Firms like OpenAI and Meta claim a compute shortage while also exploring selling compute capacity. This isn't a contradiction but a strategic evolution. They are buying all available supply to secure their own needs and then arbitraging the excess, effectively becoming smaller-scale cloud providers for AI.

By hoarding GPUs for its own models, Elon Musk's xAI inadvertently created one of the largest "neocloud" platforms. Massive deals with Google and Anthropic show that in the current crunch, simply possessing available compute is an incredibly lucrative and powerful position, almost independent of the models being built.

The intense demand and limited supply of compute and power are creating strange bedfellows in the AI industry. This dynamic forces companies with strong models but weak infrastructure (Anthropic) into partnerships with rivals who have excess compute capacity (Musk's SpaceX), fundamentally reshaping market alliances based on comparative advantage.

By renting its massive data center to competitor Anthropic, Elon Musk's SpaceX (parent of xAI) is tacitly admitting a strategy shift. Instead of competing directly on model development, it's becoming a high-margin compute provider, akin to a "new CoreWeave," and ceding the AI race.

Elon Musk's xAI initially appeared to lag, but the real victory was building massive supercomputers. By leasing this compute power to rivals like Google and Anthropic, SpaceX's 'Neo Cloud' became its primary revenue source, revealing a highly profitable infrastructure-first AI strategy.

The enormous scale of Meta's deal with specialized data center operator Nebius proves that "NeoClouds" are now critical infrastructure players. They are successfully competing with hyperscalers by offering specialized services and, crucially, available capacity, making them essential partners for AI giants.

Recognizing his Grok model lags competitors, Elon Musk has strategically shifted focus. By providing compute capacity from SpaceX's data centers to rival Anthropic, he is leveraging his strength in building massive physical infrastructure to become a critical "NeoCloud" provider, influencing the AI race through compute control rather than model performance.

Despite weak demand for its AI model Grok, SpaceXAI is expanding its data center footprint. Its key advantage is its record-breaking speed in building compute infrastructure (122 days for one facility), which it now rents to other AI companies, effectively becoming a cloud provider.

A new pattern is emerging: companies that over-invested in GPUs for proprietary AI models that didn't materialize are now leasing that excess capacity. Meta and SpaceX's entry into the cloud market creates new 'neo-cloud' competitors and signals a strategic failure in their original AI ambitions.