Jensen Huang criticizes hyperscalers for slow infrastructure planning, says neoclouds will eat their lunch

NVIDIA CEO Jensen Huang said at the All-In Summit that annual infrastructure planning cycles at hyperscalers including Amazon, Microsoft, Google, and Meta are "almost always wrong" in the AI era, and predicted a wave of smaller AI-native "neocloud" providers will fill the gap. Huang said the market may eventually need 50, 100, or even 1,000 neocloud providers, positioning them as a complement to hyperscalers for regional inference, edge deployments, and specialized training runs. Hyperscaler capital expenditures are projected at roughly $800 billion in 2026 and approximately $1.3 trillion in 2027, about 63% growth in a single year.

Photo: Tima Miroshnichenko / Pexels Jensen Huang criticizes hyperscalers for slow infrastructure planning, says neoclouds will eat their lunch NVIDIA's CEO argues that annual planning cycles at major cloud providers are fundamentally broken in the age of AI, opening the door for hundreds of agile competitors Jensen Huang has a message for the biggest cloud companies on the planet: your planning process is broken, and a swarm of smaller competitors is about to prove it. Speaking at the All-In Summit, the NVIDIA https://cryptobriefing.com/markets/nvidia/ CEO took aim at the rigid annual infrastructure planning cycles used by hyperscalers like Amazon https://cryptobriefing.com/markets/amazon/ , Microsoft https://cryptobriefing.com/markets/microsoft/ , Google https://cryptobriefing.com/markets/alphabet/ , and Meta https://cryptobriefing.com/markets/meta/ . His core argument is straightforward. These companies plan their data center buildouts once a year, but the AI market moves so fast that those plans are, in his words, “almost always wrong.” The case against annual planning Huang’s critique hits at a structural problem that gets more painful as the AI boom intensifies. Hyperscalers are projected to spend roughly $800 billion in capital expenditures during 2026, with that figure ballooning to approximately $1.3 trillion in 2027. When you lock in infrastructure plans on an annual cadence, you’re essentially placing a massive bet on what demand will look like 12 to 18 months from now. The constraints aren’t just financial, either. Power availability, land acquisition, and construction timelines all create bottlenecks that compound the problem. Enter the neoclouds Huang’s proposed solution is a new category he calls “neoclouds,” essentially AI-native cloud providers that are smaller, regionally focused, and built for speed rather than scale. These companies don’t need to plan a year in advance because they aren’t managing globe-spanning infrastructure empires. AI, tech, and the markets they move—in one daily briefing. Daily. Free. Join 34,000+ readers across crypto, finance, and policy. Huang suggested the market might eventually need 50, 100, or even 1,000 of these neocloud providers to create a distributed network of AI capacity. Huang isn’t positioning neoclouds as replacements for hyperscalers. He’s arguing they’re a necessary complement, with smaller operators serving regional inference needs, edge deployments, and specialized training runs. What this means for the infrastructure arms race The AI-native cloud segment is already growing faster than the traditional hyperscaler market, according to Huang’s characterization. Multiple smaller installations are proliferating as companies realize they don’t need to build at hyperscale to serve meaningful AI workloads. The capital expenditure trajectory tells its own story. Going from $800 billion to $1.3 trillion in a single year represents roughly 63% growth in spending. That kind of acceleration puts enormous pressure on supply chains, construction crews, and power grids. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .