Amazon just became one of only a handful of companies ever to cross a $3 trillion valuation.
But the number that stopped me wasn't the market cap.
It was AWS growing 37% year over year to $42.2 billion in a single quarter — its fastest pace in more than four years — and it still isn't enough.
Andy Jassy's own words: even after lifting 2026 capital spending to roughly $220 billion, AWS "will still not have enough capacity to meet all the demand" this year, or next. He called 2028 demand "striking."
Sit with that. The largest cloud provider on earth, spending close to a fifth of a trillion dollars, says it cannot build fast enough.
For years we framed the AI race as a contest of models and talent. This quarter reframes it as something more physical: chips, power, and data-center capacity. The bottleneck has moved from software to infrastructure. That shift matters for everyone building in the cloud. When capacity is scarce, efficiency stops being optional. Cost per token, workload placement, and architectural discipline become real competitive advantages.
The winners of the next phase may not be those with the cleverest models, but those who can actually secure compute — and use it well.
So here's my question: is the real constraint on AI now energy and infrastructure rather than algorithms? And if so, who's best positioned to solve it?
Suggested tags: aws, ai, cloud, devops