By Wall Street’s AI panic trade is already facing its first test.
Nvidia, AMD and SanDisk stocks are trying to recover after Monday’s sharp sell-off, when calls from industry leaders to slow frontier-model development triggered a rush out of AI-linked hardware.
Nvidia fell 3.4%, while semiconductor shares broadly sold off as investors quickly translated “slower AI” into weaker data-centre demand.
That assumption is now being challenged.
AMD says demand still exceeds supply, while infrastructure investors argue that inference, enterprise adoption and competition between model developers could keep spending elevated even if frontier research slows.
Slower frontier AI does not necessarily mean fewer chips
Nvidia remains the clearest proxy for the AI infrastructure trade.
The bearish logic was straightforward: if OpenAI, Anthropic and other frontier developers slow model development, hyperscalers may need fewer accelerators and less data-centre capacity.
Hendi Susanto, portfolio manager at Gabelli Funds, challenged that assumption in comments reported by MarketWatch.
He argued that if leading developers slow down, “second and third tier players will seize the opening to catch up.”
Susanto also pointed to inference, enterprise AI and physical AI as demand engines that do not depend solely on ever-larger frontier models.
That matters for Nvidia because its opportunity is no longer limited to training the next generation of foundation models.
Inference, robotics and enterprise deployment can continue consuming compute even if the frontier race becomes more cautious.
Slower model progress and slower infrastructure consumption are not necessarily the same trade.
AMD is showing little evidence of a demand slowdown
AMD provides the clearest company-level reality check, as the company’s commentary offers little support for that fear.
Piper Sandler analyst David O’Connor reiterated an Overweight rating and $600 price target after speaking with AMD, describing business momentum as “full-steam ahead.”
The firm said demand remains well ahead of available supply across AMD’s portfolio, while CPU and GPU ramps remain on track.
That matters because Monday’s selloff effectively priced in the possibility that AI customers could delay infrastructure spending.
AMD also has another route to growth. Even if overall AI capital expenditure rises more slowly, customers are still trying to diversify accelerator suppliers and reduce dependence on one vendor.
That gives AMD room to gain market share without requiring the entire infrastructure market to grow at the breakneck pace investors previously assumed.
SanDisk shows where the rebound thesis could still break
SanDisk is the useful counterweight because not every semiconductor subsector depends on the same demand drivers.
AI has lifted enterprise SSD demand as data centres require more flash storage for training datasets, inference workloads and model updates.
GF Securities analyst Jeff Pu said that the firm maintains a “high-teen QoQ NAND price expectation for 3Q26,” but expects growth to soften to low single digits in the fourth quarter as weaker mobile demand and elevated PC and handset inventories weigh on pricing.
That makes SanDisk a reminder that the AI panic trade can be overdone while the memory cycle still cools independently.
AI storage demand can remain healthy without NAND prices rising at peak rates indefinitely.
A two-day rebound cannot settle the spending debate. The real evidence will come from hyperscaler capex guidance, accelerator order books and memory pricing over the next few quarters.
The post Nvidia, SanDisk, AMD stocks rebound: is AI panic trade already cracking appeared first on Invezz
