Amazon CEO: AI Business Hits $25B Run Rate, $220B CapEx Still Can’t Meet Demand
TREE NEWS reports: Amazon CEO Andy Jassy revealed that the company’s AI business has reached an annualized revenue run rate exceeding $25 billion in just three years—a pace hundreds of times faster than AWS’s early growth—during a wide-ranging interview with a16z co-founder Ben Horowitz. Jassy disclosed that despite plans to spend roughly $220 billion in capital expenditures in 2026, Amazon still cannot satisfy demand, with 2026 capacity fully sold out, 2027 nearly sold out, and a significant portion of 2028 already committed.
The Barbell Structure of AI Demand
Jassy described AI demand as a “barbell” with two extremes: AI labs consuming massive amounts of compute, and enterprises using AI for cost reduction and productivity gains. The middle of the barbell—enterprise production workloads—remains largely untapped, and Jassy believes inference costs must fall by an order of magnitude to unlock it. This is why Amazon is investing heavily in custom silicon, including Graviton processors (40% better price-performance than x86) and Trainium AI accelerators (30-40% better price-performance).
Market Implications
Tech Stocks: Jassy’s disclosure that demand exceeds even $220 billion in spending should reassure investors worried about hyperscaler capital expenditure becoming stranded costs. The AI revenue trajectory—$25 billion in three years versus AWS’s $60 million in its first three years—validates the aggressive buildout thesis. However, the free cash flow headwind from front-loaded data center investment may pressure near-term valuations for Amazon and peers like Microsoft and Google.
Semiconductors: Amazon’s custom chip strategy poses a long-term competitive threat to Nvidia and Intel. While Nvidia remains the dominant AI accelerator supplier, the success of Trainium in reducing inference costs by 30-40% could shift enterprise workloads toward custom silicon over time.
Bonds and Rates: The scale of AI infrastructure spending—$220 billion from Amazon alone—implies continued heavy corporate debt issuance. Data center construction requires two-year lead times with 30-year asset lives, suggesting sustained demand for long-duration credit.
Commodities: Jassy’s remarks on water consumption (golf courses use 200 times more water than data centers) and power grid investment signal that energy infrastructure and cooling technology will be key commodity themes.
Key Takeaways for Investors
- AI capex is demand-driven, not speculative: Sold-out capacity through 2027-2028 suggests hyperscaler spending is responding to genuine customer commitments rather than speculative buildout.
- Inference cost reduction is the next catalyst: The enterprise AI market won’t fully materialize until inference costs drop by an order of magnitude—watch custom silicon progress closely.
- Labor disruption ahead: Jassy’s admission that “there will be fewer humans doing the same work” signals AI-driven efficiency gains will reshape corporate cost structures.
- Free cash flow volatility: Investors should expect near-term FCF pressure as data center investments precede revenue realization, with payback periods of roughly three years.
