The AI Spending Spree Comes With a Catch
Спикеры
Sam Palmisano
— Former CEO, IBM
Former IBM CEO Sam Palmisano discusses the AI spending boom and its catch. He says data centers, chips and energy are long-cycle bets, so large hyperscalers will control scarce capacity while smaller players get squeezed. He warns that AI infrastructure commitments may be more variable than they look if adoption slows, and that cheaper Chinese open-source models could constrain returns on US GPU-intensive AI.
- Sam Palmisano says data centers take 3-5 years, semiconductors about 7 years, and nuclear at least 10 years to come online.
- Large hyperscalers making near-term bets are likely to control AI data center and energy capacity.
- Announced AI infrastructure deals are mostly MOUs and may become variable commitments if adoption slows.
- Off-balance-sheet AI financing makes true debt and liability hard for analysts to assess.
- Historical infrastructure buildouts tended to correct after an installation phase.
- Chinese open-source models such as DeepSeek require less GPU capacity and can target non-US markets.
Идеи
Hyperscalers control AI capacity; others squeezed.
Because data centers, semiconductors and nuclear energy capacity take years to build, the large hyperscalers making near-term bets will lock up scarce AI infrastructure and control the market, while smaller competitors and would-be participants are likely to be squeezed out.
AI infrastructure buildout vulnerable to adoption slowdown.
The announced AI infrastructure buildout is largely MOUs and commitments that may be variable; even a modest slowdown in AI adoption would stretch the assumed adoption curve and cause data center and energy capacity buildout to be delayed or cut, leaving developers and bondholders to absorb shortages and forcing a revision of today's aggressive AI infrastructure plans.
Chinese open-source AI takes global share.
Chinese open-source AI models such as DeepSeek are a lower-cost alternative that needs less GPU capacity than US proprietary models; China can deploy them at massive scale and low cost to capture the roughly 75% of the global market outside the US, taking share from US AI models and constraining returns on US GPU-intensive AI capacity.
Chinese open-source AI takes global share.
Chinese open-source AI models such as DeepSeek are a lower-cost alternative that needs less GPU capacity than US proprietary models; China can deploy them at massive scale and low cost to capture the roughly 75% of the global market outside the US, taking share from US AI models and constraining returns on US GPU-intensive AI capacity.
Далее
This Bloomberg Markets video, published August 22, 2026, features Sam Palmisano discussing SKYY, AI data center and energy buildout, FXI, US AI GPU capacity. 4 trade ideas extracted by AI with direction and confidence scoring.
Speakers: Sam Palmisano · Tickers: SKYY, AI data center and energy buildout, FXI, US AI GPU capacity