Amazon is committing $220 billion this year to AI infrastructure, but getting servers, data centres and electricity to arrive in step is proving difficult. Business Insider reports that AWS is taking ownership of some Nvidia server racks earlier, while power constraints at some sites leave equipment waiting to be switched on.
AWS AI Watch analysis
What happened
At a September staff meeting, AWS infrastructure chief Prasad Kalyanaraman said the company’s AI business was “hugely free cash flow negative” and expected free cash flow to stay negative for “the next couple of years or so”, according to Business Insider. The report draws on a meeting recording, internal documents and people familiar with the work.
An internal email projected about $5 billion of Nvidia GB300 racks arriving in August, followed by roughly $3 billion each month. Under a new “Early Rack Receipt” process, Amazon takes ownership earlier in the supply chain. Some sites had room for racks but not enough power to run them, the report says. Amazon says staging finished hardware near deployment sites is standard practice and disputes that AWS lacks a forecasted path for getting equipment into data centres.
Why it matters
AI infrastructure spending only starts earning its keep once the capacity is usable. Delays in power, construction or equipment installation can tie up cash before revenue arrives. The timing problem is not just about servers: Kalyanaraman described labour shortages that led AWS to fund housing near some Texas data-centre projects.
Amazon points to strong demand: AWS revenue grew 37% in the second quarter and its backlog reached $496 billion. But a backlog is not the same as installed, powered capacity, and Amazon still expects demand to exceed available capacity in 2027.
Our read
This is the less glamorous side of the AI race, where a rack can be delivered on time and still be no use without electricity. The story is not that Amazon’s bet will fail; it is that the scale of the build makes coordination and cash timing part of the technology challenge. For cloud customers, the practical question is whether AWS can turn its investment into usable capacity when demand arrives.
What to watch
- Whether AWS reports progress on power availability and data-centre commissioning.
- Whether its expected capacity shortfall in 2027 changes.
- How Amazon’s spending and cash-flow outlook develop as the buildout continues.
Discussion spark: When cloud companies build AI capacity ahead of demand, should investors accept years of negative cash flow as the price of securing future supply, or expect clearer proof that the infrastructure can be powered and put to work?
Sources and evidence
- Amazon knows how to build big. The AI boom is testing just how big it can go. – Business Insider (19 October 2026, 09:00 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)