Then Jensen Huang walked in with a forecast large enough to drown out the funeral music.
Nvidia to the Rescue
Nvidia () arrived just as the artificial-intelligence trade was beginning to look like a grand investment thesis in search of a business model.
For several weeks, investors had been circling the same uncomfortable questions. Was the AI boom becoming an enormous construction project with no tenants? Were companies spending billions on computing capacity simply because they feared being left behind? And if the machines were becoming more intelligent by the month, what would happen to the software companies whose business models appeared to be standing directly in their path?
Then Jensen Huang walked in with a forecast large enough to drown out the funeral music.
Nvidia expects revenue to grow by roughly 70% in fiscal 2028, while unconstrained demand for its 2027 products is reportedly running above 100%. Blackwell, Rubin, Vera and Groq are not struggling to find customers. They are struggling to find enough capacity to meet them.
That is a very different problem from an AI spending cycle running out of road. The queues are still forming outside the data centres. The customers are still waiting for their hardware. For at least one session, it was Chips Ahoy.
Nvidia surged almost 9%, its best day since April 2025, adding roughly $442 billion to its market value. The 100 rose 1.4%, the broader Nasdaq gained 1.6%, and the advanced 0.7%.
From a distance, it looked as though the AI trade had fully restored investor confidence. Up close, the picture was far less convincing. Ten of the eleven S&P 500 sectors declined, with technology, up 3.4%, doing almost all the heavy lifting, while healthcare and consumer stocks fell about 1%.
Nvidia had not lifted all boats. It had thrown a towline around the index and dragged it higher, leaving much of the market still treading water. But that narrowness should not obscure the more important shift taking place inside technology. Nvidia did more than reassure investors that chip demand remains powerful. It reopened the debate over where the economic value of AI will ultimately settle.
For one day, at least, the answer was not confined to the chips. It was in the software.
Salesforce () surged 23%, CrowdStrike () gained 20%, and also rallied sharply. The companies that had recently been pushed to the edge of the AI narrative, treated as the old machinery that a new generation of autonomous systems would make obsolete, suddenly found their footing again.
The software sector had been cast aside as the ill-fitting middle child of the AI revolution. The machines were exciting. The chips were scarce. The data centres were strategic. Software, meanwhile, stood in the doorway, wondering whether anyone still needed it.
Then it found its mojo.
Salesforce gave investors evidence that AI can be monetized rather than simply bolted onto an earnings presentation. CrowdStrike and offered an even more tangible route to revenue, showing how the rise of autonomous AI agents could create fresh demand for cybersecurity, identity management and enterprise controls.
The more independent these systems become, the more businesses will need to monitor, authenticate, and restrain them. A digital workforce that can act on its own will also require a digital security guard capable of checking where it has been and what it has touched.
That was the part of the session that deserved the most attention. Nvidia confirmed that the hardware bottleneck remains real. The software companies began to demonstrate that the value may eventually spread beyond the chipmakers.
The market was not simply rewarding earnings beats. It was listening for evidence that AI demand could remain visible over several years, and that software companies might capture part of the value being created rather than merely defend themselves against extinction.
The software apocalypse, it turns out, may have been scheduled too early.
That is the constructive case. The less comfortable question is whether the spending required to sustain the next phase of the boom will eventually generate enough revenue to justify it.
Nvidia can show that its customers are still buying. It cannot force every company across the economy to divert free cash flow into another enormous capital-expenditure cycle. At some point, investors will want to see earnings flowing through the system rather than simply purchase orders flowing into Nvidia.
Demand is real, but demand alone does not guarantee attractive returns. The market has seen many industries enjoy extraordinary order growth before discovering that the cost of meeting that demand was almost as extraordinary as the revenue itself.
The session’s internal behaviour suggested that investors understood the distinction. Momentum baskets underperformed, heavily shorted shares were squeezed, and parts of the AI complex surrendered their early gains. Data-centre and optical-networking stocks lagged, while software led and semiconductors followed.
The rotation had a certain elegance to it. Investors were not buying everything with an AI label. They were moving toward the companies offering the clearest earnings visibility, the strongest positioning dynamics and the most believable route from artificial intelligence to cash flow.
Software stole some of the spotlight from chips, but the light was still concentrated in a handful of large names.
That was visible in the broader market. The S&P 500 finished higher, but most of its members did not. It was not a rising tide. It was a narrow current running through the largest and most liquid stocks, carrying the headline indexes while leaving much of the market standing on the shore.
The bond market was even less willing to join the celebration.
Treasury yields rose by roughly 2 to 3 basis points across the curve as investors weighed resilient economic data, sticky inflation, higher oil prices and the continuing fiscal supply problem. Initial jobless claims fell by 4,000 to 203,000, slightly below expectations, confirming that the labour market remains firm. The July goods-trade deficit widened to $118.8 billion, while man Sachs lowered its third-quarter GDP tracking estimate by 0.1 percentage point to 2.7%.
That is an awkward environment for richly valued growth stocks. The economy remains strong enough to support earnings, but not soft enough to guarantee easier monetary policy. Inflation is no longer raging, but it has not been defeated. Fiscal concerns have not disappeared, and oil is beginning to lean back into the inflation argument.
The seven-year Treasury auction was middling at best. Demand was sufficient to move the paper, but hardly strong enough to suggest investors are eager to add duration while oil, inflation and government borrowing are all pushing against the long end.
Bessent’s expanded buyback programme has helped arrest the long-end rout by supporting older and less-liquid Treasury issues. But it has also raised an awkward question: is the Treasury market finding a sustainable equilibrium, or is official support simply keeping the furniture upright while the floor continues to tilt?
The long end has not been persuaded that the problem is solved. It has merely been given fewer reasons to panic today.
That leaves Kevin Warsh as the next major test for the Nvidia-led relief rally.
The rates market had been leaning toward a September hold, but that assumption is no longer as comfortable as it was. Firm data, sticky underlying inflation and higher oil prices have pushed the probability of a hike to roughly one in three, with at least one increase still priced into the remainder of the year. That is not an outright tightening signal, but it is a meaningful shift for a technology complex whose valuation remains highly sensitive to the cost of money.
The Fed’s recent commentary has kept that hawkish drift alive. Jeff Schmid says policy is not restraining the economy, while Beth Hammack argues that officials need to act to contain price pressures. Susan Collins offers the softer counterpoint, describing rates as mildly restrictive. The Fed is not speaking with one voice, but it is speaking loudly enough to keep the bond market nervous — and to make Warsh’s Jackson Hole message unusually important.
Warsh will therefore have to thread a narrow needle at Jackson Hole. Goldman Sachs’ Rich Privorotsky’s “modal view is that Jackson Hole is a push to nothing.” Goldman does not expect Warsh to provide strong hints about the Fed’s September decision. Instead, he may acknowledge the better June and July inflation data, reiterate the 2% target and discuss the broader work of the Fed’s five task forces — covering communication, the balance sheet, economic data, productivity and jobs.
In other words, Goldman sees a speech that is neutral at worst and mildly dovish at best: no policy shock, no clear September signal and no attempt to derail the Nvidia-led relief rally.
Apollo’s Torsten Slok is looking at the same speech through a different lens. He expects Warsh to deliver an economic outlook with a “hawkish flavor”, partly to keep long-term yields in check. Warsh would not need to signal an imminent rate hike to unsettle markets. He would only need to make clear that the Fed remains fully committed to price stability and is not prepared to validate easier financial conditions simply because technology stocks have enjoyed a spectacular day.
That is the narrow needle: say enough about improving inflation to avoid choking the equity rally, but sound sufficiently firm on the 2% target to prevent the bond market from treating Jackson Hole as an invitation to push yields higher. Rich sees a push to nothing. Torsten sees a hawkish edge. The market is now waiting to find out which version of Warsh steps to the microphone.
The timing is particularly important because oil is pushing in the opposite direction.
rose as hopes for a quick reopening of the Strait of Hormuz weakened. The Trump administration has reportedly indicated that it is not interested in returning to the terms of the June agreement with Iran, while a tanker was struck by an unknown projectile during the week. Tehran has also made clear that any agreement on navigation would not necessarily mean an immediate reopening of the chokepoint, through which roughly a fifth of global oil supplies normally pass.
Diplomacy remains the market’s preferred escape route, with Qatar, Iran and Oman working toward a framework for managing navigation. But the difference between an agreement in principle and actual physical flows remains substantial. Only five commodity vessels transited on Tuesday, compared with a ten-day average of fifteen.
The dollar remains strangely unmoved overnight despite the prospect of higher US rates, suggesting that elevated yields alone are no longer enough to generate unquestioned confidence in the currency. Gold is benefiting from the same fiscal and debasement anxiety that keeps the long end of the Treasury curve unsettled.
Nvidia has shown that the AI trade is still alive. Demand is real, supply remains tight, and software companies are beginning to demonstrate that artificial intelligence may create revenue rather than simply destroy business models.
But the rally has not yet broadened, the bond market has not relaxed, and oil has not delivered the all-clear.
For now, Jensen Huang has bought the market another day. More importantly, he has reminded investors that the AI trade is not a single bet on chips. It is a chain, and software — the link everyone had begun throwing overboard — has suddenly found itself back on deck.
The question now is whether that broader AI chain can stay intact once the applause fades, the bond market has its say, and Kevin Warsh steps to the microphone.