Stack map · updated 2026-08-24
The gateway layer— the routing, caching, fallback and spend control between your app and the models — is consolidating fast in 2026. Stripe is folding in OpenRouter; Vercel, Databricks, Cloudflare and even Ramp are building their own. It's the least glamorous and most strategic part of the AI stack: whoever owns the meter owns the margin. Here's who's competing, segmented by who they really are.
Living map of the AI-gateway / model-routing layer. Facts cross-checked from public 2026 sources; deal figures reflect reported ranges and are labeled.
Startups whose whole product is the routing/observability/spend layer between apps and models. Two of the biggest just got acquired.
Independent → Stripe (acquiring)
Model marketplace + unified API over 500+ models / 80+ providers; routing, fallback, unified billing
$113M Series B @ $1.3B (CapitalG, May 2026)
Reported $7B+ Stripe deal; ~5.5% markup pricing
managed
Palo Alto Networks (acquired)
AI gateway + observability, guardrails, caching, fallback — now agent security/governance
$15M Series A (Elevation) pre-deal
Closed May 2026; folded into Prisma AIRS
managed
BerriAI (YC)
Open-source unified API/proxy across 100+ providers; self-host or managed
$1.6M seed · ~$7M ARR
The developer default for self-hosting
open-source + managed
Helicone (YC W23)
Open-source LLM observability + gateway (proxy)
seed
open-source + managed
Kong
API-gateway incumbent extended for LLM traffic; open-sourced AI Gateway
$175M @ $2B valuation
Rides 30k+ existing Kong deployments
managed + self-host
Not Diamond
Intelligent model routing / evaluation-driven selection
$2.3M
$0.05 per 1M tokens routed
managed
TrueFoundry
Enterprise AI gateway + MCP gateway + ML infra
managed + self-host
Martian
Model-routing research; now positioned as a research org more than a product
Watch — product focus shifted
managed
Unify (YC W23)
Started as an LLM router; pivoted to AI agent teammates
$8M
Pivoted away from pure routing
managed
Sapiom
Agent cost-routing to cheapest capable model (covered in Proof of Stack)
$35M Series A (Dragonfly; Anthropic co-invest)
Cut one customer's bill ~10×
managed
Dev-platforms and data clouds bundling a gateway to keep builders inside their ecosystem.
Vercel
Unified model access inside the Vercel/AI-SDK stack; routing, fallback, observability, public model leaderboards
GA (runs on Fluid compute). Angle: developer-experience lock-in — keep AI-app builders inside the Vercel stack, prototype to production.
managed
Databricks
Model serving + gateway + agent governance inside the data platform
Shipped after reportedly losing the OpenRouter bid
managed
AWS · Microsoft · Google
Hyperscaler model access layers with routing/guardrails
managed
CDN incumbents moving up-stack — inference at the edge, gateways at the network layer.
Cloudflare
Gateway with caching, analytics, rate-limiting; pairs with Workers AI inference
The most mature edge-native gateway
managed
Akamai (Zuplo-powered)
Zuplo-powered AI & API gateway on Akamai Cloud — native OpenAI / Claude / Gemini + API/MCP governance
Angle: edge distribution + API/AI monetization (AccuWeather is the proof customer). Zuplo tech, Akamai reach. Announced Dec 2025.
managed
Fastly
Edge platform; AI-delivery positioning
Edge play; dedicated gateway unconfirmed
managed
The surprise entrants — payments and spend platforms owning the metering layer. Whoever owns the meter owns the margin.
Stripe
Model gateway folded into a payments giant — metering + billing for AI tokens
Collison: 'Tokens are the central currency for AI'
managed
Ramp
Launched its own LLM gateway (spend-management angle)
Fintech eating the AI-spend layer
managed
The money backing the gateway/routing layer — worth tracking, because they signal where the next moat forms.
The Autopilot Index tracks the tech stack behind companies run by AI — gateways, agents, sandboxes and the money moving them. Weekly, sourced, free.
A living map, updated as the space moves. Deal figures are labeled and sourced in the companion edition. Not investment advice.