Nvidia’s $96.2B Quarter Says the AI Boom Is Still Underbuilt
Nvidia made $59.7 billion in profit in one quarter. If you still think AI is mostly slide decks, lousy chatbots and Silicon Valley cosplay, the market just handed you a very expensive correction.
Nvidia made $59.7 billion in profit in one quarter. If you still think AI is mostly slide decks, lousy chatbots and Silicon Valley cosplay, the market just handed you a very expensive correction.
On August 26, Nvidia reported $96.2 billion in revenue for the quarter ended July 26, 2026 — up 106% on the same period a year earlier. Its data-centre business alone did $89.0 billion. That is not a promising technology trend. That is industrial-scale demand arriving with a baseball bat. ([investor.nvidia.com](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx))
The number that actually matters
Wall Street expected another strong Nvidia quarter. It got something more valuable: proof that the AI spending boom has not hit the brakes.
Revenue rose 18% from the prior quarter. GAAP net income came in at $59.688 billion, up 126% year on year. Gross margin was 75%. Read that again. Three quarters of every dollar Nvidia brought in after direct costs remained. Plenty of businesses can grow quickly. Very few can grow at this scale while producing economics that would make a casino owner blush. ([investor.nvidia.com](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx))
Then management did the thing markets actually cared about. Nvidia forecast $108 billion, plus or minus 2%, for its current quarter — ahead of analyst expectations. More importantly, CFO Colette Kress said the company expects revenue to grow about 70% in the fiscal year ending January 2028. Reuters reported that analysts had been expecting roughly 44% growth. Nvidia almost never gives this sort of year-ahead guide, so this was not an accidental bit of optimism tossed into an earnings call. ([marketscreener.com](https://www.marketscreener.com/news/nvidia-forecasts-quarterly-revenue-above-estimates-ce7858d9d18bf621))
That is why this is the tech story that matters on Friday, August 28. Nvidia is not merely selling chips. It is giving the market a forward order book on the biggest corporate capital-spending race in decades.
This is no longer just Google, Microsoft and Meta
The lazy version of the AI story is that a handful of American giants are buying each other’s cloud services, feeding a circular boom and hoping nobody asks when the revenue turns up.
There is a grain of truth in that. Nvidia itself has faced criticism over investments, financing arrangements and partnerships involving the companies that buy its hardware. Jensen Huang had to defend the company’s $50 billion investment in frontier AI labs, while sceptics have understandably asked whether some AI money is simply doing laps around the same small ecosystem. ([axios.com](https://www.axios.com/2026/08/27/nvidia-ai-earnings-stock))
But dismissing the whole thing as circular financing misses the bigger shift.
Nvidia says demand is broadening beyond hyperscalers. Reuters reported that AI labs could account for roughly a quarter of Nvidia’s business next year, while “neo-cloud” providers including CoreWeave and Nebius are expected to exit 2026 with more than eight gigawatts of Nvidia GPU capacity, up from three gigawatts at the end of 2025. Amazon Web Services also committed to deploy an additional 2 million Nvidia GPUs across its global infrastructure in 2027 and 2028. ([marketscreener.com](https://www.marketscreener.com/news/nvidia-forecasts-quarterly-revenue-above-estimates-ce7858d9d18bf621))
Then there is sovereign AI: governments deciding that compute capacity is too strategically important to outsource entirely to America’s cloud giants. Enterprise demand matters too. Every bank, miner, retailer, logistics company, drug company and insurer with a boardroom is now being asked some version of the same question: What are we doing with AI, and why are we not moving faster?
That does not mean all their projects will work. Most will be ordinary. Some will be bloody useless. But the infrastructure buying is real before the productivity gains are evenly distributed. That is usually how major platform changes work.
The railways got built before many routes earned their keep. Fibre was laid before plenty of websites deserved broadband. The internet had a bubble, then became more economically important than the optimists imagined. Smart operators can hold both ideas in their head at once: there can be stupid prices in parts of a boom, and a very real boom underneath it.
The overlooked bottleneck is not intelligence. It is supply.
The most interesting part of Nvidia’s result was not the 70% growth forecast. It was the reason management says it cannot grow faster.
Supply.
Nvidia said its outlook is constrained by the availability of components, particularly memory. Kress said memory costs would pressure margins and Reuters reported management expects gross margins to bottom around 71% to 72% in the fourth quarter. Yet Nvidia’s response was not to tone down demand expectations. It was effectively: demand is much bigger than what we can physically ship. ([marketscreener.com](https://www.marketscreener.com/news/nvidia-forecasts-quarterly-revenue-above-estimates-ce7858d9d18bf621))
This matters because it changes where the value sits.
For the past few years, the obvious trade was the model makers: OpenAI, Anthropic, Google and everyone else building clever systems. The more durable opportunity may be the boring, expensive plumbing around them: advanced memory, networking, power generation, cooling, data-centre construction, grid connections, fibre and the software that runs inference efficiently.
Nvidia is increasingly selling that whole factory, not just the gold-rush shovel. Its Vera Rubin platform is now ramping into production, and the company has announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR aimed at mobilising more than $500 billion of third-party capital for AI infrastructure, subject to definitive agreements. ([investor.nvidia.com](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx))
That should make founders pay attention. The opportunity is not always to invent the next foundation model. More often, it is to solve a nasty operational problem created when a large organisation tries to deploy one.
The contrarian take: Nvidia’s success is a warning to AI startups
Everyone sees Nvidia’s numbers and concludes, “I need an AI company.” That is how people end up building yet another thin wrapper around somebody else’s model, with no distribution, no proprietary workflow, no customer pain worth paying for and a pitch deck full of the word “agentic.”
Don’t do that.
Nvidia’s result says the scarce asset is not the ability to call an AI model. That is becoming abundant. The scarce assets are power, compute, trusted data, distribution, customer relationships and the competence to redesign work.
The best businesses will not sell “AI.” Customers do not wake up wanting AI. They want a faster claims process, less fraud, fewer stockouts, lower customer-acquisition costs, a better sales conversion rate or an analyst who can do the work of five without creating five times the risk.
This is why the AI boom can remain genuine even if plenty of AI startups get cleaned out. The rail network did not need every rail company to survive. The internet did not need every dot-com to survive. Infrastructure platforms create enormous winners, then ruthlessly expose everyone who mistook a feature for a business.
Nvidia also has one advantage most of its customers do not: it is already making absurd amounts of cash. It returned approximately $26 billion to shareholders in the latest quarter through buybacks and dividends, while retaining about $99 billion on its repurchase authorisation. The frontier labs and cloud builders, by contrast, are still spending colossal sums to secure capacity. ([investor.nvidia.com](https://investor.nvidia.com/news/press-release-details/2026/NVIDIA-Announces-Financial-Results-for-Second-Quarter-Fiscal-2027/default.aspx))
That gap matters. In any arms race, the supplier selling ammunition with a 75% gross margin is usually in a better mood than the generals paying for it.
What this means for you
If you are an investor, stop treating “AI” as one trade. Nvidia’s quarter reinforces that the stack has layers. Chips are one layer. Memory, networking, energy, cooling, construction, data-centre operations and enterprise software are others. Work out where demand is proven, where margins are defensible and where capital intensity will crush the weak players.
If you are a founder, do not build a generic AI product because it is fashionable. Pick one expensive workflow in an industry you understand. Measure the current cost in labour, delay, errors or lost revenue. Use AI only where it produces a result a customer can verify in dollars, hours or risk reduced. Your moat is not the model. It is the workflow, data and distribution you own.
If you run an established business, make a list on Monday morning of the 10 processes where your best people waste time moving information between systems, checking documents, preparing first drafts or chasing routine decisions. Choose one. Give it an owner. Set a 60-day test with a hard commercial target. No innovation theatre. No committee of 14 people producing a strategy PDF.
And if you are tempted to call this whole thing a bubble, fair enough — keep asking hard questions. Just do not let scepticism become an excuse for inaction.
Nvidia has just told you that demand for AI infrastructure is still outrunning supply. You do not need to buy Nvidia shares to benefit from that fact. But you do need to decide whether you will use this shift to make your business faster and more valuable, or wait until your competitors do it to you.