NVIDIA CEO Jensen Huang is sounding the alarm—and the opportunity bell—on just how fast AI computing demand is accelerating. Speaking about what’s coming next for the industry, Huang laid out a forecast that makes it clear NVIDIA expects the AI boom to intensify, not cool off, through the rest of the decade’s current window.
The biggest change, according to Huang, is that AI is moving from an era dominated by training into one increasingly driven by inference. In simple terms, instead of focusing primarily on building and teaching new models, the market is rapidly shifting toward running those models at enormous scale for real-world use. And that shift is creating what he described as “wild” demand for compute.
Huang points to a stunning metric to show just how extreme the ramp has become: he claims compute requirements have surged by up to 1,000,000 times in only two years. Whether you’re talking about AI assistants, enterprise automation, coding tools, content generation, or large-scale search and analytics, inference workloads can multiply quickly once a model becomes widely deployed. That’s why the conversation is no longer just about who can train the best model—it’s about who can serve the most tokens, with the lowest latency, at the best cost.
NVIDIA believes it is positioned to capture the center of this demand. Huang has already highlighted the company’s momentum with newer platforms like Blackwell and its roadmap beyond that, and now he’s pushing expectations even further. He reportedly projects NVIDIA could generate more than $1 trillion in revenue across 2025 to 2027, framing that outlook as conservative because of how central NVIDIA has become to modern AI infrastructure.
One notable sign of how tight the market is: Huang also discussed rising spot pricing for older GPU generations, including Ampere and Hopper. When even several-year-old AI accelerators are becoming more expensive on the secondary market, it suggests a deeper compute bottleneck—organizations are grabbing whatever performance they can get, not just waiting for the latest hardware. For NVIDIA, that kind of supply-and-demand imbalance typically translates into sustained pricing power and continued urgency from buyers.
A large share of this demand is being driven by hyperscalers and cloud-native adoption, where massive data centers need fleets of GPUs to keep up with customer workloads. At the same time, Huang emphasized that sovereign AI initiatives are growing quickly as well. Governments and regions are investing in their own domestic AI capacity, with activity increasing across places like the Middle East and the European Union. These projects can be especially significant because they often involve large, strategic infrastructure commitments designed to last for years.
Huang also credits NVIDIA’s close ties with leading AI labs and major model builders—naming partners such as Anthropic and OpenAI—as major drivers of ongoing infrastructure expansion. As these organizations scale real-world usage, inference becomes a constant, always-on cost center, pushing demand for more GPUs, more servers, more networking, and better efficiency across the entire stack.
Efficiency is a key part of NVIDIA’s pitch. Huang has highlighted improvements in performance per dollar—often discussed in terms like token economics, or how many AI outputs you can generate for a given cost. If a new platform can deliver meaningfully better token-per-dollar results, it becomes hard for AI companies and cloud providers to ignore, especially when they’re competing on speed, price, and reliability.
Some analysts and industry watchers were skeptical of NVIDIA’s earlier revenue expectations when the company presented them, but Huang’s latest comments effectively raise the bar again. The message from NVIDIA is clear: the inference wave is just getting started, compute demand is expanding at a breathtaking rate, and the company expects AI infrastructure spending to remain aggressive for the next several years.
If you want, I can rewrite this again in a more dramatic, consumer-friendly tone or a more business/finance tone while keeping the same facts and intent.
