Compute Power Lock-in Race——OpenAI's "Money Burning Machine" vs Anthropic's "Strategic Alliance"

Whoever can lock in the next-generation computing power will be able to outpace in this IPO marathon?

In the AI era, compute power is oil, and data centers are oil fields. Both companies are investing heavily in this track——but the way they spend money and their backers determine who can tell a more convincing growth story after the IPO.

OpenAI's compute expenses are staggering. The company has signed a $300 billion computing commitment with Oracle, and established a $100 billion hardware alliance with NVIDIA, but behind these dazzling numbers is a compute lock-in method without strategic allies——purely exchanging cash for compute power. The high inference costs are expected to reach $14.1 billion by 2026, with a gross margin of only about 33%, far below most tech companies. For every dollar earned, they spend $1.4 to $1.7 in costs, and this unit economics model will face severe challenges in the public market.

Anthropic, on the other hand, has taken a completely different approach. It has reached an exclusive leasing agreement for the Colossus 1 supercomputing cluster with SpaceX——this supercluster, equipped with over 220,000 NVIDIA GPUs, is the core support for Anthropic's inference services. In return, Anthropic will collaborate with SpaceX to develop several gigawatts of "Orbital AI compute power." Simultaneously, Anthropic has signed a five-year agreement with Google to pay $200 billion for using Google Cloud services and chips, a deal that accounts for over 40% of Google's latest financial report's future revenue commitments.

A deeper strategic difference is: OpenAI's relationship with Microsoft is shifting from "honeymoon" to "cold war," with fierce battles over the definition of AGI, profit sharing, and equity structure. Meanwhile, Anthropic has deeply tied the interests of Google, Amazon, and SpaceX through equity—these giants are shareholders, customers, and compute power providers all at once. When compute power is the most scarce resource, the difference between "own compute network" and "purely buying compute" will determine the competitive moat of these two companies after the IPO.

#Polymarket每日熱點