Multiple AI Executives Push Back on Computing Power Glut Theory: Demand Is Limitless, Energy Is the Bottleneck
Claire Weston
Chip-stock volatility has fueled fears of an AI compute glut, but several infrastructure executives say the opposite is true: demand far outstrips capacity and the real constraint is energy — one supplier's orders are booked out five years, and corporate AI spending is shifting from 'maximize usage' to a more sustainable 'maximize value.'
What spooked the market?
Chip stocks swung sharply after Meta said it would sell spare AI compute and Musk's xAI announced plans to rent out excess capacity.
Together, the headlines fed a single fear: hyperscalers may have overbuilt.
Samsung previewed a big profit jump this week, yet its stock fell — investors questioned whether AI infrastructure spending can last.
This means → The market's core anxiety is not "AI doesn't work" but "too much money spent, and capacity may be surplus."
What do executives on the ground actually see?
Former Intel CEO and current Playground Global partner Pat Gelsinger told CNBC that AI demand is "almost infinite" and the real bottleneck is energy supply, not compute demand.
Marc Boroditsky, chief revenue officer at Nebius, which builds data centers with Nvidia GPUs, said: "Demand far exceeds our ability to fulfill it — that has been the norm for some time."
Andrew Feldman, CEO of Cerebras Systems, called the Meta and xAI sell-offs "isolated cases" and stressed that industry-wide compute demand far outpaces capacity, with gaps in data centers and upstream inputs alike.
What is the hardest evidence?
Michael Hurlston, CEO of photonic-interconnect supplier Lumentum — a company that makes components using light rather than electricity to move data between chips and data centers — revealed that orders are booked out five years.
Lumentum's stock has risen roughly 600% over the past year — capital markets voted with real money for undersupply.
Sungyun Park, CEO of Samsung- and SK Hynix-backed chip startup Rebellions, echoed the view: AI infrastructure momentum remains strong; Meta's and xAI's moves do not equal industry-wide overinvestment.
In plain terms = When the shovel-makers are sold out five years ahead, the gold rush is far from over.
How is corporate spending changing?
Many companies were in a "maximize usage" phase — encouraging staff to use frontier models from OpenAI and Anthropic regardless of cost.
As cost advantages of open-source alternatives from DeepSeek and Alibaba come under closer scrutiny, CFOs are demanding stricter returns on AI spending.
Boroditsky calls this new phase "maximize value": CFOs tightening budgets are really searching for value.
This means → Companies are not cutting AI budgets — they are learning to spend smarter.
Will open-source models replace frontier ones?
Feldman offered a different lens: as enterprise AI deployment matures, tasks of varying complexity will be matched to different tiers of models and compute.
Frontier and open-source models will be complementary, not substitutes — simple tasks run on cheaper models; complex reasoning still needs top-tier compute.
In plain terms = Not every package needs a chartered jet, but the jet's workload does not shrink because of that.
This reflects a structural shift in compute demand, not a contraction in total volume.
What should investors watch next?
Whether the rationalization of enterprise AI spending signals long-term demand health or a growth slowdown — both sides have a case right now.
This means → The next verification window is the capital-expenditure data that major cloud providers will report — the numbers will show whether money is still flowing in.
Short-term chip-stock swings reflect sentiment; the medium-term direction depends on real corporate procurement pace.
Content is for reference only, not financial advice.