Akamai Inference
by Akamai Technologies
Managed open-weight model inference on Akamai Cloud with an OpenAI-compatible API, serverless or dedicated GPUs
Akamai Inference is a managed service for running open-weight and custom AI models on Akamai Cloud through an OpenAI-compatible API, with serverless pay-per-token or dedicated GPU endpoints. It is for platform and engineering teams that want production inference close to users without operating their own GPU serving stack.
Akamai Inference is a managed model-serving service on Akamai Cloud, the public cloud Akamai built after buying Linode for about $900M in 2022. Akamai announced it in beta on October 27, 2026, together with Akamai Agent Sandbox, as part of its AI Cloud platform. The service exposes an OpenAI-compatible API, so existing OpenAI-based code moves over by changing the base URL and API key and picking a catalog model. Akamai runs the GPUs, serving, autoscaling and model lifecycle. There are two modes: Serverless Inference on a shared autoscaling endpoint, to be billed per million input and output tokens, and Dedicated Inference on GPU capacity reserved for one customer with a chosen region and GPU configuration, to be billed per GPU-hour by the minute including idle time. The launch catalog has 12 open-weight models from 8 billion to 2.8 trillion parameters, including NVIDIA Nemotron 3 Ultra, OpenAI gpt-oss-120b and Google Gemma 4, plus embedding models that connect to Akamai's managed databases for retrieval-augmented generation. Akamai states zero data retention: prompts and responses are not stored or used for training. Serverless is free during the beta, and Dedicated is in limited availability. Underneath sits the Akamai Inference Cloud built with NVIDIA in October 2025, using RTX PRO 6000 Blackwell Server Edition GPUs, with Blackwell Ultra clusters added at this launch and Vera Rubin planned for 2027. Akamai says its network spans more than 4,350 edge locations. It competes with inference clouds such as Together AI, Fireworks AI and DeepInfra and with edge platforms such as Cloudflare Workers AI.
Platform engineering lead already on Akamai's CDN or security stack who wants open-weight model serving on the same vendor and network
Swap an OpenAI client's base URL to run open-weight models on managed GPUs, with zero data retention and a choice of shared or reserved capacity
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Usage-based, Contact for pricing
- Target Market
- CTOs, Platform Engineering Teams, Enterprise Developers, ML Engineers
- Deployment
- Cloud-only, API-based, Edge-first
- Team Size
- 500+
Key Features
- ✓OpenAI-compatible API
Existing OpenAI SDK code switches by changing the base URL and API key, which keeps migration cost close to zero.
- ✓Serverless Inference
A shared autoscaling endpoint billed per million input and output tokens, with no GPU infrastructure for the customer to manage.
- ✓Dedicated Inference
A private endpoint on reserved GPUs with chosen region and configuration, avoiding shared-tenancy queues and rate limits for steady workloads.
- ✓Open-weight model catalog
Twelve launch models from 8B to 2.8T parameters, including Nemotron 3 Ultra, gpt-oss-120b and Gemma 4, plus embedding models.
- ✓Zero data retention
Prompts and responses are not stored or used to train models, which simplifies privacy review for regulated data.
- ✓Akamai Agent Sandbox companion
Agents run on Akamai Functions while their generated code runs in WebAssembly isolation with limits on compute, memory and outbound connections.
- ✓Audit logging
Logs control-plane operations and inference calls for security review, currently in limited availability.
Capabilities
Use Cases
- •Swap a closed model for open weights
A product team points its OpenAI client at Akamai and serves gpt-oss-120b instead, cutting dependence on one proprietary model provider.
- •Low-latency customer-facing assistant
A consumer app serves chat responses from Akamai's distributed GPU regions so users in several geographies get faster replies.
- •Grounded RAG over enterprise data
Engineers use Akamai embedding models with managed databases to ground answers in internal documents without sending data to another cloud.
- •Reserved capacity for steady traffic
A team with predictable volume buys Dedicated Inference to control model versions and deprecation timing and avoid shared rate limits.
- •Safe execution of agent-written code
An agent built on Akamai Functions runs generated code inside Agent Sandbox so a bad script cannot reach arbitrary network destinations.
Ideal For
Best For
- ✓Teams moving OpenAI-compatible applications to open-weight models such as gpt-oss-120b or Gemma 4 without rewriting client code
- ✓Latency-sensitive inference for user-facing apps that benefit from Akamai's distributed network
- ✓RAG applications that want embedding models next to Akamai managed databases and object storage
- ✓Enterprises with existing Akamai contracts that want to consolidate AI serving with an incumbent vendor
Not Ideal For
- ✗Teams that need frontier proprietary models; the catalog is open-weight and custom models only
- ✗Large-scale training or multi-node GPU clusters; SemiAnalysis rated Akamai's GPU cloud Underperforming in ClusterMAX 2.0, citing no managed Slurm and a Kubernetes engine not tuned for GPU servers
- ✗Production workloads that need a published per-token price and SLA today, since Serverless is still in beta with pricing not yet announced
Integrations
Deployment
Market Analysis
Pros
- ✓Drop-in OpenAI-compatible API keeps switching cost low
- ✓Zero data retention stated on the product page
- ✓Serverless tier is free during beta, which makes evaluation cheap
- ✓Published GPU instance prices for the underlying Blackwell hardware
Cons
- ✗Still beta: Serverless pricing is unannounced and Dedicated has limited availability, and OpenTools notes no versioned compatibility matrix was published
- ✗Akamai's product page warns that tool calling and structured outputs vary by model, so an OpenAI-compatible endpoint is not a guarantee of feature parity
- ✗SemiAnalysis's ClusterMAX 2.0 rated Akamai/Linode GPU cloud Underperforming, citing missing high-end GPUs, no managed Slurm and a Kubernetes engine not optimized for GPU servers
- ✗No independent latency or throughput benchmarks were published at launch
Pricing
Serverless Inference (beta)
$0
- ✓No charge during beta
- ✓Per-million-token pricing planned at limited availability
- ✓Shared autoscaling endpoint
Dedicated Inference
Contact for pricing
- ✓Reserved GPU capacity
- ✓Per-GPU-hour billing by the minute, including idle time
- ✓Choice of region and GPU configuration
RTX PRO 6000 Blackwell GPU instance (self-hosted)
From $3.00/hr
- ✓1 GPU, 96 GB VRAM
- ✓Up to 8 GPUs at $24.00/hr
- ✓Egress overage $0.005/GB
Serverless Inference costs nothing during the beta; Akamai says it will move to per-million-token pricing at limited availability but has not published rates. Dedicated Inference bills per GPU-hour by the minute, idle time included. For self-hosted serving, Akamai's GPU price list shows RTX PRO 6000 Blackwell instances from $3.00/hr for one GPU to $24.00/hr for eight; HGX B300 is quote-only and prices vary by region.
Security & Compliance
Sources
This page was written from 7 sources, 5 on domains other than akamai.com.
- 1.akamai.com — akamai inferencevendor
- 2.akamai.com — gpuvendor
- 3.siliconangle.com — akamai expands its ai cloud platform with dedicated inferenc
- 4.itbrief.co.uk — akamai launches ai inference agent sandbox tools
- 5.opentools.ai — akamai inference agent sandbox beta spin sdk serverless octo
- 6.clustermax.semianalysis.com — akamailinode
- 7.hn.algolia.com — search
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Grafana Assistant
AI observability agent in Grafana Cloud that writes PromQL, LogQL and TraceQL from plain English, investigates incidents and proposes code fixes as pull requests
Inferact Platform
Managed vLLM inference from the creators of vLLM, on your own GPUs or on Inferact Cloud
Ollama
Run open-weight LLMs locally with one command, then burst to Ollama's cloud on the same API
Perplexity SPACE
Open-source Firecracker microVM sandbox for AI agents, the Apache 2.0 runtime behind Perplexity Computer