Gelsinger Says AI Needs 10,000x Efficiency Gain
Former Intel CEO Pat Gelsinger told a Scottsdale summit that incremental chip gains cannot meet AI's computing demands.
Updated on Nov. 21, 2026 in Semiconductors

Live Poll
Do you think the U.S. is headed in the right direction on energy for AI?
Former Intel CEO Pat Gelsinger argued at the COSM Technology Summit in Scottsdale, Arizona, that AI requires a 10,000-fold improvement in computing efficiency. He later wrote on X, "Is Moore's Law dead? I say it's time to wake it up!"
Why it matters
Gelsinger contends that incremental chip improvements cannot satisfy AI and compute demands, pointing to strain on U.S. energy capacity that has grown only in the low single digits over the past decade. The argument frames power supply and computing architecture as limits on the next phase of AI development.
Gelsinger cited NVIDIA Kyber racks using the Blackwell NVL72 architecture at 600kW, enough to power about 500 homes, while new gas turbines face eight-year supply backlogs and nuclear projects take a decade to build. How quickly these power constraints can be addressed remains unclear from the summit remarks.
The players
Pat Gelsinger
Former Intel CEO and current general partner at Playground Global, who has backed U.S. lithography and nuclear energy ventures.
Playground Global
Venture firm that Gelsinger joined as a general partner in March 2025, backing deep-tech companies.
The details
During a fireside chat titled "Reality Check: Beyond Chips" with George Gilder and Bret Swanson, Gelsinger pointed to optical interconnects and silicon photonics as a path to move data with light rather than copper. He also noted that major AI foundation models ship roughly every four weeks, and a Harvard study found U.S. datacenters consume as much electricity as Chile.
Timeline
March 2025: Gelsinger joined venture firm Playground Global as a general partner.
2026: Gelsinger spoke at the COSM Technology Summit fireside chat in Scottsdale, Arizona.
The Tech Race
Gelsinger's call to "wake up" Moore's Law places him against the view that the decades-long doubling of transistor density has run its course. The argument reflects a broader competition over whether gains will come from chip scaling, photonics, or new energy infrastructure.
Power limits on AI datacenters could affect the cost and availability of cloud computing and AI services used by consumers and businesses. Gelsinger's argument also points to optical interconnects as a possible future shift in how chips move data.
The takeaway
Gelsinger's remarks suggest the next gains in AI may depend as much on energy supply and data movement as on transistor scaling. Readers following chip policy should watch how power and interconnect constraints are addressed.
Further reading
For broader context on the chip industry's next steps, see Semiconductors coverage.
Source note: This article includes information reported by Wccftech.
Live Poll
Do you think the U.S. is headed in the right direction on energy for AI?






