NextFin News - US equity futures edged lower on Sunday night as the bill for the artificial-intelligence boom came back into focus, with Nvidia Corp. raising prices on key data-center hardware and Alibaba Group Holding Ltd. launching a $10.2 billion share sale to fund its own AI buildout. The combination puts a question under the spending pace that has powered the market's advance: the same capital-expenditure surge that lifted stocks is now being scrutinized for its cost, its funding, and who ultimately pays.
The benchmark S&P 500 closed Friday at 7,674.37, up 0.43%, with the Nasdaq Composite also gaining 0.43% to 26,180.45 and the Dow Jones Industrial Average adding 517.80 points, or 0.98%, to 53,277.01. The rebound came after a week in which all three major indexes were on track for losses of about 2%, snapping a three-week winning streak for the S&P 500 and the Nasdaq. Even so, the index remains roughly 2% below its closing record of 7,798.99 set on August 13, and the Cboe Volatility Index stood at 15.13 at Friday's close, down from 16.01 the prior session. Futures pointed to a softer open to a week that will test whether investors are willing to keep funding an AI buildout whose price tag keeps rising.
The Bill for the AI Boom Arrives in Three Parts
Three developments, arriving within days of one another, frame the tension now running through equities. First, Nvidia — the supplier at the center of the AI infrastructure trade — raised prices again. Its RTX PRO 6000 Blackwell workstation GPU, a 96GB card launched in March 2025 at $8,565, now lists at $16,000 on the company's own US marketplace, an increase of roughly 87% in sixteen months with no hardware change in between. The card first moved to $13,250 in mid-2026 before the August increase. Across the consumer lineup, price adjustments have been broad: the RTX 5060 moved from $369.99 to $469.99, a 27% increase, and the RTX 5070 from $659.99 to $899.99, up 36%.
Second, Alibaba said it would sell 710 million shares at HK$112.7 each, raising about HK$80 billion ($10.2 billion) — a 3.6% discount to Friday's close — in what would be the largest primary follow-on offering ever by a Hong Kong-listed company. The deal, with Morgan Stanley, HSBC, UBS and CICC as joint bookrunners, was oversubscribed and its size was increased after strong demand, including from sovereign wealth funds. The proceeds are earmarked for AI-related development.
"In order to be able to capture that future growth, we first need to make these capex investments to build out the necessary compute capacity," Alibaba chief executive Eddie Wu said on the company's earnings call.
Third, the Canadian dollar slipped as US trade tensions intensified. The United States is imposing 50% tariffs on roughly $20 billion of Canadian goods — covering dairy, alcohol, electronics, building materials, apparel and motor vehicles — effective August 19. Canadian Prime Minister Mark Carney said Ottawa would match the duties "dollar for dollar" after last-ditch talks in Washington failed; US Trade Representative Jamieson Greer said Canada declined to finalize a deal under terms agreed earlier in the week. The tariffs apply even to goods covered by the US-Mexico-Canada trade agreement and carry no expiry date.
Taken together, these are not random headwinds. They are three faces of the same problem: the AI buildout is running into the real-world constraints of supply, funding and geopolitics. Input prices are rising. Funding is being tapped from equity markets at a discount. And the trade architecture that underpins North American supply chains is fracturing at the same time.
The Price of AI Is Rising at Both Ends of the Chain
The first-order read of Nvidia's price increases is simple: demand exceeds supply, so the seller prices it. But the mechanism that matters for the market runs deeper. When the key input to the AI economy — accelerated compute — becomes more expensive, two things happen simultaneously. Hyperscalers' capital efficiency deteriorates: each dollar of capex buys fewer floating-point operations than it did a year ago. And the inflation signal from the chip layer feeds into the broader price level, which is exactly the channel through which a sector-specific boom becomes a macro problem.
Consider the scale. Estimates for hyperscaler AI-related capital expenditure for 2025 began the year around $250 billion and now sit above $405 billion. For 2026, a CreditSights estimate puts combined capex for the top five hyperscalers — Amazon, Alphabet, Meta, Microsoft and Oracle — at roughly $750 billion, up about 67% year over year and marking a third consecutive year of growth exceeding 60%. Morgan Stanley's strategists have gone further, lifting their 2026 forecast to $805 billion and their 2027 estimate to $1.1 trillion. At Nvidia's second-quarter fiscal 2026 earnings call, chief executive Jensen Huang put the annual spending of the top four hyperscalers at about $600 billion. Independent composite estimates that add the big four plus Oracle land closer to $417 billion, with individual company guidance in the $65 billion to $120 billion range per firm.
The gap between Huang's $600 billion and the $417 billion composite is itself informative. Part of it is scope ambiguity — what counts as hyperscaler capex, and whether global supply-chain buildouts are included. Part of it is narrative incentive: the supplier benefits from framing the race as larger and more urgent than the buyers' own disclosures suggest. Investors should treat the higher number as the bull case, not the base case.
Here is the uncomfortable arithmetic. If the price per unit of compute rises 87% inside a product generation while capex dollars rise at a similar or faster clip, the number of units delivered does not keep pace with the dollars spent. That is the definition of deteriorating capital efficiency — and it is the metric that will eventually decide whether this buildout is an investment boom or an investment bubble. The market has not yet priced that distinction, because revenue growth at the chip supplier has so far arrived on schedule. The second-order question is whether the buyers' revenue will arrive on the same schedule.
Structural Buildout, Cyclical Funding Friction — Keep Them Separate
This is the call the market needs to make, and conflating the two legs of it is the most common error. The demand for AI compute is structural: it is driven by a technology shift in how software is built and how enterprises process information, backed by rules, industry structure and a multi-year capex cycle that cannot be reversed on its own. Once a data center is permitted, powered and racked, the spending does not simply pause because sentiment sours. The hyperscalers are locked into a capacity race with each other; falling behind on compute is an existential competitive risk, which is why they keep spending even when near-term returns are unclear.
The funding of that buildout, however, is cyclical. Equity issuance at a discount — Alibaba's 3.6% discount to the prior close is the latest example — is how markets ration capital when internal cash flow cannot cover the ambition. Alibaba's cloud numbers show why the story is not one-sided: revenue from AI Cloud and Compute Services reached 48.4 billion yuan in the June quarter, up 45% from 33.4 billion yuan a year earlier, the fastest pace in 22 quarters and an acceleration from 40% external growth in the March quarter. Segment profit more than doubled, rising 133% to 5.63 billion yuan and taking the margin to 12%. That is real inflection, not a mirage.
But a 45% growth rate in one division still has to justify a $10.2 billion equity raise, and the capital intensity is climbing faster than the revenue. Fiscal 2026 capex ran $18.3 billion across four quarters, averaging about $4.6 billion per quarter. The June quarter alone hit $9.98 billion — more than double the prior run rate in a single period. If that pace holds, fiscal 2027 capex lands near $40 billion, a 118% step-up in capital intensity inside one year. The market's willingness to absorb that supply at favorable prices is a cyclical variable that mean-reverts with liquidity conditions.
The evidence for the structural leg is the durability of the driver: enterprise AI adoption, sovereign AI programs, and the competitive necessity of owning compute capacity. The evidence for the cyclical leg is the funding mix shifting toward external capital and the rising cost of every input. A structural claim without evidence of permanent regime change is a story; a cyclical claim without demonstrated mean reversion is a guess. Both tests are met here, which is why the correct posture is not "the AI trade is broken" or "the AI trade is unstoppable." It is: the destination is structural, the path is cyclical, and the path is where the volatility lives.
Short-knife close: the buildout is real; the financing is getting harder.
The Physical Constraint: Power Is the Real Bottleneck
There is a second-order constraint that most equity investors are still underweight: the buildout is no longer limited by chip supply alone. It is limited by power. Microsoft has disclosed an order backlog of roughly $80 billion in Azure demand that it cannot fulfill because of power constraints. That number reframes the entire capital cycle. When the binding constraint moves from semiconductors to the electrical grid, the capex multiplier expands beyond the chip layer into utilities, transformers, switchgear, copper and construction — and the timeline stretches, because permitting and grid interconnection queues move on a political clock, not a product cycle.
This is where the trade dimension bites. Tariffs on building materials and electronics land directly on the cost of turning a data-center shell into working capacity. A 50% duty on a category of electrical or construction inputs does not just raise the price of that input; it raises the cost per megawatt of deliverable capacity, which forces hyperscalers to spend more to hit the same capacity target, which asks the equity and debt markets to fund a wider gap. The transmission chain is complete: chip prices up, trade costs up, power constrained, funding tapped at a discount. Each link pushes the breakeven on AI revenue further out.
The cross-asset footprint confirms the market is starting to price this. In recent sessions the 10-year Treasury yield traded around 4.74% and the 30-year near 5.27%, while gold futures climbed 1.97% to $4,661.60 an ounce — the classic signature of a market hedging both inflation and fiscal-risk exposure at the same time. West Texas Intermediate crude held near $86.64 a barrel, supported by geopolitical tensions and the prospect of stronger demand from data-center power generation. The bond market is not yet screaming, but it is no longer whispering.
The Strongest Case Against the Trade, and the Signal That Breaks It
The bear case deserves its due, because it is coherent and it is backed by a long line of capex cycles that ended badly. The argument runs like this: Jensen Huang's $600 billion figure is not validated by the hyperscalers' own guidance, which points to roughly $400 billion to $450 billion for the combined group; Alibaba is turning to equity markets because internal cash cannot keep up with its ambition; and history — the fiber-optic overbuild of the late 1990s, the shale capital-expenditure boom of the 2010s — says that when an industry spends ahead of demand, the correction is severe and the suppliers get re-rated first. On this read, Nvidia's price hikes are not a sign of pricing power but of a supplier front-loading revenue before the cycle turns.
That case is serious. But it rests on an analogy that does not fully hold. The fiber overbuild failed because demand for dark fiber never arrived at the price builders assumed. The shale boom broke because the commodity price that justified drilling fell below the cost of production. AI demand has arrived — cloud revenue is accelerating, AI product revenue is growing at triple-digit rates, and the buyers are not leveraged startups building redundant networks. They are cash-generative platforms with near-monopoly distribution, competing for a scarce input whose scarcity is technical, not financial. The question is narrower and more falsifiable than "is this a bubble": is AI revenue per compute dollar rising fast enough to justify a multi-trillion-dollar installed base?
Which brings us to the signal that would prove the structural-buildout thesis wrong. Watch the four largest US hyperscalers' combined quarterly capital expenditure growth. If it falls below 10% year over year for two consecutive quarters, or if AI-related revenue per dollar of capex fails to rise through 2026, the regime-shift call breaks and the cyclical-bubble read takes over. Until then, the base case is continued spending with rising volatility in the funding layer.
Who Benefits, Who Is Exposed, and What to Watch
Cashing the mechanism into concrete implications: the beneficiaries of a still-rising capex tide are the suppliers with pricing power — the accelerated-compute layer, high-bandwidth memory, custom silicon designers, and the power-and-cooling infrastructure that sits behind every rack. Utilities with data-center load growth and the electrical-equipment makers that serve them sit on the other side of the same trade. The exposed are the buyers whose capital efficiency deteriorates fastest, and any company whose cost base leans on tariff-hit inputs — building materials, electronics and motor-vehicle suppliers in North America are the first line of exposure.
Across asset classes, the setup favors real assets and inflation hedges over duration, as long as the 10-year yield holds above roughly 4.5%. Gold's move toward $4,700 is the market's own vote on that positioning. Equities are not excluded, but the leadership should narrow from "the companies that spend the most" to "the companies that monetize the spend." That is a rotation, not a rout — unless the funding layer breaks first.
Split by time horizon, the picture is mixed. In the short term, sentiment and liquidity dominate: a soft futures open, a VIX near 15, and a week of data ahead mean volatility can spike on any hint that spending is slowing or that funding costs are rising. In the medium term, fundamentals decide: the next round of hyperscaler earnings reports will show whether capex guidance moves up again or finally flattens. In the long term, the structural question dominates: whether AI revenue per compute dollar rises enough to justify the installed base being built.
Three scenarios frame the path. The base case: capex keeps climbing but at a decelerating rate, equity issuance absorbs the gap, and the market grinds higher with elevated sector rotation. The upside case: AI monetization accelerates faster than capex, capital efficiency improves, and the suppliers re-rate again. The downside case: hyperscaler guidance flattens for two consecutive quarters, the tariff war widens beyond Canada, and the funding layer — not the technology — breaks first.
The week ahead will tell investors which path they are on. The market has spent two years rewarding the companies that spend the most on AI. It is about to start rewarding the companies that spend the smartest.
Explore more exclusive insights at nextfin.ai.
