Pat Gelsinger told AI infrastructure buyers at Ai4 2026 that today's GPUs fall short. The former Intel CEO said they use power inefficiently and face computational limits. He spoke with Sachin Katti, OpenAI's head of compute. Gelsinger argued buying more accelerators will not fix AI's real problems. Power, memory, networking, and economics still fail. He said the cost of producing intelligence needs a 10,000 times improvement. That will come from advances across semiconductors, memory, packaging, networking, and data center architecture all moving together. Katti highlighted bottlenecks in data center construction, energy supply, and manufacturing capacity. Both speakers said future AI growth depends as much on electricity as on better chips. Gelsinger called chips the fuel of a token-driven economy where demand for intelligence is unlimited. The constraint is on the supply side. Silicon vendors stand to gain most from rebuilding the full compute stack.
Pat Gelsinger told AI infrastructure buyers at Ai4 2026 that today's GPUs fall short. The former Intel CEO said they use power inefficiently and face computational limits. He spoke with Sachin Katti, OpenAI's head of compute. Gelsinger argued buying more accelerators will not fix AI's real problems. Power, memory, networking, and economics still fail. He said the cost of producing intelligence needs a 10,000 times improvement. That will come from advances across semiconductors, memory, packaging, networking, and data center architecture all moving together. Katti highlighted bottlenecks in data center construction, energy supply, and manufacturing capacity. Both speakers said future AI growth depends as much on electricity as on better chips. Gelsinger called chips the fuel of a token-driven economy where demand for intelligence is unlimited. The constraint is on the supply side. Silicon vendors stand to gain most from rebuilding the full compute stack.