Micron, Nvidia, AMD stocks at risk as Anthropic, OpenAI leads push to slow AI growth

Anthropic CEO Dario Amodei urged a slowdown in AI model development, raising concerns for AI-linked stocks including Nvidia, Micron, AMD, SanDisk and Western Digital. Nvidia ended the week at $218, down 6.8% from its month-to-date high of $234, reflecting a more cautious investor stance toward the AI trade.

Analysis

The immediate transmission channel is valuation rather than orders: NVDA carries the greatest exposure to any reduction in the assumed duration of frontier-model training spend, while MU’s HBM revenue is buffered by qualification cycles and pre-committed supply. A broad AI-capex de-rating would nevertheless pressure AMD disproportionately because its accelerator share-gain thesis requires customers to diversify incremental clusters, not merely maintain existing deployments. The market should distinguish a change in model-release cadence from a reduction in hyperscaler infrastructure budgets; the former may defer demand but does not necessarily eliminate power, networking, memory, and inference requirements.

Over the next 1-3 months, the relevant catalyst is whether hyperscalers alter 2027 capex commentary or whether enterprise AI workloads show weaker token/inference growth. NVDA is vulnerable to multiple compression if investors lower terminal growth assumptions, even absent a near-term revenue-guide cut; AVGO, MRVL, ANET and TSM would be cleaner confirmation vehicles if the concern becomes an actual infrastructure-spend revision. Over 6-18 months, a slower training race could shift spend from compute-intensive frontier training toward storage-heavy data retention and inference optimization, relatively improving the mix outlook for WDC/SNDK versus GPU-centric names.

The contrarian case is that safety-driven pauses by leading labs raise barriers to entry: compliance, evaluation, and secure deployment costs favor scaled cloud providers and established silicon ecosystems. That outcome could consolidate demand into the largest buyers rather than reduce it. The thesis is falsified by unchanged hyperscaler capex guidance, continued HBM tightness, and no deterioration in NVDA/AMD backlog or lead-time commentary during the next earnings cycle.

Key Decisions for Investors

• Use the current uncertainty to express a 1-3 month relative-value trade: short NVDA versus long MU in equal dollar amounts. NVDA has greater long-duration multiple risk, while MU has nearer-term memory-cycle support; exit if NVDA materially outperforms MU by 10% after the next major hyperscaler capex update or if MU guides HBM demand below prior expectations.
• For downside protection rather than an outright bearish call, buy NVDA 2-3 month put spreads approximately 5%/15% below spot, funded only partially by selling an upside call spread. This targets a sentiment-led de-rating while limiting loss if infrastructure spending remains intact; reassess after the next NVDA earnings and cloud-capex reports.
• Do not add directional shorts in WDC or SNDK solely on this signal. Instead, monitor NAND pricing, cloud storage demand, and management commentary for evidence that AI data-retention/inference workloads are offsetting any training slowdown; a confirmed mix shift would support a long WDC/SNDK basket versus AMD over 6-12 months.
• Set an alert for any downward revision to Microsoft, Alphabet, Amazon, or Meta capex plans. A coordinated reduction would validate a broader semiconductor-spend unwind and justify extending the short basket to AVGO, MRVL and ANET; absent that evidence, treat the move as positioning volatility rather than a fundamentals break.