Amazon CEO Andy Jassy advocates for corporations to spend “aggressively” in AI at now to maximize future financial benefits.

In his annual letter to Amazon shareholders released on Thursday, Jassy said that “substantial capital” is necessary to maintain the momentum of AI research and meet consumer demand for AI goods.

He said that Amazon must allocate these funds now if it anticipates substantial returns on its investment in the future.

Jassy’s remarks follow Amazon’s announcement in February during its fourth-quarter results conference about intentions to allocate over $100 billion for capital expenditures in 2025.

Jassy said that the “vast majority” of the amount will be allocated to AWS AI capabilities.

“We maintain the conviction that AI represents a singular reinvention of all existing paradigms,” Jassy said in his shareholder letter.

The demand is unprecedented, and our customers, shareholders, and company will benefit from our aggressive investment at this time.

Jassy said that the primary expenditures associated with AI are now data centers and chips; however, he noted that, with time, these infrastructural costs would diminish.

Jassy said, “As demand in AWS escalates, the necessity for additional data centers, chips, and hardware increases (and AI chips are significantly more costly than CPU chips).”

“We allocate this capital initially, despite the long-term utility of these assets.”
Jassy cited Amazon’s own Trainium2 processors as evidence that the costs of AI infrastructure would decrease over time.

He said that these chips provide a 30%-40% superior price-performance ratio compared to the present GPU-powered computing instances typically available today.

Trainium2 was launched in late 2024.

Jassy said that future AI pricing dynamics would evolve as training costs decrease, redirecting funds towards inference, or the deployment of AI models.

“We perceive a significant urgency to reduce the cost of inference for customers,” Jassy said.

“More cost-effective chips will be beneficial. However, inference will significantly enhance in efficiency over the next several years because to advancements in model distillation, rapid caching, computational infrastructure, and model designs.

According to Jassy’s shareholder letter, Amazon is now developing over 1,000 generative AI applications.

He said that Amazon’s AI income is increasing at “triple-digit” percentages annually and is a “multi-billion-dollar annual revenue run rate.”