Back to News
TerraPower’s nuclear reactor has a secret weapon for powering AI data centers
BusinessWednesday, August 19, 2026 at 04:30 PM

TerraPower’s nuclear reactor has a secret weapon for powering AI data centers

Curated Summary

TerraPower, the nuclear startup founded by Bill Gates, is set to announce its first dedicated data center project this year, with construction expected to break ground in 2027.

The company’s 345-megawatt molten salt-cooled reactor offers a distinct advantage over traditional nuclear plants by utilizing energy storage to handle the rapid load fluctuations of AI workloads.

While standard nuclear reactors struggle to ramp up or down quickly, TerraPower’s design allows it to complement intermittent sources and smooth out the power demands of GPU-intensive training and inference.

This approach addresses a critical bottleneck for AI infrastructure, where natural gas turbines are failing under stress and battery banks are adding significant costs.

Meta has already agreed to purchase eight of TerraPower’s Natrium power plants, signaling strong industry confidence in this solution.

By leveraging its ability to operate at peak capacity while managing variable loads, TerraPower positions itself as a key player in powering the next generation of AI data centers.

Want the full story?

Read Original Article

via TechCrunch AI

#nuclear-energy#data-centers#terraform-power#ai-infrastructure#meta

Related Articles

Starcloud raises $250 million for orbital data centers as launch options dry up
TechCrunch AI
BusinessAug 21

Starcloud raises $250 million for orbital data centers as launch options dry up

Starcloud, a startup building satellites for in-orbit AI inference, has raised a $250 million extension to its Series A, bringing the round to $420 million and valuing the company at $2.3 billion. The new funding, led by Manhattan West Ventures with participation from Nvidia ($25 million) and Cisco, will support a larger manufacturing facility and the development of its Starcloud-3 spacecraft, designed for SpaceX's Starship. CEO Philip Johnston is also securing launch capacity amid a tightening rocket market, as Falcon 9 is scheduled to end by 2028 and alternatives like New Glenn and Vulcan are not yet flying regularly. Starcloud plans to launch two Starcloud-2 compute satellites on rideshare missions in 2027 for customers including US government agencies. The company has requested FCC permission for 88,000 spacecraft, betting on Starship to lower launch costs and make orbital data centers competitive with terrestrial ones. The funding highlights growing investor interest in space-based AI infrastructure, even as launch logistics remain a major bottleneck.

1 min readRead
When AI designs a drug, who gets the credit?
MIT Technology Review
BusinessAug 21

When AI designs a drug, who gets the credit?

The article explores the legal and ethical questions around AI-generated drug inventions, focusing on a patent dispute involving Insilico Medicine's AI-discovered molecule for pulmonary fibrosis. It highlights that US patent law currently recognizes only humans as inventors, as established in the DABUS case where an AI was denied inventor status. The piece quotes patent attorney Sarah Korman and discusses how laws will likely need to evolve as AI becomes more autonomous in discovery. Ryan Abbott, who brought the DABUS test case, warns that misattributed inventors could invalidate AI-related patents. Overall, the story underscores the growing tension between technological capability and existing intellectual-property frameworks.

1 min readRead
AI data startup Micro1 reaches $500M gross run rate amid AI training boom
TechCrunch AI
BusinessAug 21

AI data startup Micro1 reaches $500M gross run rate amid AI training boom

Micro1, a four-year-old AI data labeling startup, has surged to a $500M gross annual run rate in just eight months, up from $100M, with net run rate between $150M and $200M. The growth reflects booming demand for unique AI training data from top labs, though Micro1 trails rivals Mercor and Handshake. The company is expanding margins by generating synthetic data and selling off-the-shelf datasets to multiple customers, reaching gross margins of 80-90%. Founder Ali Ansari publicly stated Micro1 does not sell data to Chinese model makers, addressing controversy over off-the-shelf data exports. Micro1 pivoted from AI recruiting to data labeling and is also building robotics pre-training datasets. The startup may have recently raised a new round at a higher valuation than its $500M Series A last September.

1 min readRead