Global chip stocks slid sharply as investors balked at rich valuations built on AI euphoria, erasing roughly $500 billion in market value and putting the sector’s yearlong rally on notice. The selloff hit Asia first, where South Korea’s KOSPI fell as much as 6.2% with Samsung Electronics and SK Hynix leading declines, Japan’s Nikkei dropped 4.5%, and Taiwan Semiconductor Manufacturing Co. (TSM) slipped more than 3%. The pullback is stoking debate over whether the AI trade is cooling or merely resetting after a parabolic run.
Semiconductor valuations have sprinted ahead of fundamentals this year, with multiples expanding even as supply chains normalize and new capacity comes online. Nvidia (NVDA), the de facto proxy for AI spending, became a magnet for momentum, pulling up Advanced Micro Devices (AMD), Broadcom (AVGO), and the entire supply chain from memory to equipment. That dynamic works in reverse when positioning unwinds. Once investors began to question how quickly AI-driven revenue converts to earnings — and how long hyperscaler capex can run at current clips — the air pocket showed. The result: a swift derating across names that carried the most premium.
The overnight slide in Asia did the heavy lifting, sending a clear message that enthusiasm had outpaced near-term visibility. Korea’s decline was led by SK Hynix, the high-bandwidth memory leader whose fortunes are tied to AI server demand, and Samsung, which straddles memory and foundry. In Japan, chip-adjacent heavyweights and equipment names slumped in sympathy. TSMC’s drop echoed a cautious tone from customers tightening unit orders even as they prioritize AI allocations. When the largest fabricator blinks, downstream suppliers and device makers typically follow. Futures tied to U.S. chip ETFs, including VanEck Semiconductor ETF (SMH) and iShares Semiconductor ETF (SOXX), pointed lower as traders priced in a catch-down move at the open.
Institutional leaders have started to telegraph a higher tolerance for a shakeout. “We should welcome the possibility that there would be drawdowns, 10 percent to 15 percent, that are not driven by some sort of macro cliff effect,” Morgan Stanley CEO Ted Pick said, framing the slide as a healthy deflation of excess. Caution is spreading beyond talk. Hedge fund manager Michael Burry disclosed a sizable short against Palantir (PLTR), a high-profile AI software winner, and the stock fell 7.94%, a reminder that skepticism is no longer taboo. With gross leverage elevated in tech and factor crowding intense, de-risking can feed on itself as funds trim winners to protect YTD gains.
Not everyone sees a fundamental break. “The selloff appears to be largely positioning-driven, with recent outperforming names taking the worst of the move,” said Jon Withaar of Pictet Asset Management. That view syncs with order books at leading suppliers that still show tight allocations for AI accelerators, robust backlogs for next-generation servers, and persistent demand for high-bandwidth memory. Even so, the market is revisiting a key question: how broad and durable is the monetization of AI workloads beyond a handful of megacap buyers? If the answer is narrower than hoped in the near term, multiples need to reflect it, even if the structural thesis is intact.
Two forces are in play. First, the cost of capital. Higher real yields reduce what investors will pay for long-dated growth, and semiconductors are among the most duration-sensitive equities. Second, supply elasticity. TSMC, Samsung, Intel (INTC), and memory leaders have leaned into capacity, backed by demand forecasts shaped by hyperscaler roadmaps. That can compress pricing power if deployment timelines slip or if customers prioritize utilization over unit growth. Export controls and shifting geopolitics remain a wild card for select names with China exposure. Together, these risks warrant lower peak multiples than what the market assigned during the summer melt-up.
If this derating persists, the next leg likely hits capital equipment and specialty suppliers that benefited from aggressive capacity plans. ASML (ASML), Applied Materials (AMAT), Lam Research (LRCX), and KLA (KLAC) rode the wave of next-node and HBM expansions; they can feel the pinch fastest if customers nudge deliveries or stretch lead times. Memory is another valve. SK Hynix and Micron (MU) have enjoyed a sharp price recovery tied to AI servers, but broader DRAM and NAND demand still needs a synchronized handset and PC rebound to sustain margins. In the U.S., watch second-tier enablers in packaging, substrates, and power management, which can be early indicators of order moderation.
The core bull case remains that AI is a capex supercycle, not a one-off upgrade. That’s still credible. But there is a gap between infrastructure buildout and revenue recognition in downstream software and services. Some buyers are shifting from pilot-phase to production at a slower clip than multiples imply. That lag matters for chipmakers if the next wave of orders skews toward efficiency upgrades rather than pure capacity adds. The market is adjusting to a more nuanced slope: still up and to the right, but with pauses where utilization and unit economics catch up to hype.
Three things. First, guidance clarity. Near-term order visibility from Nvidia, TSMC, and memory leaders can reset expectations without killing the narrative. Second, capex discipline. If management teams signal a willingness to phase spend based on utilization milestones, investors will pay up for durability, not just speed. Third, rates relief. Any cooling in inflation or softer growth prints that pull yields down would ease the duration squeeze on high-multiple tech. In that scenario, cash-generative compounders like Broadcom and Qualcomm (QCOM) could outperform high-beta names as leadership rotates within the sector rather than out of it.
This reset looks less like an end to the AI trade and more like a forced calibration after an aggressive chase. The sector needed proof that earnings could grow into valuations set by a single narrative. Today’s selloff — with Asia setting the tone and high flyers taking the brunt — brings the conversation back to fundamentals: unit demand, pricing power, capacity discipline, and the cost of capital. If the next few earnings cycles deliver steady order books and measured spend, chips can re-rate at healthier levels. If not, the market will keep testing the edges of the AI story until the numbers do the talking.
