patches each for a defined window. Another publishes frequent edge releases, while its stable releases have come only from a commercial vendor since February 2024. A third concentrates development in one sponsoring company, and its latest major release removed several features.

When one sponsor funds most of the work, its commercial priorities set the roadmap. Before treating community backing as a support plan, check contributor diversity, the stable-release policy, and who patches the version you run.

7.
Architect's Validation

7. Architect's Validation

No organization weights the decision criteria equally, and a few deployment-specific questions decide outcomes in service mesh. Before shortlisting, pressure-test four things in order.

First, the data plane: run your own workload mix under the sidecar, ambient, or node-level model you will operate, and measure proxy cost and upgrade disruption, not only latency. Second, default security posture: install fresh and record what the mesh enforces before any configuration. Third, failure behavior: take the control plane down and observe what keeps working, for how long, and what new workloads can do. Fourth, estate reach: mesh one workload outside Kubernetes and one workload in a second cluster, and confirm that identity and policy follow.

The table below indicates when to promote specific criteria to first-order requirements and when to let them recede.

Prioritize when…

Treat as secondary when…

Sidecarless / Ambient / eBPF Architecture Support and Resource Efficiency: proxy cost at thousands of pods is a budget line, or dense clusters cannot carry a proxy per workload

Service counts are modest, and per-pod proxy cost is small next to the workloads themselves

mTLS Enforcement Posture and Policy Rollout Safety: zero-trust mandates require strict mTLS everywhere, and enforcement must be staged without breaking unmeshed callers

The mesh serves a bounded, fully meshed environment where strict mode can be set once

Enterprise PKI and CA Integration and FIPS Cryptography: workload certificates must chain to an existing enterprise CA, or regulated workloads require validated cryptography

The mesh CA is acceptable as a trust root, and no FIPS obligation applies

Non-Kubernetes and Multi-Runtime Support: VMs, bare-metal hosts, or container services outside Kubernetes will carry production traffic for years

The estate is Kubernetes-only, and legacy hosts are scheduled for retirement

Multi-Cluster and Hybrid Cloud Management and Control Plane Resilience: services span clusters, clouds, or data centers, and one cluster's failure must not stop the others

A single cluster or region hosts the workload, and cross-cluster failover is not required

Gateway API / GAMMA Conformance and Extensibility: portability across distributions matters, or custom request-path logic must run in the proxy

One vendor's configuration model is an accepted long-term commitment, and no custom filters are planned

AI and Agentic Traffic Governance: AI agents and model servers will call internal services and tools through the mesh

AI traffic is handled entirely by a separate gateway outside the mesh

Adapting the Radar to Your Organization. Start from your three or four nonnegotiable criteria and re-read the vendor capsules against those alone. Let that reading reorder the field before you consider overall placement. The placements in this report reflect the market's overall direction, not your architecture. A product positioned modestly here may be the right fit for your weighting.

8.
Horizon

8. Horizon

Capabilities Likely to Become Table Stakes (12–24 Months)

Several capabilities that separate products today are on track to become standard expectations. A sidecarless option is first. Several products already run ambient and sidecar data planes together at general availability, and a few more are sidecarless by design. Alpha or preview modes appear in other products and in the managed cloud service.

Gateway API routing for east-west traffic is next, with GAMMA routes stable in some products and alpha or absent in others. Buyers will increasingly treat a proprietary routing model as a portability cost. Hybrid post-quantum key exchange for mesh mTLS is moving the same way: it is already on by default in some products, configurable in others, and in release candidates elsewhere. SPIFFE workload identity that spans containers, VMs, and bare metal through SPIRE is common enough that its absence will need explaining.

Control plane outage behavior is converging as well. Proxies that keep serving on last-known-good configuration until workload certificates expire are now the common baseline. Buyers will therefore press vendors for stated certificate grace periods and redundancy guidance rather than accept the behavior on trust. Native Kubernetes sidecars as the default injection mode, and diagnostics that check the live mesh before an operator applies a change, round out the list.

Capabilities Likely to Emerge as Differentiators

Early features in today's products already point to the next round of differentiators, beginning with governance of AI agent and model traffic inside the mesh. Today most of it ships as companion products. The first vendors to bind agent identity, tool authorization, and token budgets to mesh workload identity in a production release will set the reference. Ambient multicluster is second: a single distribution offers it at general availability, while the upstream project carries it at beta and another distribution at Technology Preview.

Policy safety is third. Simulating proposed policy against live cluster data, generating policy from observed flows, and dry-running authorization changes before enforcement each ship in individual products today. Buyers will reward the vendor that turns these into one supported workflow. Finally, regulated cryptography will separate the field as buyers expect FIPS-validated builds and post-quantum key exchange to work together, a combination few products document.

Expected Market Consolidation or Fragmentation

Consolidation here follows lineage rather than size. The project families that Market Context describes hold most of the field, and a few independent platforms and a managed service stand apart. Ownership has already shifted: some commercial mesh vendors now sit inside larger infrastructure companies, and one vendor's managed mesh service reached end of life in November 2025.

Within families, products are narrowing scope rather than merging. Recent releases have removed features, retired gateway types, deprecated control plane modes, and moved AI functions into separately sold products. None of this points to a specific acquisition, but the consequence for due diligence is clear. Buyers should establish which capabilities live in the SKU they are purchasing, which depend on a companion product, how the release line maps to its upstream project, and what the exit path is if the sponsor's priorities change.

9.
Value Chain

9. Value Chain

A service mesh sits between the application platform and the security and operations estate. The mesh's job is to make every call between workloads authenticated, encrypted, routed, and observable without changing application code.

Upstream producers and consumers

The service mesh

Downstream estate

Microservices on Kubernetes; VMs, bare-metal hosts, and container services outside Kubernetes; CI/CD, GitOps, and progressive delivery pipelines; AI agents, MCP servers, and model-serving endpoints

Control plane issuing identity, policy, and routing configuration; data plane of sidecars, node-level proxies, waypoints, or kernel programs; workload identity and certificate issuance; mTLS and L7 authorization; traffic management and resilience; telemetry, flow records, and audit events

Enterprise CAs and secret stores; identity providers; observability, tracing, and SIEM platforms; API, ingress, and AI gateways at the mesh edge; CNI and cluster networking; managed Kubernetes and VM platforms

The differentiation in this field lives at two seams. The lower seam is the network. Whether the mesh runs as sidecars above the CNI, as ambient node proxies alongside it, or inside the CNI itself decides resource cost, upgrade mechanics, and which encryption and identity features are generally available.

The upper seam is the edge of the mesh, where east-west policy meets API gateways, ingress, and AI gateways. Products that express both seams through Gateway API resources and SPIFFE identity give buyers portable configuration. Products that use console-defined objects or vendor-specific languages trade that portability for an integrated model.

The mesh tends to enter the architecture at predictable moments. These include a zero-trust mandate that requires workload identity and mTLS between services, a move from one cluster to many, a platform refresh, or a cost review of sidecar overhead at scale. Exit is easier than in most infrastructure categories when the product stays close to its upstream project, because the same resources and identities carry over to another distribution. It is harder where routing, policy, and extensions live in proprietary models. Choose the mesh based on how well it handles both seams, and confirm before purchase that policy, identity, and certificates can move to another product without rewriting every service's configuration.

10.
Appendix A — Decision Criteria

10. Appendix A — Decision Criteria

This evaluation groups its decision criteria by the accountability lens they serve.

Brown Money

  • BM1. Resource Efficiency and Data Plane Overhead: How far operators can cut data plane cost, by sending less traffic through L7 proxies or by setting per-workload proxy resource limits.

  • BM2. Sidecarless / Ambient / eBPF Architecture Support: Whether a generally available data plane serves workloads without a proxy in each pod, or with a lighter per-workload variant.

  • BM3. Sidecar-to-Ambient Migration and Coexistence: Whether workloads move between data plane models one namespace or workload at a time, with coexistence and a way back.

  • BM4. Control Plane Scalability: Whether one control plane's capacity is stated in proxies, services, or endpoints, from testing or a named deployment.

  • BM5. Control Plane Resilience and Blast Radius: What keeps working when the control plane is lost, and how far a single failure reaches.

  • BM6. Operational Lifecycle and Upgrades: How traffic keeps flowing during upgrades, through version skew support, canary revisions, staged operators, and pre-upgrade checks.

  • BM7. Platform and Distribution Support: Which Kubernetes releases, managed services, and on-premises distributions the product supports.

  • BM8. Day-2 Tooling and Stack Integration: Whether a CLI or console diagnoses proxies, certificates, and live traffic, and exports metrics and traces.

Green Money

  • GM1. Multi-Cluster and Hybrid Cloud Management: Whether cross-cluster services are discovered automatically and secured by mesh identity under the same policy as local calls.

  • GM2. Non-Kubernetes and Multi-Runtime Support: Whether VMs, bare-metal hosts, and container services outside Kubernetes get identity, policy, and telemetry from the same control plane.

  • GM3. Multiprotocol Support: Which protocols the mesh understands beyond HTTP, including gRPC, WebSocket, messaging, and database traffic.

  • GM4. Progressive Delivery and Advanced Traffic Control: Whether weighted and header routing combine with mirroring or fault injection, and canaries promote or roll back on mesh metrics.

  • GM5. Resilience and Circuit Protection Depth: How the mesh contains a failing service, through circuit-breaker limits, outlier ejection, and active health checks.

  • GM6. Extensibility and Programmability: Whether customers add request-path behavior through a supported filter API or their own Wasm filters, without forking the proxy.

  • GM7. Gateway API / GAMMA Conformance: Whether ingress and east-west routing use generally available Gateway API resources, including HTTPRoute attached to a Service.

  • GM8. Load Balancing and Locality Routing: Whether traffic stays in the caller's zone while healthy, fails over outward automatically, and weights endpoints by latency.

GRCS (Governance, Risk, Compliance & Security)

  • GRCS1. mTLS Enforcement Posture: Whether mTLS covers all meshed connections by default or is enforced mesh-wide by one policy that blocks plaintext.

  • GRCS2. Workload Identity and Federation: Whether workloads receive rotated SPIFFE identities, attested through SPIRE where needed, that federate across trust domains.

  • GRCS3. Enterprise PKI and CA Integration: Whether certificates come live from an enterprise CA such as Vault, cert-manager, SPIRE, or a cloud private CA.

  • GRCS4. L7 Authorization Granularity: Whether requests are authorized by method, path, headers, and JWT claims against the caller's verified identity, per route.

  • GRCS5. FIPS and Regulated-Environment Cryptography: Whether regulated buyers have a supported FIPS path, through builds on validated modules or mesh-wide cipher restrictions.

  • GRCS6. Audit Logging and Compliance Reporting: Whether traffic records carry authorization outcomes and configuration changes leave an actor-attributed audit trail.

  • GRCS7. Supply Chain and Patch Integrity: Whether each supported version line receives CVE-tracked patches, with SBOMs, provenance, and stated remediation timelines.

  • GRCS8. Policy Rollout Safety: Whether authorization runs in audit or shadow mode, and strict mTLS phases in per namespace, before enforcement.

Emerging Technology

  • EM1. AI and Agentic Traffic Governance: How far the mesh governs AI agent and model traffic, from identity-scoped route authorization to tool allow-lists and token budgets.

  • EM2. eBPF Data Plane Maturity: Whether eBPF programs in the node kernel capture or act on mesh traffic, and how much of the data plane they carry.

  • EM3. AI-Assisted Policy and Operations: Whether the product analyzes live mesh state and proposed changes, and reports failing checks against the affected resources.

  • EM4. WebAssembly Extensibility Maturity: Whether a supported Wasm extension resource loads Proxy-Wasm modules into the proxies with versioning and verification.

  • EM5. Ambient Multicluster Maturity: Whether multicluster operation works with a generally available sidecarless data plane, not only per-workload proxies.

  • EM6. Post-Quantum Cryptography: Whether mesh mTLS negotiates a hybrid ML-KEM key exchange such as X25519MLKEM768 in a released version.

  • EM7. Runtime Security and Observability Convergence: Whether telemetry and flow records name source and destination workloads by verified identity.

11.
Methodology

11. Methodology

*Vendors marked with an asterisk did not participate in our research process for the Radar report, and their capsules and scoring were compiled via desk research.

For more information about our research process for Radar reports, please visit our Methodology.

12.
About Ivan McPhee

12. About Ivan McPhee

Formerly an enterprise architect and management consultant focused on accelerating time-to-value by implementing emerging technologies and cost optimization strategies, Ivan has over 20 years’ experience working with some of the world’s leading Fortune 500 high-tech companies crafting strategy, positioning, messaging, and premium content. His client list includes 3D Systems, Accenture, Aruba, AWS, Bespin Global, Capgemini, CSC, Citrix, DXC Technology, Fujitsu, HP, HPE, Infosys, Innso, Intel, Intelligent Waves, Kalray, Microsoft, Oracle, Palette Software, Red Hat, Region Authority Corp, SafetyCulture, SAP, SentinelOne, SUSE, TE Connectivity, and VMware.

An avid researcher with a wide breadth of international expertise and experience, Ivan works closely with technology startups and enterprises across the world to help transform and position great ideas to drive engagement and increase revenue.

13.
About GigaOm

13. About GigaOm

GigaOm provides technical, operational, and business advice for IT’s strategic digital enterprise and business initiatives. Enterprise business leaders, CIOs, and technology organizations partner with GigaOm for practical, actionable, strategic, and visionary advice for modernizing and transforming their business. GigaOm’s advice empowers enterprises to successfully compete in an increasingly complicated business atmosphere that requires a solid understanding of constantly changing customer demands.

GigaOm works directly with enterprises both inside and outside of the IT organization to apply proven research and methodologies designed to avoid pitfalls and roadblocks while balancing risk and innovation. Research methodologies include but are not limited to adoption and benchmarking surveys, use cases, interviews, ROI/TCO, market landscapes, strategic trends, and technical benchmarks. Our analysts possess 20+ years of experience advising a spectrum of clients from early adopters to mainstream enterprises.

GigaOm’s perspective is that of the unbiased enterprise practitioner. Through this perspective, GigaOm connects with engaged and loyal subscribers on a deep and meaningful level.

Discover how to put this research to work

Bring your questions to a GigaOm analyst and get guidance specific to your environment and decisions.

Sent with: GigaOm Radar for Service Mesh