Amazon Chief Jassy Defends AI Spending

Amazon chief executive Andy Jassy has defended the billions the company and others are spending on artificial intelligence, saying it was important to spend “aggressively” now to reap rewards in the future.

Jassy’s statements on the matter in his annual letter to shareholders come after Amazon said in February it planned to spend $100 billion (£76.5bn) on capital expenditures this year, with the “vast majority” going toward Amazon Web Services cloud AI capabilities.

In the letter, Jassy said “substantial capital” is required to keep up with the pace of AI innovation and demand, calling the technology a “once-in-a-lifetime reinvention of everything we know”.

Massive spending

“The demand is unlike anything we’ve seen before, and our customers, shareholders, and business will be well-served by our investing aggressively now,” he wrote.

He said the greatest costs now are data centres and AI acceleration chips, but noted that costs for AI services are expected to decrease with time.

“The faster demand grows, the more data centres, chips and hardware we need to procure (and AI chips are much more expensive than CPU chips),” he wrote.

He said that while the spending was made “upfront”, the assets would remain useful “for many years”.

He said Amazon’s own Trainium2 chips offer 30 to 40 percent better price-performance than current GPU-powered computing instances generally available today, referring to an in-house designed chip released late last year.

Other advances in AI would continue to lower the costs for inference, or providing AI services, in contrast to training models, he said.

Better price-performance with hardware will help, he said, but “inference will also get meaningfully more efficient in the next couple of years with improvements in model distillation, prompt caching, computing infrastructure, and model architectures”.

AI ‘opportunity’

Google chief executive Sundar Pichai last week reiterated the company’s plans to spend $75bn this year on data centre capacity, saying the AI opportunity is “as big as it gets”.

Amazon and other large tech companies are exploring ways of providing the massive power needs of their AI data centres, including using existing nuclear capacity and building new nuclear plants.

Last week a US energy regulator rejected a request by Amazon energy partner Talen over a rejected deal that would have seen an AWS data centre buy up to 1 GW of power directly from Talen’s colocated Susquehanna nuclear facility in Pennsylvania.

In rejecting the deal last year, regulator FERC expressed concern that diverting large amounts of power away from the regional grid could destabilise energy supplies.