I was listening when Jensen Huang returned to his new mantra and the investor line hung on the words: “Compute equals revenues.” You could feel the certainty in the room — and the rest of the market’s quiet skepticism at the same time. If you trade in bets and narratives, that tension is where money gets made and reputations get tested.
I’ll walk you through what Nvidia actually said, why the hyperscalers’ spending matters to Nvidia’s future, and the blind spots still poking holes in the narrative — in plain language you can use.
Data-center revenue swallowed the quarter whole.
Real world observation: Nvidia’s data-center business dominated the latest results. The company reported record full-year revenue of $193.7 billion (€162.0 billion), with the data-center segment delivering a staggering part of that haul — a quarter result of $62.3 billion (€52.1 billion) and data-center sales making up roughly 91% of total revenue for the period. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026?utm_source=openai))
I don’t sugarcoat this: Nvidia isn’t selling games and chips the old way anymore. You buy a Blackwell or Rubin GPU, and a cloud customer buys racks of them; those racks become the physical rails of modern AI. Jensen Huang framed it as a new economic axiom: tokens cost compute, compute drives revenue. He repeated the phrase so often it stopped being slogan and started to read like a business model.
What did Nvidia actually report on the earnings call?
Short answer: blowout numbers, tempered by investor caution. The quarter beat expectations on revenue and margins, but the market yawned once executives had to square aggressive hyperscaler capex with uncertain long-term enterprise adoption. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026?utm_source=openai))
Hyperscalers are spending like there’s no exit sign.
Real world observation: Amazon, Alphabet, Meta and Microsoft announced capex plans that are off the charts for 2026. Collectively, hyperscalers guided capital expenditures near the $650–$700 billion range for the year — roughly $700 billion (€585.5 billion) in industry commitments — numbers that have the market asking whether this is build or bubble. ([fortune.com](https://fortune.com/2026/02/17/ai-tech-red-flag-capex-hyperscalers-cash-flow-negative-evercore/?utm_source=openai))
Gartner backs the scale of that spending with a forecast that global AI spending will hit about $2.52 trillion in 2026 (≈ €2.11 trillion). That’s not just software licensing; a huge chunk is infrastructure — servers, networking, the whole dirty business of real-world scale. ([gartner.com](https://www.gartner.com/en/newsroom/press-releases/2026-1-15-gartner-says-worldwide-ai-spending-will-total-2-point-5-trillion-dollars-in-2026?utm_source=openai))
Here’s the thing I want you to feel: you’re watching the industry pour concrete for a highway system before most drivers have cars that can use it. It’s like an all-you-can-eat buffet for GPUs — appetites are enormous, but who exactly fills the plates tomorrow is still an open question.
Why are hyperscalers spending nearly $700 billion on AI capex?
They’re buying capacity because models are getting hungrier. Larger models, denser inference workloads, and the push for 24/7 latency guarantees mean more racks, more datacenters, and faster chips. Nvidia benefits because it sits at the supply choke point for that compute. ([gartner.com](https://www.gartner.com/en/newsroom/press-releases/2026-1-15-gartner-says-worldwide-ai-spending-will-total-2-point-5-trillion-dollars-in-2026?utm_source=openai))
Analysts are waving red flags for cash flow.
Real world observation: Evercore flagged that the scale of capex could push free cash flow into the danger zone for some hyperscalers. Analysts warned that heavy AI build-outs may temporarily make cash flow look negative for the largest players. ([fortune.com](https://fortune.com/2026/02/17/ai-tech-red-flag-capex-hyperscalers-cash-flow-negative-evercore/?utm_source=openai))
That’s why you saw short-term pushback in Nvidia’s stock after the call: profit and revenue can be robust, but if the buyers of your product are burning cash to build capacity, that changes the risk profile. Goldman Sachs and others have echoed the view that revenue upside needs to arrive soon to justify multibillion-dollar data-center builds.
Will hyperscalers’ capex push free cash flow negative?
It’s possible in the near term for some names; Evercore lays out that scenario plainly. But whether that’s structural depends on whether enterprises outside Big Tech finally extract measurable productivity from AI at scale. If that happens, revenue follows compute; if it doesn’t, you’ve just built an expensive piece of real estate with underused machines. ([fortune.com](https://fortune.com/2026/02/17/ai-tech-red-flag-capex-hyperscalers-cash-flow-negative-evercore/?utm_source=openai))
Enterprise adoption is the missing link.
Real world observation: Surveys and vendor execs are saying the same thing — companies use AI, but it hasn’t reshaped core operational processes for most firms. A new NBER firm-level survey found about 70% of firms report employing AI, yet over 80% say it made no measurable difference to productivity or employment in the recent window. ([nber.org](https://www.nber.org/papers/w34836?utm_source=openai))
OpenAI’s own COO Brad Lightcap recently admitted that “we have not yet really seen enterprise AI penetrate enterprise business processes,” which is unusually candid coming from one of the model vendors. That’s a reality check: models are powerful, and platforms like OpenAI Frontier or Anthropic’s Claude Cowork may lower the barrier, but adoption at the workflow level — the messy reconnection of data, security, IT, and incentives — remains the bottleneck. ([techcrunch.com](https://techcrunch.com/2026/02/24/openai-coo-says-we-have-not-yet-really-seen-ai-penetrate-enterprise-business-processes/?utm_source=openai))
Huang’s defense is a technical one: tokens are the new unit economics. More tokens consumed equals more compute used, which equals more revenue. If every line of software becomes tokenized, then compute demand — and Nvidia’s revenue — scales. That’s his wager. It’s a high-stakes poker table play: buy the rails first, hope the riders show up.
OpenAI and China remain strategic blind spots.
Real world observation: Nvidia disclosed the OpenAI investment hasn’t closed and China shipments are only slowly resuming. Huang said the $100 billion investment reported last year still hasn’t been finalized; filings make no promise a deal will close. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026?utm_source=openai))
On China, regulators have allowed limited shipments of certain H200-class chips, but Nvidia executives are not modeling meaningful revenue from China back into near-term guidance. That uncertainty hangs over the company precisely because China was once a dominant market share area for Nvidia’s datacenter chips. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026?utm_source=openai))
So what should you watch next?
Real world observation: Watch enterprise rollouts, not just capex headlines. Quarter-to-quarter revenue from cloud customers is necessary but not sufficient; look for signals that non-tech sectors — finance, healthcare, manufacturing, logistics — show measurable revenue or productivity uplift from agentic AI.
Operational signs to monitor: Fortune 100 CFOs guiding incremental revenue tied to AI, consulting and SI engagements moving from pilots into contract-backed projects, OpenAI and Anthropic signing enterprise deals with production SLAs, and hyperscalers reporting stabilizing free cash flow. These are the things that will either make Huang’s catchphrase feel prophetic or make it read like optimistic marketing.
If you want the short checklist: 1) enterprise case studies with dollar outcomes, 2) sustained hyperscaler utilization rates, 3) clarity on China sales, and 4) any final move on the OpenAI investment — those four items will tilt the risk-return calculus materially.
Whether you’re watching as an investor, operator, or curious thinker, one fact is unavoidable: Nvidia sold the market a new economic story — and now it has to prove the math. Will compute truly equal revenues, or have we built a highway that most cars won’t use?
Sources: Nvidia earnings release; Gartner AI spending forecast; Evercore coverage via Fortune; TechCrunch reporting on OpenAI; NBER working paper on firm-level AI adoption; midforex currency rates for USD→EUR conversion. ([nvidianews.nvidia.com](https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-fourth-quarter-and-fiscal-2026?utm_source=openai))