Everything About AI
NextFin News - Meta Platforms CEO Mark Zuckerberg joined Nvidia CEO Jensen Huang in pushing back against calls for a coordinated slowdown in AI development, arguing that safety is a competitive advantage individual labs can pursue on their own rather than a reason for an industry-wide pause. The two executives' remarks, delivered within days of each other, mark the clearest split yet in Big Tech over how to respond to warnings that increasingly capable AI systems could outpace human control — and they helped lift chip stocks after a selloff left the semiconductor-heavy Philadelphia index nursing its worst weekly losses in months.
The Situation: Two CEOs, One Message, Different Angles
The debate erupted after Anthropic CEO Dario Amodei urged AI companies to slow the pace at which they improve model capabilities, a call that OpenAI CEO Sam Altman and xAI's Elon Musk endorsed within hours. Their argument is that the technology is advancing faster than companies can manage its risks, and that a coordinated pause would buy time to put safeguards in place. Amodei has gone further, asking for antitrust permission to let rivals coordinate a slower pace of development.
Zuckerberg pushed back in an X post, saying every lab has the responsibility and incentive to move at the pace required to train its models safely, and the ability to take its own actions to ensure that happens. In his view, safety is becoming the defining competitive edge:
"Trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models. Any lab that doesn't focus on alignment will fall behind."
Huang took the same side from a different angle. Speaking at Salesforce's Dreamforce conference, he rejected new laws aimed at slowing AI development. "We don't need any new laws. We need new regulation," he said, framing safety as an engineering discipline rather than a legal constraint. "Safety is an engineering problem." Companies should test their own products extensively and hold back anything they are not confident about: "If we're not confident about the safety of the product, then don't release it," he said. "You pace yourself until you are confident you're releasing something that the market would appreciate."
The two positions converge on one point: speed and safety are not mutually exclusive. "I think speed, and safe products... it's a false choice," Huang said. "You can definitely have both at the same time."
The pushback also has political cover. In an on-stage phone call with Huang earlier in the week, President Donald Trump dismissed AI existential-risk warnings as "a hoax" that plays into the hands of political opponents and China. "We're not going to let that happen," Trump said, with Huang replying, "You're right. We're not going to let that happen, sir."
Markets reacted swiftly. Nvidia shares climbed as much as 4.8% to $209.18 during Wednesday's session before paring some gains, as traders treated the CEOs' comments as an attempt to talk the market off the ledge. The Philadelphia Semiconductor Index, which bore the brunt of the previous session's selling, is still up nearly 60% in 2026. The S&P 500 and Nasdaq Composite were both about 2% below all-time highs, leaving little margin for a growth scare.
Why the combination matters: Zuckerberg speaks for a hyperscaler that buys the chips; Huang speaks for the company that sells them. When both the buyer and the seller of AI infrastructure argue against a slowdown, the message to investors is that the physical buildout has more than one defender.
The Picks-and-Shovels Layer Has the Most to Lose
The market's anxiety is not evenly distributed, and that asymmetry is the key to reading the selloff. "Markets are punishing the picks-and-shovels layer harder than the hyperscalers because it's the layer most exposed to a slowdown in the rate of capability improvement," said Erik Kratz, chief investment officer and co-head of wealth at Arena Private Wealth in Chicago.
The mechanism is straightforward. If AI progress slows, a hyperscaler such as Microsoft, Amazon, or Alphabet can simply stop adding capacity and harvest the returns from the data centers it already built. The chip supplier has no such option: a cancelled order is revenue that does not come back, and a factory built for demand that never arrives is a stranded asset. That asymmetry is why semiconductor and infrastructure names fell hardest when the slowdown narrative took hold, even as the hyperscalers held up better.
The stakes are large enough to justify the jitters. The AI hyperscalers — Microsoft, Alphabet, Amazon, Meta Platforms, and Oracle — are expected to spend around $795 billion on capital expenditures this year and nearly $1.08 trillion in 2027, according to BofA Global Research. UBS puts the 2026 figure at roughly $673 billion, a 76% increase, before growth decelerates to 25% in 2027 and just 6% in 2028. The two estimates differ on the level but agree on the direction that matters: a boom that is still expanding, but at a sharply slowing rate.
"AI semiconductor and infrastructure stocks have long priced in an uninterrupted capex boom, leaving virtually zero margin of error for an industry-imposed speed limit," said James Humphries, managing partner at Mindset Wealth Management in Indianapolis. That is the real vulnerability: not that capital expenditure will collapse, but that the growth rate slows enough to break valuations built on perpetual acceleration. Empirical Research has highlighted the same gap, warning that "either the capex trajectory of the hyperscalers will be upgraded again, or the revenue growth pencilled in for their suppliers will have to come from elsewhere."
A Voluntary Slowdown Has No Enforcement — Which Is Why It Is Likely Cyclical
The central question is whether the slowdown threat is structural or cyclical. The answer turns on enforcement, and enforcement is exactly what is missing.
Amodei has asked for antitrust permission to let rivals coordinate a slower pace of development. FTC Chairman Andrew Ferguson responded that everyone should be "deeply suspicious" of AI companies seeking antitrust exemptions while lobbying for new regulations. Without that exemption, any coordination among competitors on the pace of development would run straight into antitrust law — which is precisely why it is unlikely to happen. The companies calling for a pause are asking regulators for legal cover to do something that is currently illegal; until that cover arrives, the pause is rhetoric.
This is a cyclical sentiment shock, not a structural break. A structural regime change requires a binding constraint — a law, a treaty, a physical limit on power or chips. None of those exists today. What exists is a conversation, and conversations do not cancel orders. As Chuck Carlson, chief executive officer at Horizon Investment Services, put it: "I need to see something concrete that, in fact, there is a slowdown versus just talk."
The reversion mechanism is incentives: each lab that slows unilaterally hands market share to the ones that do not. Zuckerberg made that logic explicit, arguing that competition and liability give companies enough reason to act on safety individually. Meta's own behavior illustrates the point. The company delayed the release of its Muse AI agent earlier this year to strengthen its security — without asking anyone else to wait. "We didn't call for everyone else to do this before we would," Zuckerberg said. "We just did it as part of our day-to-day work because it was clearly the right thing for people and for us."
Meta is also directing the "significant majority" of its computing resources toward products that serve users' immediate needs rather than pursuing recursive self-improvement — the capability Amodei singled out as carrying the greatest loss-of-control risk. Meta Superintelligence Labs already engages independent evaluators in several areas, a practice Zuckerberg called "industry best practice," adding that "other labs can just do this too."
That is the difference between a structural regime change and a cyclical pause. A cyclical claim needs a short-term driver and a demonstrated tendency to revert. The driver here is sentiment: a cluster of existential-risk warnings from researchers, followed by a public call from three high-profile CEOs. The reversion mechanism is competitive incentives plus the absence of legal cover. On that reading, the selloff is a sentiment cycle, and sentiment cycles revert once the narrative meets reality — in this case, the reality of capex guidance that is still pointing up.
Safety as a Moat — The Second-Order Case for Building
There is a second-order argument hidden in Zuckerberg's remarks, and it runs counter to the market's first read. If trust and alignment become the capabilities that differentiate agents, then safety investment is not a tax on growth — it is a moat. Labs that ship unsafe products face "significant" liability, Zuckerberg noted, and reputational damage that compounds across enterprise customers who will not trust an agent that misbehaves.
Kratz extended the logic to capital markets: "The buildout doesn't stop because the CEOs asked for guardrails. If anything, a credible safety framework makes the long-duration capex easier to underwrite." A decade-long data-center investment is easier to finance when the regulatory endpoint is clear. Uncertainty, not safety, is what raises the cost of capital — and that is the second-order implication the market has not fully priced. The first-order effect of slowdown talk is negative for chip stocks. The second-order effect is that a credible, industry-led safety framework could reduce regulatory tail risk and actually support the capex cycle by making it financeable at a lower cost.
Huang's framing — "safety is an engineering problem" — is an attempt to steer the industry toward that outcome: self-regulation through testing and release discipline, rather than statutory speed limits. It is also a defense of Nvidia's own growth option: the AI boom has turned the company into the world's largest chipmaker and most valuable company, and efforts to slow the development of more capable models pose a direct risk to that future revenue stream. The incentive alignment is transparent, which is why the counter-thesis deserves its due.
The Counter-Thesis: Zero Margin of Error
The strongest case against this optimism is also the simplest: the market does not need a real slowdown to sell off. It only needs the growth rate to decelerate. Hyperscaler spending is expected to slow from 76% this year to 6% by 2028 under UBS's forecast. If revenue growth for chip suppliers is priced for the 76% world and reality delivers the 6% world, multiples compress even if absolute spending keeps rising. A company can spend more every year and still disappoint a market that priced in acceleration.
Humphries' "zero margin of error" line captures the risk precisely. Nvidia's stock trades at a forward earnings multiple that assumes the AI capex boom continues uninterrupted. Meta and its peers are pouring hundreds of billions into data centers on the bet that AI revenue will eventually catch up. If monetization lags, the pressure will show up first in guidance, then in orders — and by then the picks-and-shovels rerating will already be underway. The Bank for International Settlements warned in June that weaker-than-expected returns on AI investments could reduce funding availability and eventually reverse the current capital expenditure boom — a reminder that the debt markets, not the CEOs, may set the ultimate speed limit.
There is also a public-opinion constraint that neither CEO can talk away. A Gallup poll found that seven in 10 Americans oppose constructing AI data centers in their local area, including 48% who are strongly opposed. Half of opponents cite excessive resource use — 18% each mentioning water and energy — while 16% mention pollution and about one in five cite quality-of-life effects such as traffic and higher utility bills. Grid interconnection queues and local permitting fights are real physical frictions, not sentiment. These are the structural legs of the bear case, and they do not require a CEO cartel to bite.
Finally, Meta carries its own credibility burden. The company agreed in August to pay up to $18 billion to settle lawsuits from 48 states alleging it designed Facebook and Instagram to addict children, without admitting wrongdoing. A company that just settled a child-safety case on those terms is not an unimpeachable voice on voluntary safety restraint — and investors will weigh its words accordingly.
Outlook: What Would Prove the Cyclical Read Wrong
The base case is that the slowdown narrative proves cyclical. Without antitrust cover, coordination is illegal; without enforcement, voluntary restraint is unstable; and with China in the race, no US executive can afford to unilaterally disarm. Hyperscaler capex should keep rising through 2027, even if the growth rate decelerates from this year's 76% pace.
The downside case is a guidance shock. If two or more hyperscalers cut their 2027 capital-expenditure guidance by double-digit percentages during earnings season, the picks-and-shovels bear thesis is confirmed and semiconductor multiples should compress. The upside case is that a credible safety framework emerges from industry self-regulation, reducing regulatory tail risk and making long-duration infrastructure financeable at a lower cost — the Kratz scenario.
Time horizons diverge, and they should be judged separately. In the short term, sentiment and headlines will drive semiconductor volatility; the gap between what is priced and what is said will dominate. Over the medium term, earnings and capex guidance will decide whether the buildout holds — and the BofA estimate of nearly $1.08 trillion in 2027 spending is the benchmark to beat. Over the long term, the structural question is not whether AI slows; it is whether the electricity, grid capacity, water, and public consent exist to keep building.
The falsifying signal for the cyclical view is specific and observable: if hyperscaler capex guidance for 2027 falls materially below the current roughly $1 trillion consensus — a collective cut of 15% or more — then this is not a sentiment cycle and the picks-and-shovels rerating has further to go. Concrete order cancellations or data-center construction delays, the evidence Carlson says he needs to see, would confirm the same turn.
The AI slowdown debate is less about safety than about who owns the option on the future — and for now, the companies that build the chips still hold the microphone, the capex, and the political ear. That is a strong hand, but it is not an unbeatable one: the moment guidance breaks, the narrative flips.
Explore more exclusive insights at nextfin.ai.
