Nvidia CEO Jensen Huang says hyperscalers' annual infrastructure planning cycles are too slow for the pace of AI demand, and he wants smaller "neocloud" providers to fill the gap. He made the case at the All-In Summit as Nvidia itself projects roughly 70% revenue growth for its coming fiscal year.
Jensen Huang says the way the world's biggest cloud companies plan their data centers is broken. Speaking at the All-In Summit, the Nvidia CEO argued that the rigid, once-a-year infrastructure planning cycles used by hyperscalers such as Amazon, Microsoft, Google and Meta no longer match how fast AI demand moves.
Annual planning can't keep pace
According to Crypto Briefing, Huang said those year-ahead plans are "almost always wrong." Locking in a buildout on an annual cadence means betting on demand 12 to 18 months out, and the bet gets harder every year: hyperscalers are projected to spend roughly $800 billion in capital expenditures during 2026, rising to approximately $1.3 trillion in 2027. That jump works out to roughly 63% growth in a single year. Power availability, land acquisition and construction timelines compound the problem further.
Huang's answer: "neoclouds"
Huang's proposed fix is a category he calls neoclouds — smaller, regionally focused AI cloud providers built for speed rather than scale, so they don't need to plan a year ahead. He suggested the market could eventually need 50, 100, or even 1,000 of these providers to form a distributed network of AI capacity. He frames neoclouds as a complement to hyperscalers rather than a replacement, serving regional inference needs, edge deployments and specialized training runs.
Separately, Nvidia itself is projected to see roughly 70% revenue growth in its upcoming fiscal year, according to Crypto Briefing.
Whoever builds the infrastructure, Huang's argument comes down to speed: the AI market won't wait for annual planning cycles to catch up.
Sources: Crypto Briefing, Crypto Briefing
Trading involves risk.
Risk Disclaimer
