Power, Hardware, and Materials Innovation Race to Scale AI Infrastructure

인프라/플랫폼 | Wed Jul 22 2026 00:00:00 GMT+0000 (Coordinated Universal Time) | 4 sources

Data center power demand is surging while Nvidia Vera Rubin, next-generation lithography, and other AI infrastructure innovations advance across the board.

Analysis

[BloombergNEF] projected data center power consumption to quadruple by 2035 ^[1]

• Data centers expected to consume 20% of total U.S. electricity
• Capacity to expand by 200GW over the next decade
• AI training and inference to account for half
• 34% of PJM grid allocated to data centers

[Nvidia] unveiled Vera Rubin next-generation integrated CPU-GPU system ^[4]

• Hybrid superchip with 1 CPU per 2 GPUs
• NVL72 rack system features liquid-cooled plug-and-play design
• 10x tokens per watt vs. Grace Blackwell
• OpenAI has already deployed one rack

[xLight / Pat Gelsinger] moved to reignite Moore's Law with next-generation lithography ^[3]

• Former Intel CEO Gelsinger joins Playground Capital as partner
• Joins xLight board to support new lithography technology development
• Secured U.S. government investment
• Targets performance beyond 13.5nm wavelength light sources

[Syensqo] advanced materials innovation for AI data centers and semiconductors ^[2]

• Developing polymers and elastomers with enhanced purity and plasma resistance
• Addressing high-voltage architectures and thermal management materials
• Applying EV cooling technology to data centers
• Evolving component materials including connectors
• capacitors
• and HDDs

Sources

• [1] Data centers expected to use 4x more electricity by 2035 - TechCrunch AI
• [2] Advancing next-gen AI with materials science innovation - MIT Technology Review AI
• [3] This Former Intel CEO Wants to Jumpstart Moore’s Law With Light - Wired AI
• [4] Nvidia Wants to Own Every Chip Inside AI Data Centers - Wired AI