BofA’s Vivek Arya calls AI “slowdown” fears “noise” after a semiconductor sell-off
He points to near-100% AI infrastructure utilization and a 14.8x 24-month-forward SOX multiple as demand stays tight.
Semiconductor and AI-related stocks fell Monday after Anthropic CEO Dario Amodei urged restraint in frontier model development, with Sam Altman and Elon Musk backing the call. Bank of America’s Vivek Arya argued the safety-driven “slowdown” debate is tradable “noise,” citing tight AI infrastructure and valuations that already price in deceleration.
AI Safety Push Triggers a Semi Sell-Off, and BofA Calls It “Noise”
The latest AI safety flare-up hit semis like a demand shock, even though the underlying trigger was rhetoric, not a supply-chain print. Over the weekend, Anthropic CEO Dario Amodei published an essay urging the industry to slow the pace of frontier model development, and the call drew public backing from OpenAI CEO Sam Altman and Elon Musk. Semiconductor and AI-related stocks sank Monday in the wake of that debate.
Bank of America semiconductor analyst Vivek Arya’s pushback is that the market is trading the wrong variable. “The key question is not whether AI executives want more restraint. It is whether commercial and geopolitical competition will allow it,” he said. Arya described the controversy as “noise” inside a “secular bull market,” and framed the medium-term base case as continued AI capital expenditures, with annual AI capex potentially rising above $3 trillion by 2030.
Arya tied that view to physical-market signals rather than executive intent. He said AI network capacity is operating near 100%, and rental prices are rising even for older-generation accelerators, a pattern he argued is inconsistent with a cooling cycle because older chips typically get cheaper as new supply arrives. The mismatch, in his framing, is that equities are “effectively pricing a future spending slowdown” while the compute layer is still running hot.
Valuation is the second leg of the “noise” call. Arya cited the Philadelphia Semiconductor Index as up 67% in 2026, but still trading around 19x forward earnings, roughly in line with the S&P 500, even as semiconductor earnings were described as up 139% year-over-year. Looking further out, he cited a 24-month-forward multiple of 14.8x earnings for the chip index, described as an 11% discount to the S&P 500 alongside projected growth of about 30%, and noted the index corrected roughly 17% after nearly doubling in the first half.
Competition, Not Restraint, as the Real Capex Driver
Arya’s mechanism is a prisoner’s dilemma, a setup where everyone might prefer collective restraint but each participant has an incentive to defect because slowing alone is punished. In AI terms, frontier model development is the work on the most advanced systems at the edge of current capability, and AI capex is the long-duration spend on data centers, accelerators, and networking required to train and run them. If one lab or one platform pauses, it risks losing customers, talent, and perceived leadership to the next actor willing to keep spending.
He extends that logic beyond a single company’s governance posture. American labs compete with Chinese developers, hyperscalers race specialized cloud providers, and governments fund sovereign AI systems. Even if new safety agreements change how systems are developed, Arya’s point is that they do not automatically reduce the infrastructure required to build them. The economic prize is large enough that coordination is possible in theory and fragile in practice.
That is why he treats the safety-driven “slowdown” narrative as a sentiment catalyst rather than a durable capex pivot. In the same note, Arya reiterated a constructive stance on semis while explicitly separating near-term drawdown risk from the AI-demand question: “We remain bullish on semis, though we expect near-term volatility from rates, Mideast tensions, the Nov. 2 U.S. midterms, and growing resistance to data center expansion,” he said.
What I’m Watching Next: Volatility Catalysts That Could Override the “Noise” Call
The cleanest way for the “noise” thesis to break is if restraint rhetoric turns into enforceable limits. Follow-up statements or policy proposals from major frontier labs that move from general calls for caution to concrete training or deployment caps would matter more than another round of op-eds.
The other tripwire is the physical market. Arya’s argument leans on tightness signals like near-full utilization and rising rents for older accelerators. If rental pricing for older-generation chips rolls over, or if utilization is no longer described as near 100%, that would be the first evidence that the demand side is actually easing.
Macro and politics can still swamp the tape even if AI demand holds. Arya flagged rates, Middle East headlines, and the Nov. 2 U.S. midterms as near-term volatility sources, and he also called out local permitting and community resistance to data-center expansion as a practical constraint that can slow buildouts without any change in AI ambition.
How I'm Reading BofA dismisses AI slowdown fears. Semis
The part that decides whether this is “noise” is not the debate itself, it is whether the compute market loosens. If AI network capacity is still effectively pinned near full utilization and even older accelerators are getting more expensive to rent, then the equity sell-off is trading narrative risk against a still-tight physical constraint, and that tends to mean choppy repricing rather than a clean demand unwind.
The real test is whether restraint becomes coordination. If major labs and their buyers can translate safety language into binding limits that competitors actually follow, capex expectations can reset. If they cannot, the capex race remains the default, and the Monday-style drawdowns stay more about volatility regimes than a structural slowdown in AI infrastructure spending.
