NVIDIA Stock and the Hundred-Fold Compute Whisper

While Wall Street fixated on a multi-billion-dollar headwind, the real story was in how the very nature of AI demand was quietly changing.

If you’re kicking yourself for missing NVIDIA (NVDA) stock’s 56.4% climb from mid-2025 to mid-2026, you’re in good company. The move felt like it came out of nowhere, especially with a series of negative headlines brewing. But the clues were there, hiding in the footnotes and earnings call slides. The real signal wasn’t in the top-line numbers; it was in the changing vocabulary.

The Conversation Was Shifting From Training To Thinking

For years, the story of AI was about training large models. It was a land grab for raw power. But by early 2025, management had started telling a different, more nuanced story. They were talking about “reasoning AI.”

More than a new buzzword, this represented a fundamental shift in the workload. On its February 2025 earnings call, the company noted that this new wave of “long-thinking reasoning AI can require 100x more compute per task” than the simple, one-shot answers we were used to. A few months later, in May, they elaborated: these new AI agents required “hundreds to thousands more tokens per task.” The demand was not only greater but also different in kind – substantially more intensive. They even pointed to Microsoft, which had already seen a “five-fold increase” in tokens processed in a single quarter. The physics of the AI data center were changing, and NVIDIA was building the engine for it.

What Were The New Chips Actually Doing?

The most concrete clue was buried in the commentary around the company’s new Blackwell architecture. Typically, a new, powerful chip generation is first acquired by customers for the most demanding task: training the next behemoth AI model. But management shared a key detail in that same February call.

They stated that many of the early Blackwell deployments were “earmarked for inference.” A first for a new architecture. Read that again. Instead of training, customers were immediately deploying the new silicon to meet surging real-world demand – driven by two forces compounding each other.

The first was scale: companies were shipping AI to millions of users, and simply running these models consumed enormous compute. The second was reasoning: queries are becoming far heavier, with NVIDIA noting long-thinking AI could require 100x more compute per task than a simple one-shot answer.

More users, each one burning far more compute. It was a quiet but significant signal that the market’s center of gravity was moving, and the capital was following.

This surge in inference demand meant data centers needed far more chips, far sooner than analysts had modeled. Revenue forecasts were quietly, then suddenly, too conservative.

Inference is bringing in more competition for GPUs from ASICs as well as CPUs. Here is An AI Stock That Wins No Matter Which Chip Wins

The Market Was Looking The Other Way

So why did so many people miss it? Because the obvious story was a negative one. In May 2025, the company was addressing new export controls on China, leading to a loss of approximately $8 billion in revenue for the upcoming quarter from a single product line. That’s a large number that commands attention.

Yet, the underlying business was so strong it was absorbing that multi-billion-dollar revenue impact and still growing. Meanwhile, the options market was getting sleepy. In the weeks before the run, implied volatility had eased all the way down to the 19th percentile of its one-year range. Traders were pricing in less drama, not more. The real story was building, but it was doing so behind a wall of noise and a haze of complacency.

The lesson here isn’t about finding a magic bullet. It’s about listening for when a company starts describing not just how much customers are buying, but why. When the nature of the work changes, the scale of the opportunity often does, too.

So How Do You Spot The Next NVIDIA?

It is harder than it sounds, and especially hard for an individual investor with thousands of stocks to keep track of. That is exactly the gap the Trefis High Quality (HQ) Portfolio is built to fill. It weighs the quality signals across thousands of names to identify the 30 strongest, sizes and re-balances them with discipline, and has a track record of outpacing the S&P 500, S&P Mid-cap, and Russell 2000.