Server farms look like battlegrounds now: big US tech names are ramping up AI infrastructure spending, The Kobeissi Letter reports they intend roughly $850 billion for leasing data-center capacity in the coming years. That figure is hard to ignore — a sign they're buying raw compute at scale, not just polishing software (hum).
In Q1 2026 combined infra outlays jumped by about $570 billion, roughly +204% YoY, says the same piece. Microsoft and Oracle are called out as top spenders; Oracle, in particular, reportedly leads in contract volume tied to supporting OpenAI’s neural nets — deals said to exceed $250 billion. Keep in mind these are contract dollars, not immediate product launches.
There’s a catch the authors point out: all that capex hasn’t translated into a skyrocket of user-facing breakthroughs. For many people, 2026 feels slower — model updates are coming more cautiously, and most visible changes are stability tweaks and extra features to existing models rather than radical new architectures or paradigms, esp. for ordinary users.