Just launched: AI Infrastructure Optimization → Read more

Akamas

MORE COMPUTE WON’T FIX A BAD CONFIGURATION

Autonomous
Kubernetes and AI optimization

Akamas is the autonomous optimization platform for Kubernetes and AI infrastructure. Akamas applies continuous, full-stack Kubernetes optimization from the application runtime down to the GPU. Every recommendation carries the evidence behind it and is ready to apply. Akamas improves performance, fixes reliability issues, and lowers infrastructure cost.
Full stack JVM·Node.js·.NET·vLLM·pod requests·node groups·GPU memory
Calculate Savings Create Account
Kubernetes Multiple clusters
– 58 %
Infrastructure cost
2
Weeks to result
Enel
Sabre
LastMinute.com
Sisal
Raiffeisen
Sogei
TeamSystem
Navan

The Akamas Platform

Three optimization modes.
One AI Engine.

The same engine that learned to optimize Kubernetes now tunes AI inference stacks and any parameter space you point it at.

In production · Full-stack optimization

Full-stack Kubernetes Cluster to application runtime

Akamas provides continuous, full-stack optimization for production Kubernetes environments. It plugs into the observability you already run, analyzes full-stack data, and delivers ready-to-apply recommendations: pod rightsizing, Java Virtual Machine (JVM), .NET, and Node.js runtime settings, Horizontal Pod Autoscaler (HPA) policies, node groups.
  1. 1 Integrate Plug into Dynatrace, Datadog, or Prometheus. No agents, no code changes.
  2. 2 Akamas analyzes The engine processes full-stack data and produces recommendations with the evidence behind them
  3. 3 Production tuned Review and apply, or automate with guardrails, via GitOps if that's how you ship.
Explore Akamas Vivo →
Customer result Without Akamas
⚠ used > requests
App demand
Used
Requests
Allocatable
① Application efficiency ② Workload efficiency ③ Cluster efficiency
+23% Peak transaction volume
−20% Operating costs

lastminute.com case study.

Every layer at once · Automated optimization

AI Inference and GPUs LLMs, GPUs, and inference parameters

Akamas provides automated optimization for AI inference stacks. That means the model, the serving engine, the GPU, Kubernetes, and the instance underneath. It runs experiments across every layer simultaneously, tuning batching, concurrency, and the key-value (KV) cache together. The target is your real workload, not just the hardware and the model.
  1. 1 Connect the stack Integrate your inference server, Kubernetes, and GPUs in a dedicated environment.
  2. 2 Akamas iterates Experiments across every layer at once, not one at a time.
  3. 3 Validated configuration Reviewed, explained, and ready to promote to production.
Explore Akamas for AI →
vLLM on NVIDIA GPUs · benchmark Without Akamas
+57.5% Decode throughput per GPU
+54.1% Prefill throughput per GPU

Methodology in the solution page.

Test-driven · Tailored optimization

VMs and Data platforms Any stack, optimized under meaningful load, expert-configured

Akamas provides pre-production optimization for any stack with parameters. Every stack has its own tuning language, and few teams speak them all. You scope the tunable parameters and the success metric with our engineers. Studio then runs your own performance tests against each candidate configuration. Studio keeps only the configurations that measured results prove. Nothing touches production.
  1. 1 Scope with Akamas Define the tunable parameters and the goal for your stack, together with our engineers.
  2. 2 Akamas tests, optimizes, and iterates Each candidate configuration runs against your own performance test, the engine learns from the observed metrics, and recommends the best configuration.
  3. 3 Validated configuration The optimal configuration proven under your real load, reviewed, and ready to be promoted to production.
Explore Akamas Studio →

One engine, any parameter space

Java · heap & garbage collection Spark · executors Kafka · consumers Go · runtime NodeJS · runtime .NET · heap & runtime KubeVirt · VM resources AWS · EC2 GCP · GCE Azure · VMs PostgreSQL MongoDB Your custom tech

Whatever you developed in house, if it has parameters, the engine can optimize it.

WHY THIS IS HARd

The tuning space explodes faster than any human can search it

One service means roughly ten parameters with ten values each. Hundreds of interacting services share clusters and node pools, and the target moves with every release. Meanwhile ownership is scattered: platform teams tune clusters, application teams ship features, SREs manage incidents, FinOps sees costs after the fact.

Observability tells you what’s happening. Someone still has to act. That’s the missing platform capability.

1010

Possible configurations for a single service

~10 parameters × ~10 values each. Now multiply by hundreds of interacting services, and re-tune at every release.

The impact

Optimization stops being one person’s job

Today it lives with whoever happens to know the system best. It should be a capability every team shares.

Team Without Akamas With Akamas
Developers Pick a limit, ship it, hope it holds. Yes Optimal config arrives validated, in the PR.
SREs Find the reliability risk during the incident. Yes See it before, with the evidence to act on it.
Platform Engineering Static rules that age the day they ship. Yes Full-stack signal, applied at cluster scale.
FinOps A savings report nobody implements. Yes Savings tied to changes that actually get applied.
Akamas named Outperformer in GigaOm Cloud Resource Optimization report
Akamas helps us size our pods correctly and address configuration issues that often emerge with new services. It also fills a cross-team skills gap between developers and our Kubernetes administrators, delivering significant reliability improvements, and the cost savings donʼt hurt either.

Gabriele Bosisio

Head of Operations Reliability & Security, Sisal

Sisal
Akamas helped us to rapidly mature on the performance tuning front, by allowing us to find an optimal configuration for our application. This resulted in significant cost savings as well as removing barriers to replatforming.

Chris Cholette

VP Productivity and Site Reliability Engineering, Navan

Navan
Thanks to Akamas, TeamSystem has improved the efficiency of our critical microservices as we would never be able to do manually. The ability to consistently deliver the highest level of quality to our end-users at the lowest possible cost is an important differentiator for us.

Luca Montecchiani

Lead Software Architect, Product Owner, TeamSystem

TeamSystem
Within just a few hours, Akamas uncovered performance issues we had overlooked for months. This wasn’t just an improvement – it was a revelation. Akamas delivered insights we didn’t even know we needed and solved problems faster than any manual approach could.

Damjan Kumin

Chief Technology Officer, Perform IT

Akamas’ ease of integration with our CI/CD pipelines enabled us to automate the configuration deployment to quickly find optimized configurations that had not been previously found with our manual approach.

Gartner Peer Insights

Customer in Online Services

Questions, answered

What people ask about Akamas

Autonomous full-stack optimization is the continuous, AI-driven tuning of every layer an application depends on. Those layers are pod resources, runtimes like the JVM and Node.js, autoscaling policies, node groups, and GPUs. Akamas treats them as one system. A change in one layer shifts the optimum of the others.
Only if you want it to. Akamas produces ready-to-apply recommendations with the evidence behind them: you can review and apply each change yourself, or automate application with guardrails, including delivery through GitOps. Human-in-the-loop is the default, autonomy is the option.
Observability and FinOps tools show what is happening and what it costs. Akamas goes beyond visibility: it computes the better configuration and gets it applied. It plugs into Dynatrace, Datadog, or Prometheus, complementing your existing monitoring stack rather than replacing it.
Kubernetes workloads end to end: pod and container rightsizing, JVM and Node.js runtimes, autoscaling, node groups. AI inference stacks: the vLLM serving engine, GPUs, batching and cache parameters. And through Akamas Studio, pre-production optimization for any stack with parameters: Spark, Kafka, Go, KubeVirt, Amazon EC2, or software you built in house.

Blog

Tech deep dives, product news, and Akamas stories – all in one place

Your GPUs Are 95% Idle, and Buying More Won’t Fix It
Blog
Your GPUs Are 95% Idle, and Buying More Won’t Fix It

AI is reshaping every industry, and agents are at the center of that shift. Many…

The Michelin Star Dilemma and the Secret to a Perfect Service
Blog
The Michelin Star Dilemma and the Secret to a Perfect Service

In the world of fine dining, the tension in the kitchen is palpable long before…

The State of Cloud Native Optimization 2026
Blog
The State of Cloud Native Optimization 2026

A critical realization has emerged within the platform engineering community. We have excelled at creating…

Kubernetes Performance Tuning, Beyond One-Size-Fits-All: Introducing Tuning Profiles
Blog
Kubernetes Performance Tuning, Beyond One-Size-Fits-All: Introducing Tuning Profiles

Modern Kubernetes clusters are not monolithic; they host many applications with very different operational needs….

Why Your Kubernetes Cluster Autoscaler Wastes Resources, and How to Fix It
Blog
Why Your Kubernetes Cluster Autoscaler Wastes Resources, and How to Fix It

Platform engineers rely on the Kubernetes cluster autoscaler to control cloud cost. However, the autoscaler…

Beyond the JVM: Full-Stack Optimization for Node.js on Kubernetes
Blog
Beyond the JVM: Full-Stack Optimization for Node.js on Kubernetes

The promise of Kubernetes lies in seamless scalability, yet running high-performance Node.js applications often reveals…

View all
Akamas named a Leader in the GigaOm Radar for Cloud Resource Optimization v5, 2026
Akamas named an Outperformer in the GigaOm Radar for Cloud Resource Optimization v5, 2026

See for Yourself

Experience the benefits of Akamas autonomous optimization.
No overselling, no strings attached, no commitments.

Calculate Savings Calculate Savings Create Account Create Account