Volatility on Wall Street, Not in the Data Center
Chip and AI infrastructure stocks have had a rocky few weeks, but if you ask the people actually running AI compute businesses, the underlying demand story hasn't changed at all. According to a recent CNBC report, a chorus of AI executives is pushing back on the narrative that softening stock prices signal softening demand — insisting instead that the real constraint isn't customer appetite, it's supply.
"Nearly Unlimited" Demand, According to Industry Veterans
Few voices carry more weight on this topic than Pat Gelsinger, the former Intel CEO now serving as a general partner at Playground Global. Gelsinger characterized AI demand as nearly unlimited, arguing that the real bottleneck isn't chips or capital — it's energy availability. That's a notable reframing: the conversation around AI scaling is shifting from "can we build enough compute" to "can we power enough compute."
What Sparked the Sell-Off?
The recent volatility traces back to a few high-profile moves:
- Meta announced plans to sell off excess AI computing capacity.
- Elon Musk's xAI is renting out its own excess capacity.
- Samsung forecast a sharp rise in profit, yet its shares still dropped — a reaction to a stock that had already rallied more than 360% over the past year.
Taken together, these headlines were enough to spook investors into reading "capacity for rent" as "demand is cooling." But several executives on the ground say that read misses what's actually happening.
Demand Is Outstripping What Companies Can Even Build
Marc Boroditsky, Chief Revenue Officer at Nebius, said demand is exceeding what his company can currently fulfill, even as it races to build out new data centers running on Nvidia GPUs. That's echoed by Andrew Feldman, CEO of Cerebras Systems, who called the Meta and xAI capacity moves unusual rather than representative — and maintained that industrywide compute demand still far outstrips available supply.
Sungyun Park, CEO of Rebellions, added that AI infrastructure momentum remains strong, and pushed back directly on the idea that these capacity sales are evidence of hyperscalers overbuilding.
Sold Out Five Years Out
Perhaps the most striking data point came from Michael Hurlston, CEO of Lumentum, whose company makes photonics and optical components critical for data center connectivity. Hurlston said Lumentum's products are sold out for the next five years — a backlog that has helped drive the stock up roughly 600% over the past 12 months. That's not the profile of a market bracing for a slowdown.
The Real Shift: From Spending to ROI
None of this means enterprise AI spending is immune to scrutiny. Boroditsky pointed to a meaningful shift in how businesses approach AI budgets — moving away from heavy, unrestrained usage toward more ROI-focused "rationalization." Companies are increasingly weighing the cost of expensive frontier models against capable open-source alternatives like DeepSeek and Alibaba's models, looking for the best return rather than simply the most powerful tool available.
The Takeaway
The picture emerging from these executives is nuanced: raw infrastructure demand remains extremely strong — arguably supply-constrained rather than demand-constrained — even as individual companies get smarter and more selective about how they spend on AI. Stock volatility may reflect shifting investor sentiment and one-off corporate decisions, but on the ground, the people building and selling the picks and shovels of the AI boom say the gold rush isn't over. It's just maturing.
Source: Reporting based on CNBC coverage, via Binance News.