ScaleOps Just Raised $130M Because Most Kubernetes Clusters Are Half Empty

Here is a number that should make every CTO uncomfortable: the average enterprise Kubernetes cluster runs at somewhere between 30% and 50% utilization. That means companies are essentially paying double for their cloud infrastructure, and they have been for years.

ScaleOps, a Tel Aviv and New York-based startup, just closed a $130 million Series C at an $800 million-plus valuation to tackle exactly this problem. The round was led by Insight Partners, with existing investors Lightspeed Venture Partners, NFX, Glilot Capital Partners, and Picture Capital all participating. Total funding now sits at $210 million.

But the timing here matters more than the dollar amount. This raise comes at the exact moment when AI workloads are turning Kubernetes cost sprawl from a quiet inefficiency into a board-level crisis.

The Overprovisioning Tax

I see this pattern constantly. A development team is deploying a new service. They do not know exactly how much CPU or memory it will need in production, so they set generous resource requests and limits. Better safe than sorry, right?

Multiply that decision across hundreds of services, dozens of teams, and several clusters. Now you have an organization where 31% of workloads run below 25% CPU usage for 95% of the day. The safety margin becomes the default, and nobody revisits it because the application works fine.

Think of it like heating a warehouse to keep one office warm. The thermostat works. But the gas bill is absurd.

According to a recent CIO survey, 88% of Kubernetes practitioners report increased total cost of ownership in the past year. And 68% of organizations overspend on Kubernetes by 20-40% or more, often due to exactly this kind of configuration drift.

Why AI Made This Urgent

Overprovisioning a CPU-bound web service is wasteful. Overprovisioning a GPU-bound AI training job is catastrophically expensive.

GPUs cost 10-50x more per hour than standard compute instances. When teams overprovision GPU resources using the same "better safe than sorry" mentality, the waste compounds at a completely different scale. One misconfigured AI training pipeline can burn through more cloud budget in a week than a traditional microservice burns in a quarter.

ScaleOps claims their platform can deliver up to 80% reduction in cloud and AI infrastructure costs through continuous, autonomous optimization. That number sounds aggressive, but when your starting point is 30-50% cluster utilization, there is a lot of room to claw back.

What ScaleOps Actually Does

The company positions itself as building a "Cloud Operating System for the AI era." In practice, their platform covers several interconnected optimization layers:

• Pod rightsizing: continuously adjusting CPU and memory allocations based on real-time usage, not static developer estimates

• Replica optimization: scaling the number of pod replicas based on actual demand patterns rather than worst-case assumptions

• Node management: optimizing the underlying instance types and bin-packing workloads more efficiently

• Spot instance optimization: intelligently leveraging spot/preemptible instances for workloads that can tolerate interruptions, which in 2026 deliver 70-90% savings

• GPU reclamation: automatically rightsizing GPU allocations and reclaiming idle GPU resources across AI training and inference workloads

The key differentiator, at least on paper, is that these optimizations happen autonomously and in real time. Traditional tools like Kubernetes VPA require manual tuning and often cannot apply changes without restarting pods. ScaleOps claims to handle this without disruption.

The Numbers Behind the Hype

ScaleOps is reporting 350%+ year-over-year revenue growth. Their customer list includes Adobe, Salesforce, Wiz, DocuSign, and Coupa. The company, founded in 2022, has grown to 120+ employees across Israel, North America, and Europe, and tripled headcount in the past 12 months.

CEO Yodar Shafrir previously worked as an engineer at Run:ai, which was acquired by Nvidia. That background in GPU orchestration clearly shaped ScaleOps' expansion beyond traditional CPU/memory optimization into AI infrastructure management.

The platform is already available on AWS Marketplace, Azure Marketplace, and Google Cloud Marketplace. It is also FIPS-compatible for FedRAMP, which signals serious enterprise and government pipeline ambitions.

The Competitive Landscape

ScaleOps is not operating in a vacuum. Cast AI, Kubecost, and Spot (now part of NetApp) all play in the Kubernetes cost optimization space. The cloud providers themselves are building native optimization tooling. And 92% of practitioners surveyed are already investing in AI-based cost optimization tools, according to recent industry data.

But the market is growing faster than any single vendor can capture. Platform engineering salaries now approach $200,000 annually, and organizations are also absorbing the hidden costs of developer productivity lost to manual cluster management. When the labor costs of maintaining infrastructure manually start rivaling the infrastructure costs themselves, autonomous optimization becomes the obvious investment.

My Takeaway

I believe the ScaleOps raise is a strong signal, but not because of the company itself. The real story is that Kubernetes cost management has become urgent enough to command $800M valuations.

For years, overprovisioning was a rounding error in the IT budget. AI workloads changed the math. When a single GPU hour costs more than a full day of traditional compute, the margin of error shrinks dramatically. Companies that were comfortable carrying 50% waste on $500K/month cloud bills are now staring at $2M/month GPU invoices with the same utilization patterns.

Whether ScaleOps specifically wins this market remains to be seen. But the category itself, autonomous Kubernetes optimization, is clearly here to stay. The 80% cost reduction claim needs to be validated in practice at each organization. In my experience, real-world savings from these tools tend to land in the 40-60% range after you account for the workloads that genuinely need their overhead.

Still, even a 40% reduction on a seven-figure monthly cloud bill pays for itself in the first invoice cycle. And that is why investors just wrote a $130 million check.