Skip to content
AI Economics Hub
Market map

AI economics vendor map

The tools that let an enterprise see, control and prove the economics of the AI it uses, from the model call to the board pack. Organised by the problem each one solves, with every entry checked against its own sources.

Last reviewed
vendors and platforms
100vendors and platforms
categories across five control layers
13categories across five control layers
ownership, name or status changes logged since March 2025
16ownership, name or status changes logged since March 2025
listed vendors that are Tokenomics Foundation members
28listed vendors that are Tokenomics Foundation members

Vendor-neutral and unsponsored. The map describes vendors from public sources and does not rate, score or rank them. IBM-owned products are marked IBM-owned. See how the map is built. Tokenomics Foundation membership comes from the Foundation's members page and signals engagement with the discipline, not quality. By Ed Epstein, who works at IBM. The postings on this site are my own and don't necessarily represent IBM's positions, strategies or opinions. Full disclosure.

Start with your problem

What are you trying to fix?

Most buying mistakes in this market come from choosing a category before naming the problem. Pick the statement closest to yours to see the vendors that address it.

Directory

100 vendors, 13 categories

Map last updated , the most recent entry check. Entries are rechecked at least every six months.

Showing 100 of 100

Layer A: Build and run

Model Gateways & Routing

11 vendors

Problem it solves: Multi-provider access, routing policy, failover, budget enforcement at the point of consumption.

How to read this category

The gateway is where AI cost policy becomes enforceable rather than reportable. This is the layer FinOps teams most often discover they are missing.

Maps to Usage Transparency in the Five-Layer model; most relevant at maturity levels 1 to 3.

Also plays a role here:Helicone

Layer A: Build and run

LLM Observability & Evaluation

13 vendors

Problem it solves: Trace-level cost, quality, and workflow context.

How to read this category

This is where cost data meets quality data. Without it, optimisation degrades into cheapest-token chasing.

Maps to Usage Transparency and Output Quality in the Five-Layer model; most relevant at maturity levels 2 to 3.

  • Arize (Phoenix and Arize AX)

    Two products: Phoenix, a free self-hosted tracing and evaluation project, and Arize AX, the commercial platform with online evaluation, drift monitoring and enterprise support.

    • Tracing
    • Evaluation
    • Cost attribution
    Profile and sources

Layer A: Build and run

Infrastructure & Inference Optimisation

10 vendors

Problem it solves: GPU efficiency, container and workload optimisation, commitment management.

How to read this category

Infrastructure optimisers focus on GPU utilisation, Kubernetes efficiency, and cloud commitment management.

GPU cloud providers such as CoreWeave, Lambda, Together, Fireworks, Baseten, and Modal are supply-side infrastructure, not governance tools, and are therefore not profiled as directory entries. They are referenced here as context for where AI workloads run, but the vendor map focuses on buyer-side control and visibility platforms.

Maps to Usage Transparency in the Five-Layer model; most relevant at maturity levels 2 to 3.

  • Cast AI

    Tokenomics member

    Kubernetes optimisation platform with automation-led scaling and strong relevance for GPU-heavy AI infrastructure.

    Profile and sources

  • IBM Kubecost

    IBM-ownedTokenomics member

    Kubernetes-native cost allocation platform with strong GPU and workload-level visibility for AI infrastructure.

    Profile and sources

  • IBM Turbonomic

    IBM-ownedTokenomics member

    Application resource management platform that uses automation to improve utilisation and performance across cloud and container environments.

    Profile and sources

Layer B: Financial governance

FinOps Platforms & AI Cost Management

18 vendors

Problem it solves: Allocation, showback, anomaly detection, and governance across cloud plus AI spend.

How to read this category

FinOps platforms extend cloud cost governance into AI workloads, providing allocation, budgets, and optimisation workflows.

Maps to Usage Transparency and Delivery Alignment in the Five-Layer model; most relevant at maturity levels 2 to 4.

Layer B: Financial governance

TBM / ITFM Platforms

5 vendors

Problem it solves: AI cost inside enterprise service, planning, and portfolio structures.

How to read this category

TBM and ITFM platforms translate AI cost into business service models, cost pools, and portfolio planning structures.

Maps to Delivery Alignment and Portfolio Strategy in the Five-Layer model; most relevant at maturity levels 3 to 5.

  • MagicOrange

    Technology financial management platform recognised in analyst conversations for cost transparency, service insight, and business reporting.

    Profile and sources

  • Nicus

    ITFM platform for cost transparency, forecasting, and scenario modelling with explicit visibility into AI and cloud costs.

    Profile and sources

Layer B: Financial governance

AI Unit Economics & Attribution

2 vendors

Problem it solves: We can see the AI bill but not what each customer, agent, workflow or product feature costs, so we cannot price, budget or protect margin.

How to read this category

A young category that sits between the gateway and the finance team. These tools meter AI usage at transaction level and attribute it to a customer, agent, feature or workflow, so cost can be compared with revenue or value. They matter most to companies that sell AI-powered products and need cost per customer or cost per outcome. Expect overlap with FinOps platforms adding token cost, and with gateways adding metering.

Usage-based billing platforms such as Stripe Billing (which completed its Metronome acquisition on 14 January 2026), Orb and Amberflo are seller-side tools for charging your customers. They are out of scope here, but often appear in the same evaluation.

Maps to Usage Transparency and Portfolio Strategy in the Five-Layer model; most relevant at maturity levels 2 to 4.

  • Revenium

    Tokenomics member

    Metering and attribution for AI usage at transaction level, across tokens, APIs and agent workflows, with real-time spend controls.

    Profile and sources

Layer C: Consumption and workforce

Workforce AI Usage & SaaS AI Spend

7 vendors

Problem it solves: Copilot, ChatGPT Enterprise, Claude, and embedded AI SKUs are bought as software subscriptions outside the FinOps remit. Shadow AI usage is invisible to all of it.

How to read this category

For most enterprises, seat-based workforce AI is already a larger controllable line than API tokens, and almost no FinOps practice owns it.

Maps to Usage Transparency and Productivity Value in the Five-Layer model; most relevant at maturity levels 1 to 3.

  • Larridin

    Measurement layer for enterprise AI: what AI costs, what teams produce with it and which workflows to automate next.

    Profile and sources

  • Portal26

    GenAI visibility platform spanning shadow AI discovery, licence intelligence, and a 2026 AI value realisation module.

    Profile and sources

  • Surveil

    Cost, usage and licence intelligence across the Microsoft estate and multicloud, with a module for Microsoft 365 Copilot readiness, adoption and licence optimisation.

    Profile and sources

Also plays a role here:OpenAI

Layer C: Consumption and workforce

AI Coding & Developer Tools

9 vendors

Problem it solves: Per-seat and usage-hybrid AI coding spend with weak FinOps visibility.

How to read this category

AI coding tools represent one of the largest current enterprise AI spend categories, but cost governance maturity varies widely across vendors.

Maps to Usage Transparency and Productivity Value in the Five-Layer model; most relevant at maturity levels 1 to 3.

Layer C: Consumption and workforce

Engineering Productivity & AI Impact Measurement

6 vendors

Problem it solves: Proving whether AI coding spend changes engineering output, the bridge from cost to value for the largest current enterprise AI use case.

How to read this category

DORA-metric relabelling is not AI measurement. The credible platforms compare AI-assisted and unassisted work and connect it to delivery outcomes.

Maps to Productivity Value in the Five-Layer model; most relevant at maturity levels 3 to 4.

  • DX

    Developer experience and productivity platform with dedicated AI impact measurement across coding assistants.

    Profile and sources

Also plays a role here:IBM Bob

Layer D: Value and strategy

AI Value Management & Portfolio Governance

7 vendors

Problem it solves: Connecting AI usage, cost, quality, and productivity to portfolio decisions, the full five-layer problem.

How to read this category

The thinnest category in the directory, but thin coverage here is not proof that capable products do not exist. Portfolio platforms such as Planview, AI governance layers from ServiceNow and SAP, and new value modules from IBM Apptio are all moving into this space, mostly with capabilities announced in 2026 and not yet proven at scale. Check each one's value evidence: few record realised financial benefit rather than adoption or estimated savings.

Maps to Usage Transparency and Output Quality and Productivity Value and Delivery Alignment and Portfolio Strategy in the Five-Layer model; most relevant at maturity levels 4 to 5.

  • Ascerta (formerly Pay-i)

    RenamedTokenomics member

    Measures AI value, adoption and ROI, with products for workflows and portfolios (Atlas), engineering teams' use of coding agents (Forge) and self-hosted AI capacity (Convoy).

    Profile and sources

  • IBM Apptio AI Value & ROI

    Part of IBM

    IBM-ownedTokenomics member

    Links AI initiatives to business outcomes and proof metrics, tracking baseline, target and realised results against modelled AI cost.

    • Outcome tracking
    • Cost attribution
    • Portfolio planning
    Profile and sources

Layer D: Value and strategy

Process Outcomes & Benefits Realisation

2 vendors

Problem it solves: Showing, from operational data, whether AI changed how work actually runs: cycle times, exceptions, rework and cost per completed case.

How to read this category

These are outcome evidence layers, not AI cost tools. They supply the before-and-after process data that a benefits case needs, and they are most useful when joined to cost data from layers A and B.

Maps to Productivity Value and Delivery Alignment in the Five-Layer model; most relevant at maturity levels 3 to 5.

  • Celonis

    Process intelligence platform that reconstructs how processes actually run, providing the operational evidence to test whether AI changed business performance.

    • Process intelligence
    • Outcome tracking
    Profile and sources

  • UiPath Maestro

    Part of UiPath

    Orchestration of AI agents, robots and people in one process, with process mining and automation-level analytics.

    • Workflow orchestration
    • Process intelligence
    Profile and sources

Layer E: Native provider controls

Hyperscalers

4 vendors

Problem it solves: Native AI billing, tagging, and optimisation controls from cloud platforms.

How to read this category

Hyperscaler native controls are improving but remain incomplete. They are designed to keep workloads within the platform ecosystem.

Maps to Usage Transparency and Delivery Alignment in the Five-Layer model; most relevant at maturity levels 1 to 3.

Layer E: Native provider controls

Model Providers

6 vendors

Problem it solves: Direct model API billing, usage reporting, and spend controls.

How to read this category

Model providers offer varying levels of cost visibility and control. Enterprise governance depth differs significantly across vendors.

Maps to Usage Transparency in the Five-Layer model; most relevant at maturity levels 1 to 2.

  • Anthropic

    Claude API with token-based pricing, caching tiers, batch discounts, and admin APIs for granular usage and cost reporting.

    Profile and sources

  • Cohere

    Model API with token-, search-, and embedding-based billing plus billed token reporting and production and trial key separation.

    Profile and sources

Analysis

What the market is signalling to buyers

Seven patterns drawn from the entries and dated events on this page. Each ends with what we would do about it.

  1. 01

    Consolidation is fast, so contracts need exit terms

    We logged 16 acquisitions, renames or exits among vendors on or near this map since March 2025, and 14 current entries carry an ownership or name change. Products bought eighteen months ago may now sit inside a different company with a different roadmap.

    What to do: Ask for change-of-control terms, data export in an open format and a dated roadmap commitment.

  2. 02

    AI gateways are turning into security products

    Portkey now ships as part of Palo Alto Networks Prisma AIRS. Kong AI Gateway and Cloudflare AI Gateway come from API and network security, and TrueFoundry leads with an MCP gateway for agent tool access. Cost control is becoming one feature of a security purchase.

    What to do: Decide who owns the gateway, platform engineering or security, and check that budgets and attribution survive the change.

  3. 03

    Observability tools are being absorbed by data and infrastructure platforms

    ClickHouse acquired Langfuse, CoreWeave owns Weights & Biases Weave, Mintlify acquired Helicone, Cisco acquired Galileo for Splunk and Traceloop is joining ServiceNow. Trace data, the raw material for cost per outcome, is moving into larger estates.

    What to do: Prefer OpenTelemetry-based instrumentation and open-source cores so traces stay portable if the vendor changes hands.

  4. 04

    FinOps vendors are repositioning around AI, at different depths

    On 6 October 2026 the homepage titles read “The AI ROI Company” (CloudZero), “FinOps Platform for the Agentic Era” (Finout) and “the complete token engineering platform” (Faros AI). Kion added Anthropic token spend management in a point release. Messaging has moved faster than most products.

    What to do: Ask each vendor to load a month of your own model provider billing data and show allocation by team and use case.

  5. 05

    Cloud providers are building AI cost controls into their own consoles

    In 2026 Google Cloud announced spend caps that pause AI API traffic at a project budget, in private preview, and an agent that explains AI cost drivers. AWS released a FinOps agent in preview for cost questions and anomalies. Microsoft Azure is previewing per-project cost attribution for the models it sells and, in private preview, agent ROI in Foundry. Each covers only its own cloud, and most of it is still in preview.

    What to do: Use native controls for single-cloud limits and alerts, and ask what a third-party tool adds across providers, SaaS AI and seats before paying for both.

  6. 06

    Coding agents are the fastest-moving cost line

    SpaceX completed its acquisition of Cursor in August 2026, Windsurf became Devin Desktop, and new tools such as Quesma and Not Diamond exist only to explain or reduce coding-agent spend. Seat prices are giving way to usage-based charges.

    What to do: Track cost per merged pull request or per completed task, not cost per seat.

  7. 07

    Value measurement is thin here, and new to the market

    This directory lists 7 entries in AI value management, 2 in process outcomes and 2 in AI unit economics, against 18 FinOps platforms. A short list is not proof of an empty market: Planview and IBM Apptio AI Value & ROI both announced value capabilities in 2026. But only 2 assessed entries are designed to record realised financial benefit rather than adoption or estimated savings.

    What to do: Ask each vendor which kind of value evidence it records, and expect to build part of the value model yourself; see the AI Value Gap and AI ROI pages.

Timeline

Market moves

Acquisitions, renames and exits that change what you are buying. Each links to its source. Removed from the map in this edition: Productiv (ceased operations) and Zluri (now an identity governance product).

  1. 30 September 2026

    Pay-i rebrands as Ascerta

    Announced with an $18 million Series A led by Dell Technologies Capital; moves from AI cost tracking towards measuring AI value, adoption and outcomes. GeekWire

  2. 14 August 2026

    SpaceX completes acquisition of Cursor

    The most widely used AI code editor becomes part of a company that trains its own models. Cursor blog

  3. 6 August 2026

    Productiv ceases operations

    A SaaS management vendor exits, a reminder to check vendor viability before multi-year contracts. Productiv

  4. August 2026

    DoiT acquires Attribute

    The startup's AI cost attribution, covering token, model and GPU cost, is sold as a DoiT product launched in July. Terms not disclosed. Calcalist

  5. 16 July 2026

    Portkey becomes Prisma AIRS AI Gateway

    Palo Alto Networks makes its Portkey-based AI gateway generally available, about six weeks after closing the acquisition. Palo Alto Networks blog

  6. 2 June 2026

    Windsurf renamed Devin Desktop

    Cascade agent retired on 1 July 2026, replaced by Devin Local. Devin Desktop

  7. 22 May 2026

    Cisco completes acquisition of Galileo

    An AI evaluation and guardrails platform joins Splunk Observability. Cisco blog

  8. 9 March 2026

    OpenAI agrees to acquire Promptfoo

    The open-source evaluation and red-teaming tool is to be folded into OpenAI Frontier; OpenAI says the open-source project will continue. OpenAI announcement

  9. 3 March 2026

    Mintlify acquires Helicone

    An open-source LLM observability tool joins a documentation platform. Helicone blog

  10. March 2026

    ServiceNow acquires Traceloop

    OpenTelemetry-based LLM instrumentation moves towards ServiceNow AI governance. Traceloop blog

  11. 16 January 2026

    ClickHouse acquires Langfuse

    Langfuse says it stays open source and self-hostable. Langfuse blog

  12. 14 January 2026

    Stripe completes Metronome acquisition

    Usage-based billing consolidates into payment platforms. Seller-side billing is outside this map but often in the same evaluation. Stripe newsroom

  13. 6 January 2026

    Flexera acquires ProsperOps and Chaos Genius

    Cloud commitment automation and Snowflake and Databricks cost optimisation join the Flexera portfolio; ProsperOps keeps its brand. Flexera press release

  14. 14 July 2025

    Cognition takes over Windsurf

    The Windsurf product and team move to Cognition, maker of Devin. SiliconANGLE

  15. 5 May 2025

    CoreWeave acquires Weights & Biases

    Weave, the W&B LLM observability product, now sits inside a GPU cloud provider. CoreWeave press release

  16. 3 March 2025

    Flexera completes acquisition of Spot from NetApp

    Spot Eco, Ocean, Elastigroup and CloudCheckr move into the Flexera portfolio. Thoma Bravo press release

Market structure

Five control layers, from model call to business outcome

No single platform covers the whole chain. Most organisations combine one tool from layer A or E for live control, one from layer B for financial allocation, and build or buy the value view in layer D.

Layer A

Build and run

Engineering infrastructure for AI demand: model routing, spend enforcement, trace-level observability, and GPU efficiency.

34 vendors

Layer B

Financial governance

Cost allocation, showback, and financial governance from cloud workloads to enterprise portfolio views.

25 vendors

Layer C

Consumption and workforce

Workforce AI tools, seat spend, shadow AI discovery, and measuring productivity impact across engineering.

22 vendors

Layer D

Value and strategy

Connecting AI cost, usage, quality, and productivity to portfolio decisions and business outcomes.

9 vendors

Layer E

Native provider controls

Baseline billing, tagging, and optimisation controls from hyperscalers and model providers.

10 vendors

Categories also map to the Five-Layer AI Value Intelligence model and the AI Economics Maturity Model; open “How to read this category” in the directory for each mapping. Open standards and proposals, such as FOCUS, OpenTelemetry and the draft AI value data layer, are not vendors and are tracked on the standards map.

Supporting notes

Enabling technologies that change the economics

These parts of the AI stack move unit costs, sometimes more than model price does, but none is a buyer-side governance product. They are described here rather than listed in the directory. An entry is added only where a product gives the buyer a distinct economic control.

Enabling technologies, their economic effect and why they are not directory entries
TechnologyEconomic effectWhy it is not listed
Semantic cachingRedis LangCacheReturns a stored answer when a new prompt means the same as an earlier one, so output tokens are not paid for again. Savings scale with the cache hit rate.A component inside an application or gateway, not a governance product. Several listed gateways include caching.
Inference serving and prefix cachingvLLMReuses computation for prompts that share a prefix, such as a long system prompt, which cuts prefill cost and latency on self-hosted models.Supply-side infrastructure. Its effect shows up in the GPU and infrastructure costs that layer A and B tools measure.
Agent frameworksLangGraph, CrewAI, OpenAI Agents SDKDecide how many model and tool calls one task makes. Loop limits, retries and planning steps often matter more to cost per task than model price.Build tools with no buyer-side cost control of their own. Their cost appears in observability and gateway tools.
Vector databases and embeddingspgvector, Pinecone, WeaviateRetrieval quality sets how much context each prompt carries. Poor retrieval inflates input tokens; storage and re-embedding are recurring costs.Data infrastructure. Their spend belongs in cloud and FinOps cost allocation.
MCP servers and toolsModel Context Protocol serversEach tool an agent can call adds context, calls and sometimes per-call fees, and widens the surface to govern.Governed through gateways and agent inventories, such as TrueFoundry and SAP AI Agent Hub, rather than as products in their own right.

Method

How to read the map

The map organises vendors by the problem each one addresses. It describes what each product does, from public documentation; it does not rate, score or rank vendors, and listing is not a recommendation. Two descriptions help separate products that sound alike:

Spend controls it documents
A reported budget, an alert and a limit that stops consumption are different controls. Described entries show one of: not a spend control, reports spend, alerts on thresholds, enforces limits.
Value evidence it is designed to record
Which kinds of evidence a product records: adoption and usage, estimated time savings, measured process change, realised financial benefit. Adoption is not time saved, and neither is realised financial benefit.

Both are taken from public documentation for entries added or revised on 8 October 2026; other entries have not yet been described this way. Confirm capabilities and pricing with the vendor. Read how the map is built.

Gaps

Where the market is still thin

Gaps in what we could verify from public documentation, not proof that no product addresses them. Several vendors announced capabilities in these areas during 2026 that have not yet been assessed in use.

  1. 1One view from model call to business outcome. No vendor covers the full chain; expect to integrate.
  2. 2Cost attribution for agents that call several models and tools in one task.
  3. 3A cross-vendor view of workforce AI spend and shadow AI. Microsoft and OpenAI now report on their own tools, but rarely on anyone else's.
  4. 4Quality-adjusted cost per outcome. Evaluation tools now report cost beside quality, but rarely beside business outcomes.
  5. 5Live FinOps data flowing into TBM and ITFM service models.
  6. 6Forecasting AI cost from adoption curves and agent scaling, rather than last month's bill.

Scope

What is in and out

Buyer-side only

Tools an enterprise uses to govern AI it consumes. Seller-side billing platforms such as Metronome (now Stripe, completed 14 January 2026), Orb, Lago and Amberflo are out of scope.

No GPU clouds

CoreWeave, Lambda, Together, Fireworks, Baseten, Modal and similar are supply, not governance. They appear in notes where relevant.

CloudHealth not assessed

A scope decision, not a judgement. Its AI-specific cost capability was not assessed for this edition; existing customers should check current documentation.

One primary category per entry

Each product appears once, under its primary buyer problem. Secondary roles show as capability tags and cross-references, and product entries name their parent company. Bedrock and Azure OpenAI sit inside the AWS and Azure entries.

One evidence role in the AI value lifecycle sits outside this map: AI risk and model-governance platforms. They supply approval evidence rather than AI cost and value control, so they are not profiled here.

Missing a vendor, or found an error? Corrections are logged publicly on the corrections page. Changes to this map are listed in the changelog.

Related reading

Author and disclosure

I'm Ed Epstein, an enterprise account executive at IBM in the Apptio portfolio (Apptio, Cloudability, Turbonomic), and I write and edit AI Economics Hub. IBM sells products in several areas this site covers, including FinOps, technology business management and AI value measurement, and is a member of industry bodies the site discusses, including the Tokenomics Foundation. This is an independent personal publication. The postings on this site are my own and don't necessarily represent IBM's positions, strategies or opinions. No IBM, client or partner confidential information is used. IBM-owned products are marked wherever they appear, and the vendor map describes products without rating them.