Everything About AI
NextFin News - Nvidia's board authorized a $150 billion increase to its share repurchase program on Monday, lifting the remaining buyback capacity to $235 billion — the largest repurchase authorization increase on record — while SpaceX reached orbit with Starship for the first time and Meta kept pressing ahead with an AI infrastructure buildout that could cost as much as $145 billion this year. Taken together, the three developments mark the moment the artificial-intelligence boom stopped being a single-stock rally and became a capital-allocation regime with clear winners, payers, and enablers.
The split is stark. Nvidia, the company that has sold the picks and shovels, is now returning cash to shareholders at a scale no company has attempted. Meta, one of the biggest customers, is spending at a pace that would rank it among the world's largest capital projects. And SpaceX, on Monday, cleared the hurdle that makes its satellite constellation — and the launch cadence AI-era infrastructure depends on — credible. The question investors now face is not whether AI is real, but who captures the cash and who merely funds the buildout.
The Buyback That Reframes the AI Winners
Nvidia's announcement is a statement of financial maturity as much as confidence. The $150 billion authorization increase raises the remaining repurchase program to $235 billion, and the company said it expects to execute the full amount through fiscal year 2028. With about 16 months between the announcement and the end of fiscal 2028, that implies an annualized buyback pace of roughly $176 billion — enough to retire about 3.2% of the company's roughly 24.3 billion shares each year if executed at current prices near $225.
"NVIDIA's growth is being driven by a once-in-a-generation platform shift to AI and accelerated computing," Jensen Huang, founder and CEO of Nvidia, said in the company's statement. "Our cash generation gives us the capacity to invest in the technologies that advance this transformation and return capital to shareholders. This authorization reflects our confidence in the long-term opportunity ahead."
The cash behind the buyback is not hypothetical. In the fiscal second quarter of 2027, which ended July 26, 2026, Nvidia returned approximately $26.0 billion to shareholders through share repurchases and cash dividends — more than the $21.3 billion of free cash flow it generated in the quarter — while revenue rose 106% year over year to $96.2 billion. As of quarter-end, about $99.0 billion remained under the prior authorization; Monday's increase more than doubles that. At a market capitalization of about $5.45 trillion as of late September, the $235 billion authorization equals roughly 4.3% of the company's value — large, but not desperate. The trailing price-to-earnings multiple of about 28 times sits well below Nvidia's 10-year median of roughly 53 times.
What the buyback signals is that Nvidia's management sees the AI infrastructure buildout as a multi-year cycle rather than a one-time spike. Returning capital while still investing in the next platform is the classic behavior of a company that believes its cash flow is durable. The alternative — hoarding cash or overpaying for acquisitions — would have suggested management saw fewer internal opportunities. Instead, Nvidia is telling shareholders that the best risk-adjusted use of its own capital is its own stock.
There is a second-order message here that the market should not miss. A company of Nvidia's size repurchasing aggressively reduces the equity float available to index funds and momentum strategies, which can amplify volatility on any disappointment. The buyback is bullish if cash flow holds; it is a lever that magnifies the downside if AI spending slows faster than expected.
SpaceX's Orbital First: The Enabler Clears Its Hurdle
While Nvidia monetized the present, SpaceX cleared a gate that matters for the future. Starship Flight 14 lifted off at 7:48 a.m. Central Time from Starbase, Texas, and reached orbit for the first time in the vehicle's 14-flight test campaign. Elon Musk confirmed the result on the company's X platform: "First orbital flight of Starship successful!"
The mission was not flawless. One of Starship's six Raptor engines failed during ascent, and the flight was shortened from a planned 10 hours to about three hours. But the vehicle still deployed 26 next-generation Starlink V3 satellites over roughly 30 minutes beginning about 34 minutes after liftoff, at an altitude of about 269 kilometers — the first time V3 satellites have been placed into orbit rather than on a suborbital arc that falls back to Earth. The upper stage completed a controlled splashdown.
The stakes extend well beyond broadband. Starship is the launch vehicle NASA is betting on to land Artemis astronauts on the moon, and SpaceX has said it wants to retire the Falcon 9 and Falcon Heavy fleets in favor of Starship once the vehicle flies "reliably several times per week." Musk has framed that cadence as necessary to conserve "scarce SpaceX engineering and production resources" — a phrase that matters for anyone tracking the company's path to profitability as a public company following its June IPO, the largest in history.
Notably, SpaceX did not attempt to catch either the Super Heavy booster or the Ship upper stage with the launch tower's "chopstick" arms on this flight. The company had already walked back the Ship-catch plan, with Musk saying an explosion would be too large a setback at this stage. The booster, which struggled to relight its engines after separating from the upper stage on Flight 13 in July, flew with "several modifications" to its hardware and software. The decision to forgo the catch is a reminder that orbital velocity is the hard problem — and SpaceX just solved the hardest part first.
Meta's AI Push: The Payer's Dilemma
If Nvidia is the harvester and SpaceX the enabler, Meta is the payer — and it is paying at a scale that is testing investor patience. The company expects 2026 capital expenditure to land between $130 billion and $145 billion, with as much as $145 billion directed at AI infrastructure, roughly double last year's investment. That figure represents a significant share of the more than $700 billion Big Tech is projected to spend on AI in 2026.
Meta's plan is specific. The company is doubling computing capacity to 7 gigawatts this year and doubling again to 14 gigawatts in 2027, supported by 32 data centers in operation or under construction. The spending also extends to silicon: Meta plans to begin production in September of Iris, its first in-house AI chip, designed with Broadcom and fabricated by TSMC under the MTIA (Meta Training and Inference Accelerators) program. The MTIA 300 is already in production handling ranking and recommendation workloads, while the 450 and 500 variants target generative image and video inference through 2027. A custom-silicon partnership with Broadcom runs through 2029, and Meta has struck a multiyear agreement with AMD to deploy up to 6 gigawatts of Instinct accelerators.
"As AI usage in our products and businesses continues to ramp, we continue to invest aggressively in infrastructure to meet the demand," CEO Mark Zuckerberg said on Meta's second-quarter earnings call. "We expect that a significant portion of our compute is going to go towards training our models, growing our core business and delivering personal agents and new products, but we also expect to grow a large business serving large customers as well."
That last clause is the pivot point. Zuckerberg is signaling that Meta does not intend to be a pure cost center in the AI stack — it wants to sell compute. Meta's stock has risen about 14% in 2026, trailing Nvidia and Amazon, and the market has punished capex increases: after the first-quarter report, the shares fell 10% despite a 57% earnings surprise. At a market capitalization of about $1.9 trillion and a trailing price-to-earnings ratio of roughly 28 times, investors are paying for execution, not optionality.
Cyclical or Structural: What the Three Stories Actually Say
The central question is whether this is a cyclical capital supercycle that will mean-revert, or a structural regime shift that will not. The evidence points to both — operating on different time horizons, and confusing investors who treat them as one.
The cyclical leg is real and visible. AI infrastructure spending is front-loaded: data centers are being built before the applications that will fill them exist. History offers a template. The fiber-optic buildout of the late 1990s saw capital expenditure surge ahead of demand, followed by a brutal correction when capacity arrived before customers. The semiconductor cycle has swung between shortage and glut on a roughly three-to-four-year rhythm for decades. Any company caught overbuilding — including Meta — faces the risk that today's scarce compute becomes tomorrow's idle capacity.
But the structural leg is equally real, and it is what separates this cycle from the dot-com buildout. Three differences matter. First, the demand is already monetized: Nvidia's 106% revenue growth and Meta's 28% revenue jump to $60.8 billion in the second quarter show customers paying now, not someday. Second, the technology is a general-purpose platform, not a single product — AI touches search, advertising, software, drug discovery, and defense, which spreads the demand base. Third, the supply chain is concentrated: a handful of companies control the advanced packaging, the high-bandwidth memory, and the fabrication capacity, which gives pricing power to the bottlenecks rather than letting it leak to commoditized builders.
The correct read is that the spending is cyclical — it will normalize as capacity catches up with demand — but the profit pool is structural, and it sits with whoever controls the bottleneck. Nvidia owns the bottleneck today. Meta is trying to become one by building its own silicon and selling excess capacity. SpaceX is building a bottleneck in heavy-lift launch access.
The Counter-Thesis: What If the Market Is Right to Doubt the Builders?
The strongest case against this framing is that the market is simply ahead of management on the math. The bear argument, articulated by analysts who have flagged Meta's capex trajectory, is that spending $145 billion a year on AI infrastructure requires a new revenue stream large enough to justify it — and that "personal agents" and a cloud business are not yet proven at that scale. If Meta's return on invested capital on AI spend lands in the low single digits for an extended period, the stock will re-rate lower regardless of how impressive the engineering is.
That argument has force. Meta's own history shows the pattern: the market rewarded margin expansion when AI spending was a line item; it has punished the stock when the line item became the story. The 10% post-earnings drawdown in April 2026 on a 57% earnings beat is the clearest evidence that investors are discounting future returns, not current profits.
The answer to the bear is that the option value is asymmetric. If Meta's AI agents and cloud business work, the company captures a slice of a multi-trillion-dollar computing market from a roughly $1.9 trillion starting valuation. If they do not, the core advertising business — 3.6 billion daily active users, growing 3% year over year — still funds the experiment. The risk is not existential; it is a valuation multiple that assumes execution. That is a different risk than the one dot-com infrastructure builders faced, where the revenue never arrived at all.
The falsifying signal is concrete: if Meta's capital expenditure stays above $120 billion annually for two consecutive years while AI-attributed revenue remains below 10% of total revenue, the structural-shift thesis for the builders is wrong, and the spend is a cyclical overbuild. Watch the company's quarterly disclosure of AI revenue contribution and capex guidance — those two numbers, together, settle the debate.
What Comes Next: Three Horizons
Short term (sentiment and liquidity): Nvidia's buyback provides a floor under the stock, but the market will test whether management actually deploys the authorization quickly. A slow rollout would read as a confidence signal withdrawn. SpaceX's shares — as a newly public company — will react to the cadence of the next Starship flights; one failure resets the narrative. Meta will remain under pressure into its next earnings report, where capex guidance will matter more than the headline EPS.
Medium term (fundamentals): The key variable is capacity utilization. If data centers fill as fast as they are built, Meta's spending looks prescient. If utilization lags, the market will force a capex cut, and the suppliers — Nvidia, Broadcom, AMD — will feel the order slowdown with a lag. SpaceX's test of orbital refueling, the next major milestone for the Artemis architecture, will determine whether the moon-landing timeline holds.
Long term (structural): The companies that own bottlenecks win. Nvidia's position in accelerated computing, SpaceX's position in heavy-lift launch, and — if it executes — Meta's position in both silicon and distributed compute are structural assets. The companies that simply rent capacity and hope for a software payoff face the mean reversion.
Base case: AI spending normalizes to a still-elevated plateau, the bottlenecks keep pricing power, and the builders that monetize excess capacity earn their keep. Upside case: AI agents become a mass-market product faster than expected, turning today's capex into tomorrow's cash machine. Downside case: utilization disappoints, capex is cut, and the multiple compresses across the builders while the bottleneck owners give back only part of their gains.
The three stories from Monday converge on one judgment: the AI boom has entered its capital-allocation phase, where the market stops asking whether the technology works and starts asking who gets paid. Nvidia is answering with buybacks. Meta is answering with spending. SpaceX just answered with a rocket in orbit. The market will decide which answer compounds.
Explore more exclusive insights at nextfin.ai.
