What's Next for Poolside's Texas Data Center Project?

When a 2GW data center deal falls apart, you don't just lose a tenant; you lose momentum, credibility, and — as Poolside is discovering — your next funding round too.

The AI coding startup's Texas ambitions, once backed by a marquee partnership with CoreWeave and a reported $1 billion commitment from Nvidia, have unraveled with unusual speed. What looked like one of the more audacious infrastructure plays of late 2025 is now a cautionary case study in how quickly AI startup valuations can outpace operational reality.

Project Horizon: The Ambition Was Real

Announced in October 2025, Project Horizon wasn't a modest pilot. Poolside had secured 568 acres in the Permian Basin — a site with a theoretical maximum capacity of 10GW — and structured a deal with CoreWeave as the anchor tenant. The first phase targeted 250MW of capacity under a 15-year lease, scaling to 500MW. The company also planned to develop 2.5GW of solar and wind generation directly on the campus.

For context: 250MW is roughly equivalent to the power consumption of a small city. Locking that into a 15-year lease commitment signals the kind of long-range conviction that institutional investors and hyperscalers take seriously.

Poolside had raised $500 million at a $3 billion valuation in 2024. The CoreWeave partnership, combined with a reported $2 billion funding round — with Nvidia said to be contributing up to $1 billion of that — suggested the company was building toward something genuinely substantial. Founded in 2023 by Eiso Kant and Jason Warner, Poolside develops AI software oriented toward code automation, with security and accuracy standards it claims meet government-level requirements. That use case has real defensibility. The infrastructure bet, it turns out, was harder to execute than the pitch made it sound.

How the CoreWeave Deal Fell Apart

The breakdown, reported by the *Financial Times* citing five people familiar with the matter, comes down to one core failure: Poolside couldn't stand up its first cluster of chips on CoreWeave's timeline.

That's a significant operational miss. CoreWeave's entire business model is built on deploying GPU infrastructure at speed and scale — the company's competitive advantage is precisely its ability to move faster than traditional cloud providers. A partner unable to meet basic deployment timelines isn't just inconvenient; it's incompatible with CoreWeave's operational DNA.

CoreWeave's public statement was diplomatic — the companies "ultimately chose to pursue different paths for their own strategic and timing reasons" — but the substance of the FT reporting tells a more pointed story. When an anchor tenant walks away from a 15-year lease before construction begins, the strategic disagreement is usually about execution, not strategy.

For Poolside, the timing couldn't have been worse. The CoreWeave departure triggered a cascade. Nvidia's reported $1 billion contribution to the $2 billion funding round also evaporated. According to the FT, investors had grown skeptical that Poolside could train AI models competitive with established players — a fundamental credibility problem for a company whose entire valuation rests on that capability.

The Funding Math Doesn't Lie

Investor skepticism about AI model quality is worth unpacking because it gets to the heart of what makes the Poolside situation structurally different from a typical startup funding setback.

Poolside isn't just building software — it's competing in a market where OpenAI, Anthropic, Google DeepMind, and others have years of model training runs, proprietary datasets, and hardware relationships that took enormous capital to build. Convincing institutional investors to fund a $2 billion round requires a convincing answer to one question: why will your models be better, or at least differentiated enough to matter?

The fact that investors walked suggests they didn't hear a compelling answer — and that's harder to fix than a delayed chip cluster.

From an insider perspective, this pattern is familiar in AI infrastructure plays. Startups frequently conflate "access to compute" with "ability to compete on model quality." Leasing Nvidia H100s from Iris Energy and utilizing Fluidstack's infrastructure — both confirmed in Poolside's history — is table stakes, not a competitive moat. The gap between running workloads on rented GPUs and training frontier models at scale is vast, and investors with any sophistication in this space understand that distinction.

Who Might Step In — And Why It's Complicated

Poolside hasn't abandoned the Texas project. The company is reported to be in conversations with cloud providers, targeting commitments for up to 400MW of capacity. Google was named as one party in those discussions.

The catch: the FT also reports that a source familiar with Google's discussions said the company is no longer actively in talks with Poolside. That's a significant detail. Google has its own infrastructure priorities, its own AI ambitions, and limited incentive to anchor someone else's data center project unless the terms are exceptionally favorable or the strategic fit is clear.

The Permian Basin site itself retains real long-term value — 10GW of potential capacity, paired with 2.5GW of on-site renewable generation, is genuinely attractive land in a region where power availability is increasingly the binding constraint on data center development. Texas has become one of the most competitive markets for AI infrastructure precisely because ERCOT, for all its complexity, offers more development flexibility than constrained markets like Northern Virginia or Silicon Valley.

The land and the power thesis are sound. The question is whether Poolside has enough credibility left to be the developer that captures that value — or whether the site eventually trades to someone with deeper infrastructure execution experience.

What This Signals for AI Infrastructure More Broadly

Poolside's situation isn't isolated. It reflects a broader tension that's been building across the AI infrastructure space: the gap between announced capacity and operational delivery.

Over the past two years, the industry has seen an enormous volume of gigawatt-scale announcements — data centers, power agreements, GPU procurement deals. A meaningful portion of those announcements were aspirational, built on funding assumptions that required everything to go right. When anchor tenants demand execution proof before construction milestones, or when model quality doesn't hold up to investor scrutiny, those announced gigawatts don't materialize.

For developers, lenders, and landowners evaluating AI data center partnerships, Poolside's experience reinforces a lesson that's easy to forget during a bull market: the strength of a deal isn't the headline capacity number or the valuation — it's the operational credibility of the parties involved.

Startups with compelling AI narratives but thin infrastructure track records will increasingly find that hyperscalers and institutional capital want more than a pitch deck. They want demonstrated ability to deploy chips, manage power agreements, and hit construction timelines. That bar is rising.

Poolside still has time to rebuild. The company has real technology, a clear use case in government-grade code automation, and a site with genuine long-term upside. But the path forward requires finding a partner willing to bet on a startup that's already burned one marquee relationship — and doing that in a market where every credible cloud provider has more options than bandwidth to evaluate them. The next deal Poolside signs, if it signs one, will tell us everything about how much runway the company actually has left.

Explore more about InfraSale Marketplace here!

[INTERNAL LINK: AI infrastructure trends]

[INTERNAL LINK: funding challenges for startups]

[INTERNAL LINK: renewable energy in data centers]