Analogic to Invest $51.8 Billion in AI Infrastructure Despite Speed Control Debate

Despite calls for slowing down AI development speed, competition to secure computing infrastructure is intensifying. Recent noise around AI safety debates, OpenAI's model cancellation, and Oracle data center approval issues has not dampened actual investment and supply chain data, which point to acceleration instead.
Analogic announced a $51.8 billion commitment over 10 years with six partners for AI infrastructure build-out, including Google ($11.11 billion), Amazon ($11 billion), Microsoft ($3.14 billion), and Broadcom equipment leases worth $16.12 billion. Approximately 80% of the agreement is non-cancellable regardless of actual usage, meaning costs cannot be easily reduced even if demand slows or AI service expansion is delayed.
Analogic has shifted from renting cloud services to building dedicated data centers and directly leasing chips. The company is also considering purchasing memory directly from Samsung Electronics and SK Hynix. Analogic's computing expenses rose from $400 million in 2023 to $2.5 billion in 2024, reaching $7.3 billion last year, while revenue grew to $4.6 billion—a 11x increase—but still lags behind costs, resulting in an operating loss exceeding $8 billion.
The company emphasized that "the constraint on future growth is not demand but a shortage of computing resources," justifying further investment. Non-cancellable contracts reflect the severity of supply shortages. Despite risks if revenue growth and funding face issues, Analogic and OpenAI have recently launched new models like Claude Opus & Sonnet 5.5, GPT-6 Astral & Luna, and Grok 4.7.
OpenAI's annualized revenue is estimated near $7 billion, up over 70% from July's $4 billion, driven by enterprise sales doubling since July. The company aims for a $140 billion valuation with potential additional funding of at least $3 billion, much of which will likely go toward computing investments.
Supply chain constraints are worsening. Semiconductor supply shortage index hit a record high of 103.9 in September (above 100 indicates shortage), with demand outpacing supply. Among seven key AI infrastructure components, only GPUs showed balanced supply and demand; DRAM lead times have doubled to 20 weeks compared to normal 8 weeks, HDD at 50 weeks, ABF substrate materials at 48-56 weeks, and others showing similar delays.
Memory suppliers are prioritizing long-term contracts (LTA) while selling excess demand at market prices. TrendForce forecasts HBM average selling price to rise 121% next year. Power supply bottlenecks are also emerging as a critical issue for data center completion, with increased demand for on-site power generation solutions like fuel cells, solar, and gas turbines.
Markets will watch Micron's Q3 earnings report due on August 30 (US Eastern Time), which could impact memory market sentiment. With year-end rebalancing by pension funds approaching, short-term volatility may increase as some $3.3 billion in stocks might be sold off from US pension funds alone.
Korean Source
This article is an English localization of a Korean-language crypto news report. Original headline: AI 속도조절은 없다…앤스로픽 "AI 인프라에 5180억弗 투자"