Cast AI

Kimchi Coding is GA today. Six months ago, an engineering leader I respect showed me his AI coding bill. It was growing faster than his headcount. Every task, even trivial refactors, was going to the most expensive frontier model on the market. He knew it was wasteful, and the only fix available was paying someone to babysit model selection all day. We built Kimchi so nobody has to do that. It's an autonomous multi-model coding agent. It routes each task to the best-fit model and scores the code at every step. It runs on our own inference cloud, because whoever controls inference controls token cost. Against a commercial-models-only baseline it came out 2.5x cheaper, with equal or better quality. We ran it on 150 of our own developers before asking anyone else to trust it. It also ships with hard spend caps, automatic termination of runaway agent loops, and a FinOps dashboard that shows who spent what, down to the token. We spent six years squeezing waste out of Kubernetes and GPUs. Kimchi is that engine pointed at AI coding. The full story is here: kimchi.dev/blog/kimchi-co… Try it at kimchi.dev. Happy coding. From today, tokens are cheaper. Thank you to my team, to our customers, and to our investors. The journey was long, but I can tell you that it was worth it.

Cast AI is now live in Mumbai 🇮🇳 Indian enterprises and global orgs running AI workloads in India can now automate their cloud infrastructure without leaving the country. Read more: cast.ai/press-release/…

@MiniMax_AI M3 lands on @getkimchi Coding by Cast AI, the first autonomous coding agent to launch it. Frontier open-weight performance on Kimchi CLI. Read more: cast.ai/press-release/…

FinOps X 2026 was a milestone week for the industry and for Cast AI! Our cofounder and president Laurent Gil was in the room when the Tokenomics Foundation became official. He sat down with theCUBE to talk about what tokenomics actually means for infrastructure teams GPU economics, AI spend governance, and where this is all heading. Worth a watch. Our own Esmira Bayramova also took the stage, covering GPU utilization data, token economics, and what teams running AI workloads at scale need to start thinking about now. The conversation at FinOps X made one thing clear: the "figure out cost later" era is over. The FinOps community is ready to own this. We had a booth on the floor too and the energy from practitioners was exactly what you'd hope for at a moment like this. Thank you San Diego. See you at Tokenomicon! 👋

Learn more: cast.ai/blog/opspilot-…

Dev and staging are running with the same resource floor as production. Everyone in your company knows it, but nobody has ever had a practical way to fix it at scale. That changes today. Cast AI OpsPilot now generates your workload scaling policies automatically, based on what it actually finds in your cluster. Three questions to set your intent, one review step, and that's all it takes. Available now for all Cast AI customers. Full breakdown in the first comment.

Learn more: cast.ai/press-release/…

For years, Kubernetes workload optimization has meant the same thing: engineers writing policies, keeping them current, and absorbing the cost when they inevitably go stale. Cast AI OpsPilot changes that. It reads your cluster's actual state, generates workload scaling policies calibrated to what it finds, and keeps pace as your infrastructure evolves. The days of manually authored rules are behind us. Full announcement in the first comment.

For twenty years, compute followed a predictable path: wait a year, and the same power would cost less. In January, that trend officially broke. AWS increased its H200 Capacity Block prices by 15%, breaking a two-decade precedent of falling compute costs. As organizations rush to secure AI hardware, high-end compute is actually getting more expensive. Yet, the 2026 State of Kubernetes Optimization Report reveals a massive efficiency gap. While companies hoard hardware out of scarcity fears, raw utilization remains incredibly low across production environments : • Average CPU utilization: 8% • Average memory utilization: 20% • Average GPU utilization: Just 5% These are direct measurements taken across tens of thousands of production Kubernetes clusters on AWS, GCP, and Azure. It means 95% of highly competitive GPU capacity is sitting completely idle, costing dollars per hour while doing nothing. In a market where hardware costs are actively rising, static provisioning is no longer a viable engineering strategy. Our co-founder and president @laurentgil broke down the core numbers and structural issues behind this mismatch in Node Magazine. Read his full breakdown to see how your team can avoid the AI cost trap: node-magazine.com/thoughtleaders…

Even OpenAI’s Sam Altman admits there is a ton of waste in current AI spending. As organizations pour billions into infrastructure, the critical question is how to get these costs under control. Our 2026 State of Kubernetes Optimization Report provides the hard data behind this reckoning: after analyzing 23,000 computing clusters, we found that average GPU utilization sits at just 5%. Companies are stockpiling expensive chips, leaving 95% of their available capacity completely idle. We don't need to slow down AI innovation; we just need to fix how we manage the compute behind it. Thanks to Business Insider and Digit India for featuring our report's findings in the broader conversation on AI infrastructure costs. 🔗 Read the full story here: digit.in/news/general/s…

São Paulo showed up in full force 🇧🇷 We designed the first @kubeautoday Day: FinOps Edition in Latin America for deep technical growth, and the response was incredible. With nearly 1,000 registrations, the venue was packed with great energy from start to finish. The sessions were practical and loved by everyone. Speakers shared actual numbers and clear lessons on SRE and cost optimization, and we even had a speaker deliver an unforgettable FinOps rap. It was a major milestone for the regional ecosystem, bringing together @LINUXtipsBR, @CloudNativeFdn, and the @FinOpsFdn. Thank you to our speakers, panelists, and our hosts for driving the conversations. Thank you to our co-sponsors Opsteam, @chainguard_dev, @grafana, and @awscloud for collaborating with us to bring this to life. The LATAM tech community is building something special, and we are proud to support it. See you at the next one.

Our biggest G2 quarter ever: 🥇Leader in Cloud Management Platforms, Cloud Cost Management, Auto Scaling 🏆Leader across Enterprise, Mid-Market, and every major region – Europe, EMEA, India, Asia Pacific Thank you to the 190 customers who reviewed us. 🙏

We had the privilege of welcoming a delegation from the International Monetary Fund (IMF) to the Cast AI office in Vilnius 🇱🇹 As part of its visit to Lithuania, the IMF meets with selected organizations to discuss the country’s economic and business environment. This year, Cast AI was the only technology company invited to take part in these conversations. It was an open and insightful discussion on scaling global technology companies from Lithuania, the future of AI infrastructure, talent, innovation, and the broader business landscape shaping growth across the region. A proud moment for Cast AI, and a meaningful recognition of the company’s journey from Lithuania to becoming a globally recognized AI infrastructure platform.

We’re thrilled to be partnering with Hack Night London today for an intense, in-person hackathon! 🇬🇧 The energy in the room is exactly what we love about the community: 5 hours, 0 fluff, and a focus on shipping. It’s been particularly impressive to see @getkimchi being utilized for inferencing in real-time. This is exactly why we value community-driven engineering. We’re here for the demos and the breakthroughs. Let’s build! 💪

The 2026 State of Kubernetes Optimization Report is out. CPU utilization: 8%. Memory: 20%. GPU: 5%. 🤯 Organizations that closed the gap share one thing in common: they stopped treating resource optimization as a one-time configuration task and started treating it as an automated, ongoing process. An efficient Kubernetes cluster is not a configuration. It's a feedback loop. 👉 Full report: cast.ai/reports/state-…

🏆 Cast AI is #1 in Cloud Cost Management on G2 Ranked #1 out of every platform in the category, based entirely on verified customer reviews: 🌟 98% gave us 4 or 5 stars 🎯 100% say we're headed in the right direction ✅ 93% would recommend us To every customer who reviewed us – thank you! 🙌 cast.ai/press-release/…

@ALLENDigital_In built an AI platform that personalizes learning for every student 🤖 But as they scaled, GPU utilization became a challenge – and alternatives couldn't match the performance their learners needed. Enter Cast AI's AI Enabler: 💡 GPU time-slicing → multiple models, one instance, zero performance trade-offs 📈 Managed GPU deployments → automatic scaling and hibernation 🔋 On-demand + Spot mix → high availability for production The result: dramatically higher GPU utilization and infrastructure built to scale with their mission. 🔗 cast.ai/case-studies/a…

🍇 From Bordeaux vineyards to Kubernetes efficiency – Caudalie's journey continues to impress! Even in an already optimized environment, with Cast, Caudalie achieved: → 40% EC2 cost savings → Automated node lifecycle management → Less operational toil across clusters 📖cast.ai/case-studies/c…

Altruist cut cloud costs by 45% and freed 100+ engineering hours/month with Cast AI. 📉 45%+ savings ⏱️ 85–130 hrs freed monthly 💰 $13–$26 return per $1 spent Full story: cast.ai/case-studies/a…