AINews reports a $12B deal where NVIDIA licenses Poolside's model factory and hires 109 employees, while Poolside's founders stay to pivot the company. Founder quotes attribute the pivot to a failed $2B raise that cost a 40,000-GB300 cluster and identify physical data center space and contracted compute, not capital, as the frontier bottleneck.
Poolside's founders describe the binding constraint moving from capital to physical infrastructure: a failed six-week $2B raise cost them a 40,000-GB300 cluster, and they now name physical data center space and contracted compute as the limits for next-year frontier models requiring more than an order-of-magnitude scale-up. This shift explains the deal shape—NVIDIA gains the factory and technical team, while Poolside's founders retain a cash-rich company to pursue experiment-bound, science-driven AI where compute is not the only moat.
The source is a single AINews newsletter that combines aggregator commentary with founder quotes; no primary agreement documents are cited. Headline deal amounts ($12B, $1B, $6B) and Infraco's 7GW neocloud target appear in the title/metadata rather than in the quoted block text, while the 109 hires, factory license, and founder statements are stated in the body. The quotes are internally consistent but self-serving and not independently verified, so confidence in specific financial figures should be moderate-low while the direction of the strategic pivot is clearer.
NVIDIA reports that its AVO agent architecture achieved 100.00 RHAE on the ARC-AGI-3 public set, solving all 183 levels across 25 environments using Claude Opus 5 and 6,624 environment actions, about 12% fewer than VISTA's reported 7,542. The same AVO system was previously applied to GPU-kernel optimization, and both sources frame the result as evidence that the agent harness—persistent memory, supervision, and an inspect-plan-implement-evaluate loop—rather than model capability alone drives long-horizon performance.
The binding constraint moves from model capability to the agent harness: the same model family reportedly scores about 30% when evaluated as a model and 100.00 RHAE when wrapped in AVO, with persistent memory and supervision presented as the mechanisms that let evidence and feedback compound over long horizons. This reframes long-horizon ability as a systems property, with model selection as one component rather than the ceiling.
The core factual claim—a 100.00 RHAE score on the ARC-AGI-3 public set, all 183 levels in 25 environments—is asserted by NVIDIA and repeated by an independent newsletter; both rely on NVIDIA's own reporting. Evidence is strongest for the existence of the public-set result, moderate for action efficiency relative to VISTA, and weak for the 'general-purpose' transfer claim because the comparison is non-controlled, the model-only baseline comes from a different evaluation setting, and no private-set score or third-party replication is provided.