Amazon issues aggressive AI bill, with capital expenditures expected to reach $200 billion by 2026

Image from Visual China

Blue Whale News, February 6 (Reporter Wu Jingjing) — On February 5, local time, Amazon announced its fiscal year 2025 fourth quarter and full-year results. The earnings report showed that the company achieved revenue of $213.4 billion in the fourth quarter, a 12% increase year-over-year; operating profit was $25 billion. Despite steady performance in core businesses, after disclosing a massive capital expenditure guidance, Amazon’s stock price dropped over 11% in after-hours trading.

What truly caused a sharp shift in market sentiment was Amazon’s aggressive capital expenditure guidance for the future. During the just-concluded earnings call, the company clearly stated that capital spending in 2026 is expected to reach about $200 billion, exceeding Google’s previously disclosed cap of $185 billion and well above the average analyst estimate of $146.11 billion.

“We are committed to investing actively. We will invest to become the leader in this field,” CEO Andy Jassy said during the earnings call.

Amazon’s management spent considerable time explaining the logic behind this decision. Jassy stated that the additional capacity is almost immediately absorbed and monetized by customers after deployment. The company is converting capacity into revenue and cash flow at “the fastest installation speed.” He emphasized that AWS has accumulated long-term experience in data center construction, chip design, network equipment, and power optimization, and is not expanding blindly.

“When we see such strong and sustained demand signals, increasing investment is rational,” Jassy promoted AWS cloud division’s year-over-year growth of 24%, with the chip business continuously expanding: “In our core AWS business, we have proven that this investment model can generate substantial returns, and the same applies in AI.”

Regarding the demand structure in the AI market, Jassy described it as “barbell-shaped”: one end is AI-native labs and a few leading applications that consume large amounts of computing power; the other end consists of a vast number of traditional enterprise clients using AI to improve productivity and reduce costs.

He revealed that by the end of Q4, AWS’s backlog reached $244 billion, a 40% increase year-over-year and a 22% increase quarter-over-quarter, reflecting ongoing enterprise investment in AI and core cloud services. On the supply-demand side, Jassy admitted that current computing capacity remains tight. “If we had more capacity, growth could be even faster.”

Major overseas cloud service providers are competing to provide more AI infrastructure for Anthropic and OpenAI. In terms of AI infrastructure, self-developed chips have become a key focus for Amazon. During an analyst Q&A, Jassy disclosed that the “Rainier Project,” based on the Trainium 2 chip, is training Anthropic’s next-generation cloud model, with about 500,000 chips deployed so far, and plans to continue increasing this number.

He pointed out that Trainium offers 30% to 40% higher cost-performance ratio compared to similar GPUs, making it highly attractive to customers. Currently, Trainium’s annualized revenue run rate has reached several billion dollars, and capacity is “completely sold out.” Amazon has begun delivering the next-generation Trainium 3 chips, which further improve cost-performance by about 40% over Trainium 2, with nearly all related capacity expected to be booked by mid-year. Additionally, Amazon has initiated R&D on Trainium 4, scheduled for release in 2027, and has begun early discussions on Trainium 5.

Notably, before the earnings release, Amazon signaled organizational adjustments. On January 28, the company confirmed it would cut about 16,000 corporate administrative jobs, marking the second round of large-scale layoffs since October last year.

Beth Galetti, Amazon Senior Vice President, said this move is part of ongoing “streamlining of management levels and strengthening of responsibilities.” The layoffs mainly target white-collar positions in key departments such as AWS, retail, Prime Video, and HR (PXT). Affected US employees will have a 90-day internal transfer period; those unable to transfer will receive severance and benefits. From the structure of the layoffs, this adjustment is not merely cost-cutting but also aligns with high-intensity capital investment, aiming to flatten the organization and reshape efficiency.