Poolside didn't get the deal they wanted. The AI startup, known for its ambitious plan to rewrite how we build software, is back on the hunt for massive data centre capacity. This comes after high-stakes negotiations with CoreWeave, the darling of the GPU-as-a-service world, reportedly ended without a signature. It's a reminder that in the gold rush for AI, owning the pickaxe is just as important as having the vision.
The hardware bottleneck is real. You can have the most sophisticated coding models in the world, but if you don't have the H100s or Blackwell chips to run them, you're just sitting on expensive code. Poolside, which recently moved its base to Paris to tap into the European ecosystem, now has to find a new home for its compute-hungry workloads. They're looking for partners that can provide the sheer scale required to train models that don't just suggest lines of code, but understand entire software architectures.
Why the CoreWeave breakup matters for the AI market
CoreWeave has become the "it" company for AI infrastructure. Backed by Nvidia, they have access to the hardware everyone else is begging for. When a deal with them falls through, it’s usually for one of two reasons: price or priority.
Maybe the terms didn't align with Poolside's aggressive growth trajectory. Or perhaps CoreWeave found a bigger fish—think Microsoft or Meta—willing to sign a multi-year, multi-billion dollar lease for every spare rack they have. For Poolside, this isn't just a minor delay. It's a strategic pivot.
Data centres are the new oil refineries. You can't just build them overnight. They require massive power permits, cooling infrastructure, and physical security. By losing out on a partnership with a specialized provider like CoreWeave, Poolside has to look at the broader market. This includes the hyperscalers—Google Cloud, AWS, and Azure—but also smaller, sovereign cloud providers in Europe that might offer better regulatory alignment for their Paris-based operations.
The struggle for European compute power
Poolside's move to France wasn't accidental. The French government is throwing everything at the wall to become the AI capital of Europe. President Emmanuel Macron has been vocal about wanting "AI champions" born and raised on the continent. But there's a problem. Europe is behind on the physical infrastructure.
Most of the world's GPU clusters are concentrated in the United States. If Poolside wants to keep its data and training local to satisfy EU "digital sovereignty" goals, its options are limited. They need thousands of GPUs. We're talking about a scale that most regional providers simply can't handle.
This puts the startup in a tough spot. Do they compromise on their European identity and lease space in a Virginia or Texas data centre? Or do they wait for European providers like OVHcloud or Scaleway to ramp up their high-performance computing (HPC) offerings? Waiting in the AI world is a death sentence. Your model is obsolete before the first epoch finishes training.
What Poolside actually needs from a partner
It's not just about "cloud space." That's a generic term that means nothing in this context. Poolside needs low-latency networking and specialized cooling.
The hardware requirements are insane
Training a large language model (LLM) for coding isn't like hosting a website. It requires thousands of GPUs working in perfect synchronization.
- InfiniBand networking: Standard ethernet doesn't cut it. You need specialized networking to make sure the GPUs can talk to each other without delays.
- Power density: Modern AI racks pull 50kW to 100kW of power. Most older data centres would literally melt if you tried to plug these in.
- Sustainability: Especially in Europe, the optics of burning megawatts of power matter. Poolside likely wants a partner with a high Power Usage Effectiveness (PUE) rating.
The valuation pressure on AI startups
Poolside raised a massive $500 million seed round. That kind of money comes with massive expectations. Investors aren't paying for "research." They're paying for a product that can challenge GitHub Copilot or Replit.
Every month spent "hunting for partners" is a month of burn rate without progress. The AI market is currently in a "show me" phase. The initial hype is cooling, and people want to see actual tools that work. If Poolside can't secure the compute, they can't train the next version of their model. If they can't train the model, they can't ship.
It's a high-stakes game of musical chairs. Right now, there are more startups than there are chairs (GPUs). Poolside is currently standing up, and the music is still playing, but they need to find a seat fast.
Who else could step up
If CoreWeave is off the table, who's left?
Lamba Labs is a strong contender. They specialize in GPU clusters and have been growing nearly as fast as CoreWeave. Then there's the possibility of a direct deal with a hardware manufacturer or a specialized "sovereign cloud" play.
There's also the "big three." But dealing with AWS or Google comes with strings. They often want equity or "strategic partnerships" that might limit where Poolside can sell its software. Plus, the cost of egress—moving data out of their clouds—is notoriously high. Once you're in, you're stuck.
What this says about the 2026 AI outlook
The fact that a well-funded, high-profile startup like Poolside is struggling to lock down a partner tells us the "compute crunch" isn't over. Even with Nvidia pumping out chips at record speeds, the demand is insatiable.
We're seeing a bifurcation in the market. There are the "haves"—companies with direct lines to Nvidia or their own silicon—and the "have-nots" who are constantly negotiating for scraps of compute time.
For Poolside to survive, they probably need to look beyond the obvious players. They might need to piece together a "Frankenstein" cluster across multiple smaller providers or look into emerging markets where power is cheaper and data centre space is less contested. It's a messy way to build a company, but "clean" options are currently sold out.
How to track the next move
If you're watching this space, watch the job boards and the shipping manifests. You'll know Poolside found a partner when they start hiring "Data Centre Operations" staff in specific geographic locations.
They also need to be transparent with their developer community. If the training is delayed, the product roadmap slips. For a company trying to win the hearts and minds of software engineers, trust is the only currency that matters.
The next few months are critical. They either find the hardware to back up their $3 billion valuation, or they become another cautionary tale of a startup that had the brains but couldn't find the brawn.
Stop looking at the software demos for a minute. Look at the power grids and the cooling pipes. That's where the real AI war is being fought. Poolside just lost a battle, but the war for infrastructure is just getting started. If you're building in this space, secure your compute before you hire your first engineer. Otherwise, you're just paying people to wait for a server that might never show up.
