#Anthropic再签350亿美元算力协议 This $35 billion computing-power agreement involving Anthropic and NVIDIA is a landmark event signaling the intensifying AI computing arms race, reflecting structural changes and future trends in the AI computing market.

I. The “Arms Race” Logic Behind the Agreement
1. Computing power has become the core moat: The iteration and commercialization of large AI models (such as Anthropic’s Claude and OpenAI’s GPT) are driving exponential growth in demand for computing power. Leading AI companies are signing long-term, large-scale computing-power agreements to lock in future supply ahead of time, avoid the risks of shortages and soaring prices, and build a “computing sovereignty” moat.
2. Giants “staking their claims” and increasing capital investment: Leading AI companies are locking in computing power at a pace of “$10 billion per month,” while capital providers such as NVIDIA are deeply binding themselves to leading AI companies through “computing-power financialization” models (such as NVIDIA acting as a “sub-landlord” to backstop computing-power supply), forming a closed loop of “chips + capital + infrastructure” and further raising the threshold for competing for computing resources.
II. Factors Driving the Continued Intensification of the Computing Arms Race
1. AI commercialization is forcing demand for computing power: AI applications are scaling up rapidly, and AI companies are moving from “burning money for scale” to “making money to prove themselves” (for example, OpenAI’s advertising revenue has exceeded $1 billion). Delivering on commercialization requires models to iterate faster and achieve lower inference costs, thereby continuously stimulating investment in computing power.
2. Computing-power supply bottlenecks and a construction frenzy: AI computing power, especially high-end GPUs, still faces a shortage at the million-unit level, while data-center construction and power supply have long lead times. This is prompting AI companies to plan ahead, while also driving comprehensive expansion across upstream areas such as chips, liquid cooling, and optical communications, creating a positive cycle of “computing-power demand → infrastructure investment.”
3. Technology-route and ecosystem positioning: Leading companies are attempting to establish barriers around technology routes and ecosystem standards by developing their own chips (such as OpenAI) or binding themselves to specific computing ecosystems (such as Anthropic’s partnership with NVIDIA), further intensifying competition for computing resources.
III. Potential Risks and Cooling Factors
Although the computing arms race is generally heating up, it also faces certain constraints:
1. Antitrust and regulatory risks: The “computing-power financialization” model used by chip giants such as NVIDIA (for example, providing credit support to cloud providers in exchange for a share of revenue) has already raised antitrust concerns, and tighter regulation could restrict some disorderly expansion.
2. Financial and performance risks: AI startups that sign enormous long-term computing-power contracts face substantial cash-flow pressure. If commercialization progresses more slowly than expected, they may face debt risks.
3. Technology-route switching risks: As breakthroughs are made in proprietary chips (such as OpenAI’s self-developed chips), some companies may reduce their reliance on external general-purpose computing power, thereby changing the competitive landscape of the computing-power market.

Overall, the AI computing arms race will continue for the foreseeable future, but the focus of competition is shifting from simply “computing-power scale” toward “computing efficiency, green energy, and ecosystem collaboration.” $NVDA ‌