SpaceX-Anthropic contract could reprice the AI market

The AI market repricing

Chamath Palihapitiya argues that a single contract between SpaceX and Anthropic could fundamentally reprice the entire AI buildout. He suggests that if SpaceX chooses to terminate their current compute agreement, Anthropic would lose a disproportionate share of the power it needs to function. In that scenario, SpaceX could reprice the capacity at spot rates, which Palihapitiya believes could be double or triple the current cost.

This move would force every other major hyperscaler to reconsider how they supply their own models. Neocloud providers, which rent out capacity to others, would be forced to buy whatever power is available at any price to avoid bankruptcy. Palihapitiya emphasizes that this is a matter of spreadsheets and math rather than personal relationships between the companies involved.

The broader market has already begun to react to these pressures. Following recent political shifts, the market saw a single session where roughly $4 trillion in AI-linked market value was erased. This volatility reflects a growing concern over who will capture the gains from AI and who will pay for the electricity and water required to run it.

The data center bottleneck

A significant hurdle for the AI industry is the growing public opposition to data center construction. According to an October CNBC All-America Economic Survey, 63% of registered voters oppose building a data center within 10 miles of their homes. This sentiment is driving a wave of local ballot measures to restrict new construction or add strict environmental requirements.

The situation is particularly tense in states like Texas, which accounts for more than 25% of planned US capacity through 2030. While some officials have ordered halts to environmental permitting for these sites, others are campaigning on moratoriums. This political friction creates a risk of 'stranded assets,' where projects are under construction but cannot secure the final permits needed to actually open for business.

Investors are currently underweighting the risk that many projects will never reach completion. Sacks warns that even if a site is currently being built, it may still require specific air pollution permits before it can operate. If the new-project pipeline is closed or sharply reduced by 2028, it could lead to a significant drop in capital expenditure for the entire sector.

Hyperscalers as kingmakers

The current infrastructure bottleneck favors the largest tech companies, often called hyperscalers. Companies like Amazon, Meta, Microsoft, and Google already own the existing gigawatts of power and the interconnection queue positions that new entrants cannot buy. Because they already have the physical footprint, they are positioned to maintain their dominance while others struggle to find space.

These giants have multiple paths forward to maintain their lead. They can continue renting capacity as they do today, or they can divert incremental compute to their own internal models. By doing so, they can throttle their rivals in a recursive improvement loop while competing models attempt to catch up to their capabilities.

The scarcity of power and permits creates a moat for these established players. As neoclouds are squeezed, they may be forced to buy capacity from the hyperscalers at any price. This dynamic could turn the existing infrastructure owners into the ultimate kingmakers of the AI era, dictating the pace of development for everyone else.

Legislative and geographic shifts

The next Congress is expected to push for major AI legislation that could create a federal department of AI. This body would likely review and approve frontier models, with 'harms' defined to include things like disinformation. Such regulation could provide the duopoly of closed model companies with more regulatory capture or business moats.

There is also a possibility that AI development could move geographically if US restrictions become too burdensome. Calacanis suggests that if states block buildout, capacity will simply move to countries like Canada, Mexico, Malaysia, or Singapore. This would result in the US becoming a net importer of tokens while exporting the construction jobs and the tax base to other nations.

Ultimately, the binding constraint on the AI industry remains power, which politicians cannot easily conjure. While the legislative landscape will change, the physical reality of energy requirements will dictate which companies can actually scale. The outcome of the SpaceX-Anthropic 90-day termination clause will be a key indicator of how these power dynamics will play out in the real world.