Data-center, chip-manufacturing, model-capacity and physical-automation deals show AI expansion forming an interconnected infrastructure chain.
Artificial Intelligence··Morning
The scale of compute commitments
TechCrunch reports that OpenAI's infrastructure and compute commitments through 2030 have reached $750 billion, 25% above estimates from early 2026. That total includes Project Camellia, a planned $20 billion, 1,400-acre campus in Georgia. More than 3.2 gigawatts of power have been contracted for the site, with capacity expected to come online between 2028 and 2032. Supply is set to rely mainly on natural gas, supplemented by batteries and solar, while OpenAI has agreed to reduce its draw by as much as 1 gigawatt during grid peaks.[1]
The AMD-Anthropic arrangement reported by CNBC combines capacity sales to a model company with an equity commitment from the supplier. AMD plans to sell Anthropic tens of billions of dollars in AI servers and invest up to $5 billion in the company. Anthropic, in return, is to deploy 2 gigawatts of Instinct MI450 GPU capacity beginning in the first half of 2027. The investment depends on deployment milestones, so the announced ceiling is neither an unconditional amount nor money already transferred in one step.[2]
Manufacturing and physical automation
On the manufacturing side of the same compute chain, TSMC is expanding its U.S. investment to $200 billion amid political pressure. CNBC says American fabs cost 20% to 50% more to operate than facilities in Taiwan, leading the company to expect 2% to 4% gross-margin dilution over the coming years and to pass some higher costs to customers. At the same time, second-quarter revenue reached a record, gross margin was 67.7%, and profit grew 77.4%, with the report identifying AI-chip demand as a growth driver.[3]
The Atoms round covered by TechCrunch shows capital also moving beyond data centers into physical automation. The holding company that joins Travis Kalanick's CloudKitchens with heavy-industry automation unit Pronto raised $1.7 billion in a round led by a16z, with Bain Capital, Fifth Wall and Uber participating. It aims to expand automation of physical work through AI and robotics and is considering a move into mining. No valuation was disclosed for Atoms, so the size of the financing cannot be treated as the company's market value.[4]
One chain, different financial measures
Read together, the OpenAI, Anthropic-AMD and TSMC records place capital commitments across layers extending from electricity connections to server and GPU deployment and then advanced-chip manufacturing. Yet a $750 billion multiyear spending commitment, an investment of up to $5 billion tied to milestones, tens of billions in server sales, and a $200 billion manufacturing program are not equivalent accounting categories. Their timelines and conditions differ, so the figures cannot simply be added into a single daily investment total.[1], [2], [3]
Atoms adds an application layer to this picture: while the other three records describe building compute capacity and chip supply, Atoms is raising capital to automate work in the physical world with robotics and AI. Together, the four reports support a view of AI investment distributed across energy, hardware, manufacturing and automation rather than concentrated in one model launch. They do not, however, verify through a common method that every project will finish on schedule, every announced ceiling will be reached, or local capacity additions will produce a particular economic return.[1], [2], [3], [4]
