Poolside AI is scrambling to rescue its $14 billion bet on a 2-gigawatt Texas data center after CoreWeave walked away from a 15-year anchor lease and a planned $2 billion funding round, including a rumored $1 billion Nvidia anchor check, fell apart in early April 2026. The collapse, first reported by the Financial Times and confirmed by DataCenterDynamics on April 2, 2026, has forced the AI coding startup co-founded by former GitHub CTO Jason Warner to court Google and other hyperscalers for a scaled-down 400-megawatt revival of Project Horizon – a plan that, according to Finimize reporting from the same week, has now also cooled.
The unraveling of the largest single AI infrastructure project ever attempted by a startup outside the OpenAI-Oracle Stargate orbit marks a sharp inflection point for the $650 billion AI capex cycle. It exposes how quickly investor enthusiasm can sour when training-economics fail to keep pace with frontier-lab leaders, and it raises hard questions about which AI-native companies can actually finance the gigawatt-scale compute they say they need. With CoreWeave’s 40,000 Nvidia GB300 NVL72 commitment now in limbo and the 568-acre Longfellow Ranch site sitting idle in Pecos County, Poolside has become the cautionary tale of the 2026 build-out – a startup that went full-stack on power, land, and silicon at exactly the moment its financing window slammed shut.
Poolside AI’s $14 Billion Project Horizon: From Marquee Deal to Fire Sale
Project Horizon was unveiled on October 15, 2025, with Poolside positioning itself as the first frontier-AI startup to vertically integrate compute, land, and power. The plan called for an eight-phase, 250-megawatt-per-phase build on the 568-acre Longfellow Ranch site in Fort Stockton, Texas – part of the Mitchell family’s 500,000-acre property in the heart of the Permian Basin. CoreWeave was unveiled as anchor tenant under a 15-year lease for the first 250 MW with an option on a second 250 MW, and was simultaneously contracted to deliver more than 40,000 Nvidia GB300 NVL72 systems beginning December 2025. Poolside’s blog announcement framed Horizon as “the world’s largest AI training campus” and described an architecture explicitly tuned for multi-agent reinforcement-learning systems rather than conventional inference workloads.
By April 2, 2026, that vision had collapsed. DataCenterDynamics reported that CoreWeave had exited the anchor-tenant role and that the parallel $2 billion Series C – which Poolside had been raising at a $14 billion pre-money valuation, with Nvidia’s NVentures arm slated to anchor up to $1 billion – had fallen apart. The Financial Times, which broke the funding-round news, cited investor concerns that Poolside could not credibly train models that compete with Anthropic, OpenAI, and Google DeepMind at the frontier. Two weeks later, Finimize reported that Google had held discussions with Poolside about restarting a 400-megawatt slice of Horizon – roughly 20% of the original plan – but those talks had also gone quiet by the third week of April 2026, according to people familiar with the negotiations.
The reversal has been swift and embarrassing. In October 2025, Poolside’s announcement was greeted as proof that AI startups could rival hyperscalers in build-out velocity. By late April 2026, the same project sits as the most public failure of the AI infrastructure boom – a $14 billion plan now reportedly seeking any commercial partner willing to underwrite even a 200 MW commitment. Bloomberg, citing sources close to the company, reports Poolside has begun cutting its capital commitments to long-lead-time gas turbines and has paused some site preparation contracts to preserve cash through Q3 2026.
How the CoreWeave Deal Collapsed: Inside the 40,000 GB300 GPU Unwind
The CoreWeave-Poolside deal, signed in October 2025, was structured as a back-to-back transaction: CoreWeave would lease the first 250 MW phase of Horizon for 15 years and, in parallel, supply Poolside with a cluster of 40,000-plus Nvidia GB300 NVL72 systems – a deployment with a notional sticker value north of $7 billion at list price. The structure gave Poolside immediate access to bleeding-edge Blackwell Ultra silicon while giving CoreWeave a long-duration anchor tenant for a campus it did not have to build itself.
What broke the deal, according to multiple people briefed on the talks, was a basic mismatch between Poolside’s training-revenue trajectory and the cash-flow profile required to support a 15-year master lease. CoreWeave, which became publicly traded in March 2025 and has been under pressure to shore up the revenue concentration of its top customers – Microsoft alone accounted for an estimated 62% of CoreWeave 2024 revenue per its S-1 – could not justify booking another single-tenant commitment of this size to a startup whose training pipeline had not yet produced a frontier-grade model. When Poolside’s parallel $2 billion Series C began stumbling in February and March 2026, CoreWeave pulled the plug on the lease before final closing.
The 40,000-unit GB300 NVL72 commitment is now expected to be reallocated across CoreWeave’s other tenants, with sources telling DataCenterDynamics that capacity originally bound for Pecos County will instead be deployed into CoreWeave’s Plano, Texas, and Las Vegas, Nevada, sites through 2026 and 2027. For Nvidia, the unwind is awkward but not financially material – the units were already produced and contracted – though it leaves a hole in CEO Jensen Huang’s narrative that AI-native startups can credibly compete with hyperscalers on infrastructure investment. Nvidia, which had committed up to $1 billion to anchor the failed funding round, has now declined to lead any rescue financing, according to a Bloomberg report on April 24, 2026.
Why the $2 Billion Series C Imploded: Investor Concerns Over Training Economics
Poolside’s failed Series C was structured as a $2 billion primary raise at a $14 billion pre-money valuation – an aggressive 4.7x markup from the $3 billion post-money valuation set in October 2024 when Bain Capital Ventures led a $500 million Series B. The round, which began marketing in January 2026, was anchored by an expected $1 billion check from Nvidia’s NVentures, with participation pencilled in from existing backers Lightspeed Venture Partners, DST Global, FirstMark, Felicis, eBay Ventures, GitHub, Lakestar, Soma Capital, and Volo Earlybird. By the time the round was abandoned in early April 2026, only about $250 million in soft commitments had been secured, well short of the $2 billion target.
The Financial Times, which broke the story on April 1, 2026, reported that institutional investors balked at three specific concerns. First, Poolside’s flagship Malibu coding model, released in beta in mid-2025, had failed to clear the SWE-bench Verified threshold of 70% – well behind Anthropic Claude Sonnet 4.6 at 79.6% and Claude Opus 4.6 at 80.8%. Second, the company’s enterprise revenue run rate, while undisclosed, was reported to be under $50 million as of Q1 2026, a number that could not service the financing costs of a $14 billion infrastructure plan. Third, Poolside’s compute spend, even before Horizon, was running at roughly $400 million annually against gross margins that had not turned consistently positive.
The combination of weak benchmark performance, modest revenue, and high capital intensity made the 4.7x markup untenable. “There’s a difference between an infrastructure thesis and an infrastructure-only thesis,” said Sarah Tavel, general partner at Benchmark Capital, in a CNBC appearance on April 7, 2026. “Investors will pay up for compute if the model layer is winning. Poolside hasn’t made the case that the model layer is winning yet, and at $14 billion you have to be winning.”
Project Horizon Funding Timeline at a Glance
| Date | Event | Amount / Scale | Status |
|---|---|---|---|
| Aug 2023 | Poolside seed round | $26M led by Redpoint Ventures | Closed |
| Oct 2024 | Series B led by Bain Capital Ventures | $500M at $3B post-money | Closed |
| Oct 15, 2025 | Project Horizon announcement (2 GW, 8 phases, 568-acre Longfellow Ranch) | $14B+ projected build cost | Announced |
| Oct 16, 2025 | CoreWeave anchor lease + 40,000 GB300 NVL72 supply deal | 15-year lease on 250 MW + 250 MW option | Signed |
| Jan 2026 | Series C marketing begins | $2B target at $14B pre-money | Launched |
| Feb-Mar 2026 | Investor due diligence flags Malibu benchmarks, $50M ARR, $400M annual compute | Soft commitments ~$250M | Stalled |
| Apr 1, 2026 | Financial Times reports Series C fall-through and Nvidia NVentures pull-back | $2B round abandoned | Failed |
| Apr 2, 2026 | DataCenterDynamics confirms CoreWeave anchor lease termination | 40,000 GB300 reallocated | Terminated |
| Apr 16-22, 2026 | Google explores 400 MW reboot | 20% of original plan | Talks cooled |
| Apr 27, 2026 | Poolside seeks new partners; pauses gas-turbine commitments | Cash runway extended to Q3 2026 | Active |
Jason Warner and Eiso Kant: The Founders Behind Poolside’s Bet
Poolside was founded in May 2023 by Jason Warner and Eiso Kant. Warner, who serves as CEO, spent four years as CTO of GitHub from 2019 through 2022 and personally incubated GitHub Copilot – the very product Poolside now competes against. Before GitHub, he held engineering leadership roles at Heroku, Canonical (the company behind Ubuntu Linux), and 41st Parameter, and served as a managing director at Redpoint Ventures from 2022 until founding Poolside. Warner holds a bachelor’s in computer science from Penn State and a master’s from Rensselaer Polytechnic Institute.
Eiso Kant, the co-CEO and chief technology officer, previously founded Source{d}, a code-data analytics startup that Poolside acquired the technology and team from in 2023. Kant has spent more than a decade applying machine learning to source code understanding, and his Source{d} work formed the technical seed corn for Poolside’s training corpus. The two co-founders structured Poolside on a remote-first model with engineering hubs in Paris and the United States, and the company’s headcount had grown to roughly 250 employees by the end of 2025, according to LinkedIn data.
Warner’s vision, articulated repeatedly across podcast appearances in 2024 and 2025, was to build “a vertically integrated frontier lab focused exclusively on software development” – a thesis that called for owning the model, the inference layer, and ultimately the compute infrastructure required to train future models without paying hyperscaler margins. Project Horizon was the physical manifestation of that thesis. Its collapse reopens the question of whether any AI-native company smaller than OpenAI or Anthropic can credibly own its full stack.
Permian Basin Power: Why Pecos County Was Supposed to Be the Killer App
Project Horizon’s 568-acre site was selected for one core reason: stranded natural gas. The Longfellow Ranch sits adjacent to a major Permian Basin gas-processing hub, where producers routinely flare excess associated gas because there is insufficient pipeline capacity to move the molecules to market. By installing on-site aero-derivative turbines fueled directly from existing Occidental Petroleum and Targa Resources gathering systems, Poolside aimed to bring up to 2.5 GW of bridge generation online within 18 months – a fraction of the five-to-seven-year timeline ERCOT, the Texas grid operator, currently quotes for new transmission interconnects of comparable size.
The economic case was compelling on paper. Permian flared gas trades at a discount of roughly 60% to Henry Hub benchmarks, putting Poolside’s blended fuel cost in the low-single-digit cents per kWh range – well below the 5-7 cents per kWh hyperscalers typically pay for grid power in ERCOT West, and roughly half what Microsoft and Google pay for green-tariff power in Virginia or Oregon. With renewables tariffs continuing to climb and Texas grid lead-times stretching past 2030 for new gigawatt-scale interconnects, Horizon was structured as a power-arbitrage play as much as a compute play.
The location’s principal handicap turned out to be hydrology, not energy. Pecos County has minimal surface water, and direct-evaporative cooling – the cheapest cooling architecture for a gigawatt-class data center – requires roughly 1 to 2 million gallons per megawatt-day. Even with closed-loop liquid-cooled GB300 NVL72 racks, Horizon’s water budget was projected at 350,000 acre-feet per year at full 2 GW operation, exceeding the documented sustainable yield of the Pecos County aquifer system by an order of magnitude. Investor due diligence in February 2026 flagged the water cost as a material unmodeled risk, contributing to the Series C cooling.
The AI Coding Market: Where Poolside Stands Against Cursor, Copilot, and Codeium
The collapse of Project Horizon arrives at a moment when the AI coding-tools market is consolidating around a small number of fast-scaling winners. Anysphere, the maker of Cursor, raised at a $60 billion valuation in February 2026 on a reported $2 billion annualized revenue run rate. GitHub Copilot, embedded in the Microsoft 365 stack, surpassed 25 million paying users in March 2026, according to Microsoft’s Q3 FY2026 earnings call. Anthropic’s Claude Code crossed $1 billion in annualized run rate by Q1 2026, with developer-tool ARR growing more than 300% year-over-year. Against this backdrop, Poolside’s reported sub-$50 million ARR and second-tier benchmark scores have made the company an outlier on revenue per dollar of capital raised.
Stripe’s 2025 developer survey, published in January 2026, found that 73% of professional developers globally now use at least one AI coding assistant daily, up from 31% in 2024. Of that user base, GitHub Copilot holds the largest share at 46% for Claude Code, followed by Cursor at 19%, and others at lower shares per Feb 2026 survey[2][1]. With distribution heavily favoring incumbents bundled into hyperscaler stacks, building a fourth or fifth-place coding model has become a structurally harder business than it was when Poolside raised its 2024 Series B.
AI Coding Tool Competitive Snapshot, April 2026
| Product | Owner | Latest Valuation / Status | Reported ARR | SWE-bench Verified | Active Users |
|---|---|---|---|---|---|
| GitHub Copilot | Microsoft | Inside Microsoft (no standalone valuation) | $1.5B+ (estimated) | 72.5% | 25M+ paid |
| Cursor (Anysphere) | Anysphere Inc. | $60B (Feb 2026 round) | $2B | 51.7% | 10M+ daily active |
| Claude Code | Anthropic | ~$350B (tender offer indicative) | $1B+ | 80.8% | 5M+ paid |
| Codeium / Windsurf | Codeium Inc. | $3B (mid-2025) | $200M+ | 54% | 2M+ paid |
| Replit | Replit Inc. | $3B+ (late 2025) | $150M+ | 49% | 30M+ free, 1M paid |
| Poolside Malibu | Poolside AI | $3B (Series B, Oct 2024); failed $14B Series C, Apr 2026 | under $50M (reported) | under 70% | not disclosed |
Stargate, Crusoe, and Anthropic: How Horizon Compares to AI’s Other Gigawatt Bets
Project Horizon was always one of several gigawatt-class buildouts announced in late 2024 and 2025, but it stood alone among AI-startup-led projects in attempting to own the underlying real estate, power, and grid infrastructure. The Stargate Project, announced in January 2025 by OpenAI, Oracle, SoftBank, and MGX, committed $500 billion over four years and broke ground on a 1.2 GW Phase 1 in Abilene, Texas – built and operated by Crusoe Energy on a 4,000-acre site, with Oracle as primary cloud customer and OpenAI as anchor tenant. Crusoe Energy raised $686 million in late 2025 at a reported $4.6 billion valuation specifically to fund this expansion.
Anthropic, meanwhile, has structured its compute differently. Rather than owning physical assets, Anthropic signed a $50 billion compute partnership with Amazon Web Services covering 5 GW of Trainium-based capacity through 2030, and a separate $40 billion Google Cloud partnership for TPU access announced in March 2026. The combined commitments represent more than 10 GW of compute, but with zero direct real-estate or power exposure on Anthropic’s balance sheet. xAI took yet another path: Elon Musk’s Memphis Colossus 2 buildout, financed in part by a recently announced $25 billion debt facility, is now reportedly approaching 1.4 GW under contract – but xAI is leasing the campus from a developer rather than owning the underlying gas turbines.
Against this peer set, Poolside’s vertical-integration thesis was the most ambitious – and the most fragile. Stargate has hyperscaler balance sheets and sovereign capital behind it. Anthropic has $13 billion in committed cloud-credits liquidity. xAI has Musk and a roster of debt investors. Poolside, by contrast, was attempting to underwrite a $14 billion physical project with the balance sheet of a $3 billion-valued startup whose primary product had not yet broken into the SWE-bench top three. The mismatch was always there; the Series C process simply made it visible.
Expert Reactions: Industry Voices on the Horizon Collapse
Industry response to the Horizon unwind has been swift and unsparing. “This is the first real stress test of AI-startup vertical integration, and it didn’t pass,” said Gil Luria, head of technology research at D.A. Davidson, in an April 14, 2026, note to clients. “When the model layer is undifferentiated, no amount of physical-asset ownership rescues the unit economics. Poolside tried to skip a step that no AI-native company has yet figured out how to skip.”
Charles Fitzgerald, a longtime cloud-infrastructure analyst and the author of the Platformonomics blog, was blunter on April 8, 2026: “Project Horizon was a $14 billion bet that you can win the AI coding market by owning gas turbines in the Permian. The market has spoken: capital flows to model quality first, infrastructure second.”
Even sympathetic observers see the collapse as an inflection point. “I love what Jason Warner is trying to do, and the engineering team at Poolside is one of the strongest in our industry,” wrote Dylan Patel, founder of SemiAnalysis, in his April 22, 2026, briefing. “But the harsh reality is that you cannot finance a 2 GW campus with a model that loses to Claude on 60% of customer evals. The bridge they were trying to build was always 18 months too long.”
Hyperscaler executives have been more circumspect, reflecting the awkward fact that several of them are now circling the carcass for cheap capacity. A Google Cloud executive told Bloomberg on background that “the underlying assets at Pecos County have real value – the question is the price.” Sources at AWS and Microsoft have echoed similar interest in the gas-turbine and water-rights assemblage at Longfellow Ranch, even as none has yet stepped forward as a confirmed buyer.
The Macro Backdrop: AI Capex Is Slowing for Everyone Except the Top Three
Horizon’s collapse fits a wider 2026 pattern. According to Wells Fargo equity research dated April 21, 2026, hyperscaler AI capex commitments have continued climbing – Microsoft, Amazon, Google, and Meta are now collectively guiding to $650 billion in 2026 AI infrastructure spend – but startup-led infrastructure deals have stalled sharply. The same Wells Fargo note tracked 17 announced startup-led data-center projects between Q3 2024 and Q1 2026 totaling 9.4 GW; of those, only 3 GW had broken ground by April 2026, with the remainder either delayed, scaled down, or canceled outright.
This bifurcation aligns with the half-of-U.S.-AI-data-centers-delayed reporting from earlier in April 2026, which documented 7 GW of canceled or delayed capacity across the broader $650 billion buildout. The pattern suggests the 2026 AI capex story is increasingly “haves and have-nots”: hyperscalers and a handful of frontier labs with existing customer revenue can finance physical infrastructure at any scale, while the long tail of model-layer startups is being pushed back to capital-light operating models.
For Poolside specifically, the fallout extends beyond Horizon. The company’s compute spend with CoreWeave, Microsoft Azure, and Google Cloud – estimated at $400 million annually as of Q1 2026 – now has to be renegotiated against weaker bargaining use. Multiple sources report that Poolside has begun internal modeling around a “lab-only” pivot in which the company would abandon infrastructure ownership entirely, lean on existing hyperscaler relationships for inference capacity, and concentrate remaining cash on improving the Malibu coding model’s benchmarks ahead of any future fundraise.
Five Predictions for the Post-Horizon AI Infrastructure Landscape
1. Poolside will close a smaller bridge round at a flat or down valuation by Q3 2026. With more than $400 million already raised and a $50 million-or-less ARR base, Poolside has a runway problem. Industry observers expect a $200-$300 million bridge from existing investors at the same $3 billion valuation set in October 2024 – not a markdown, but explicitly not the $14 billion Horizon-era valuation either. Bain Capital Ventures and Lightspeed remain motivated to protect their positions.
2. Google or AWS will acquire the Longfellow Ranch site at a 50-70% discount to Poolside’s stated cost. The land, water rights, gas-turbine commitments, and ERCOT interconnect queue position represent real, hard-to-replicate assets. Several hyperscaler executives have signaled interest, and the absence of competing bidders means the eventual buyer will likely secure 250-500 MW of pre-permitted capacity for a fraction of greenfield development cost. Expect a deal announcement before year-end 2026.
3. Nvidia will quietly shift its startup-investment strategy toward inference-only deployments. The aborted $1 billion NVentures check into Poolside represents Nvidia’s largest near-failure as a venture investor, and the company has already pulled back from new training-cluster anchor commitments according to its April 2026 disclosures around OpenAI and Anthropic. Expect Nvidia capital to flow toward inference-heavy, revenue-positive AI applications rather than frontier-training startups for the remainder of 2026.
4. Vertical integration becomes a hyperscaler-only thesis through 2027. No AI-native startup is now likely to attempt another startup-led 2 GW campus in the next 18 months. Instead, expect frontier labs to deepen multi-cloud commitments – Anthropic’s $50B AWS Trainium plus $40B Google TPU model is the new template – while leaving land, power, and grid risk on hyperscaler balance sheets.
5. AI coding consolidation accelerates: expect at least two named acquisitions before year-end 2026. With Cursor at $60 billion, Claude Code at $1B+ ARR, and Copilot at 25 million users, the long-tail coding startups (including Poolside, Codeium/Windsurf, and Magic) become acquisition candidates rather than independent IPO stories. The most likely strategic acquirers are AWS (which lacks a flagship coding product), Salesforce, and IBM. Poolside’s Malibu model and Paris-based research team would be a natural fit for any of the three.
What Poolside Customers and Employees Need to Know
For Poolside’s enterprise customers – a roster believed to include several Fortune 500 banks, two top-five U.S. defense contractors, and a handful of European industrials – the immediate question is service continuity. Sources close to the company say existing on-prem and dedicated-cloud deployments of the Malibu model remain fully supported, with no near-term disruption to inference SLAs. The 40,000 GB300 NVL72 GPUs originally reserved for Horizon have been redirected toward CoreWeave’s existing cluster footprint, leaving Poolside customers with continued access to Blackwell Ultra-class compute through their existing contracts.
For Poolside employees, the picture is murkier. Headcount stood at roughly 250 at the end of 2025, with about 60% in engineering and 25% in research. Two people familiar with internal planning say the company has paused new hiring and frozen non-engineering roles indefinitely, but there have been no broad layoffs as of late April 2026. The company’s runway, post the failed Series C, is reportedly sufficient to operate through Q4 2026 without further fundraising, though that timeline shrinks materially if the bridge round slips past Q3.
Warner himself has remained publicly silent through the unwind, breaking his Twitter/X cadence for the first three weeks of April 2026 before posting a brief acknowledgment on April 24, 2026: “Hard moments build hard companies. Heads down on the model. More soon.” Eiso Kant, who appeared on CNBC’s “The Exchange” the week of the original Horizon announcement, has not made a public appearance since the Series C collapse.
The Texas Power Politics Angle: ERCOT, the PUC, and the Permian Hangover
Project Horizon’s collapse also has political reverberations in Texas. The Mitchell family, longtime Permian landowners and the lessors of the Longfellow Ranch site, had publicly endorsed Horizon as a model for monetizing flared gas and creating high-paying technology jobs in West Texas. The Texas Public Utility Commission, which had quietly fast-tracked Poolside’s interconnect application under its 2025 large-load expedited-review process, now faces questions from competing data-center applicants who argue the queue jump no longer serves a viable project.
State Senator Charles Schwertner (R-Georgetown), who chairs the Texas Senate Business and Commerce Committee, told the Houston Chronicle on April 18, 2026, that “the legislature will want to understand what went wrong with Horizon and whether the regulatory acceleration we extended to AI infrastructure projects is being used responsibly.” That signal – coupled with concurrent ERCOT capacity-market tightness driven by the broader $1.4 trillion U.S. utility AI buildout – suggests Texas may tighten its grid-interconnect requirements for new AI data-center projects later in 2026, with knock-on effects for every startup planning to colocate in the Permian.
Frequently Asked Questions About Poolside AI and Project Horizon
What is Poolside AI and what does it actually build?
Poolside AI is a generative-AI startup focused on software development. Its flagship product is the Malibu coding model, which is offered to enterprise customers through dedicated cloud deployments and on-premises installations. The company was founded in May 2023 by former GitHub CTO Jason Warner and Source{d} founder Eiso Kant, and is headquartered remotely with engineering hubs in Paris and the United States.
How much money has Poolside raised and at what valuation?
Poolside has raised approximately $526 million in disclosed primary funding across two rounds: a $26 million seed round led by Redpoint Ventures in August 2023, and a $500 million Series B led by Bain Capital Ventures in October 2024 at a $3 billion post-money valuation. The company’s $2 billion Series C, which would have valued the company at $14 billion, fell apart in early April 2026.
What was Project Horizon and where was it located?
Project Horizon was Poolside’s plan for a 2-gigawatt AI training campus on a 568-acre parcel of the Mitchell family’s Longfellow Ranch in Pecos County, Texas, near Fort Stockton. Announced October 15, 2025, it was designed in eight 250-MW phases and was intended to be powered primarily by on-site aero-derivative gas turbines fueled by Permian Basin natural gas.
Why did the CoreWeave deal collapse?
CoreWeave terminated its 15-year, 250-MW anchor lease in late March 2026 after Poolside’s parallel $2 billion Series C funding round failed to close. Investor concerns about Poolside’s sub-$50 million ARR, sub-70% SWE-bench Verified scores, and $400 million annual compute spend made the long-duration lease commercially indefensible for CoreWeave’s public-market shareholders. The 40,000 GB300 NVL72 GPU supply contract was unwound at the same time and the units have been reallocated to other CoreWeave tenants.
Is Google going to take over Project Horizon?
Probably not at the original 2 GW scale. According to Finimize and Bloomberg reporting in April 2026, Google held discussions with Poolside about restarting roughly 400 MW of the campus – about 20% of the original plan – but those talks had cooled by late April 2026. Google, AWS, and Microsoft are all reportedly evaluating the Longfellow Ranch site at a steep discount to Poolside’s original cost basis, but no deal had been announced as of the publication of this article.
How does Poolside compare to Cursor, GitHub Copilot, and Claude Code?
Poolside trails the leading AI coding tools on every public metric. GitHub Copilot leads on distribution with more than 25 million paid users; Cursor (Anysphere) leads on revenue with a reported $2 billion ARR at a $60 billion valuation; Claude Code leads on benchmark performance with an 80.8% SWE-bench Verified score. Poolside’s Malibu model has not officially crossed 70% on the same benchmark and is reported to have less than $50 million in ARR.
Is Poolside going out of business?
Not in the near term. The company is reported to have sufficient cash runway to operate through Q4 2026 without additional fundraising, and has paused new hiring and large infrastructure commitments to extend that runway. Industry observers expect Poolside to close a $200-$300 million bridge round from existing investors before Q3 2026, with strategic-acquisition scenarios likely to surface if the company cannot demonstrate benchmark progress over the next two quarters.
What does this mean for AI infrastructure investment overall?
Project Horizon’s collapse is the clearest signal yet that the 2026 AI capex cycle is bifurcating. Hyperscalers (Microsoft, Amazon, Google, Meta) are accelerating their $650 billion combined 2026 spend, while startup-led infrastructure projects are stalling. Roughly 7 GW of announced AI data-center capacity has already been delayed or canceled across the broader U.S. buildout, and Horizon adds another 2 GW to the at-risk column.
Related Coverage
- Cursor’s $60 Billion Valuation: Inside Anysphere’s $2B Revenue Surge and the AI Coding Gold Rush
- Half of U.S. AI Data Centers Delayed or Canceled: Inside the 7 GW Capacity Crisis
- Big Tech’s $650 Billion AI Infrastructure Bet: Inside the Capex Surge Reshaping the Global Economy
- Amazon’s $5B Anthropic Bet: 5GW Trainium and $100B AWS Pact [2026]
- Google’s $40B Anthropic Bet: 5GW TPU Compute, $350B Valuation [2026]
- Claude Code vs GitHub Copilot 2026: 80.8% vs 72.5% SWE-bench and a $10 Price Gap
- Duke Energy’s $103B Plan: 14 GW Power Surge for AI Data Centers [2026]
- The $1.4 Trillion Grid Overhaul: How 51 Utilities Are Reshaping America’s Power System
External References
- Poolside: Why we’re building a 2 gigawatt AI campus in Texas (October 2025)
- TechCrunch coverage of AI infrastructure startups
- The Verge: AI and data-center reporting
- Jason Warner GitHub profile (Poolside CEO, former GitHub CTO)
- CoreWeave (former Project Horizon anchor tenant)
- Nvidia (NVentures was the planned $1B Series C anchor)
This analysis was published on April 28, 2026. All financial figures, deal timelines, and benchmark scores reflect publicly reported data as of that date. The fact pattern around the Series C collapse and CoreWeave anchor termination is based on Financial Times and DataCenterDynamics reporting from April 1-2, 2026, and follow-on Bloomberg and Finimize coverage through April 24, 2026.
