Semiconductor stocks tumble on AI slowdown calls, but capex outlook stays strong
In brief
- Semiconductor stocks fell 3.4–6% on September 14 after AI executives called for development caution
- NVIDIA, AMD, Broadcom, and Intel posted declines; PHLX semiconductor index slid 5.9%
- Bank of America projects $795 billion combined 2026 capex from major cloud firms, rising to $1.08 trillion in 2027
- Analysts: slower AI iteration extends hardware upgrade cycles rather than reducing demand
The trigger: coordinated caution from AI labs
Executives at Anthropic, OpenAI, and xAI coordinated remarks pointing to safety concerns around the pace of AI advancement. The market read this as a signal that demand for new chips might soften—a reflexive sell-off in a sector that's bet heavily on uninterrupted acceleration.
The reaction was visceral. Traders heard "less compute needed" and acted. Yet the underlying math tells a different story.
Why the slowdown may not crater demand
A slowdown in training new frontier models would primarily affect the R&D budgets of AI developers, not the infrastructure orders already locked in with chipmakers. That distinction matters. Broadcom CEO Hock Tan confirmed on September 14 that the company's AI revenue targets remain unchanged, signaling confidence that capex commitments won't crater.
The data supports this view. NVIDIA reported data-center revenue of $89 billion for the quarter ending in August 2026, a 117% year-over-year increase. Its guidance calls for roughly 70% revenue growth in fiscal 2028. These aren't the numbers of a company bracing for a demand cliff.
The real floor: hyperscaler spending
BofA Global Research projects that combined 2026 capital spending from Microsoft, Alphabet, Amazon, Meta, and Oracle will reach approximately $795 billion, climbing to nearly $1.08 trillion in 2027. That's the trajectory the market needs to watch. Analysts flagged on September 15 that if AI development genuinely slows down, existing hardware remains useful longer, extending the upgrade cycle. Slower iteration doesn't kill demand—it spreads it out, with companies buying chips to scale horizontally as volume grows.
The key metric to monitor is what hyperscalers actually commit to spending. As long as the capital expenditure trajectory climbs toward $1.08 trillion in 2027, the chip supply chain has a floor. Sentiment swings are noise. Capex commitments are signal.
