Anthropic has quietly committed to a staggering $517 billion in computing power contracts over the last eleven months, according to a report by The Information. Since October 2025, the AI lab has secured at least 14.8 gigawatts of new capacity, adding to its existing one to two gigawatts. This aggressive expansion includes plans for proprietary data centers, signaling a major shift from relying solely on third-party cloud providers. While impressive, this total still trails OpenAI’s ambitious 30-gigawatt target for 2030, though direct comparisons are muddied by the long-term nature of Anthropic’s agreements.
The financial scale of these deals dwarfs current income streams. Anthropic’s annualized revenue recently topped $65 billion, per Bloomberg, while OpenAI sat above $40 billion as of July. Neither company can fund these massive infrastructure bets through revenue alone right now. This disconnect between spending and earnings highlights the capital-intensive nature of the current AI arms race, where securing future compute is prioritized over immediate profitability. Investors are effectively betting that model capabilities will scale linearly with hardware access, a premise that remains unproven at this magnitude.
Ironically, this spending spree follows explicit warnings from Anthropic CEO Dario Amodei earlier in 2026. He cautioned rivals about “reckless risk” and argued that competitors “don’t really understand the risks they’re taking.” Now, his own company is racing to catch up, mirroring the very behavior he criticized. OpenAI CEO Sam Altman has similarly pushed back against what he calls “unsustainable silliness” from neo-cloud providers, suggesting that rapid technical progress could render today’s expensive hardware investments obsolete before they pay off.
This pivot suggests that fear of being left behind outweighs concerns about financial prudence or technical efficiency. If Altman is right and breakthroughs in algorithmic efficiency reduce the need for raw compute, Anthropic could be sitting on billions in stranded assets. Yet, in a market where capability is king, opting out of the hardware race is seen as an existential threat. The industry is no longer just building models; it is building the physical backbone of the next economy, regardless of the cost.
