Elon Musk confirmed on X on Saturday that a secretive SpaceX foundry near Bastrop, Texas, will cast blades and vanes for gas turbines, TechCrunch reported. Casting those parts is the bottleneck in turbine production, he wrote. Doing it in-house could bring new gas turbines online as much as 18 months sooner. GE Vernova, which builds the turbines, says AI infrastructure demand has taken up nearly all of its capacity through 2030. The International Energy Agency expects worldwide data center power consumption to about double by then. Due diligence analyst Corey Trinetti reported that SpaceX bought roughly 830 acres near its Bastrop Starlink factory between March and June.
Read at TechCrunch ↗
OpenAI holds warrants in SB Energy now valued at an estimated $5.5 billion, awarded as part of the arrangement that secured OpenAI as a data center tenant, The Wall Street Journal reported from draft IPO documents. The warrants were worth $3.6 billion when granted in January. OpenAI put another $500 million into the company earlier this year. After the offering, which could raise $5 billion to $7 billion as soon as next month, OpenAI is expected to hold a single digit stake. SB Energy, controlled by SoftBank, has nearly nine gigawatts of contracted computing capacity and no operating data centers. Its data center unit expects a contracted backlog above $400 billion. SB Energy flagged its heavy reliance on OpenAI as a risk to investors.
Read at Investing.com ↗
Nvidia's reported $13 billion acquisition of Hugging Face, the main repository where developers share open-weightOpen-weight modelAn AI model whose trained parameters are published so anyone can download, run and modify it, as opposed to one reachable only through the developer's API. The main policy tension is that open weights spread capability widely and cannot be recalled once released. AI models, remained unconfirmed on Aug. 28, TechCrunch reported. That is above the $12.9 billion figure reported by The Information in AIPD's August 27th edition. Those talks follow a $6 billion agreement with model builder Poolside that moves most of its staff to Nvidia. Two weeks earlier, Stripe bought OpenRouter, which routes open-weight models to businesses, for more than $7 billion. Nvidia already ships its own open-weight Nemotron models, without large uptake. Just 6% of companies use open-weight models, per spending data from Ramp, and inferenceInferenceRunning a trained AI model to answer a query, as distinct from training it. Inference is where most ongoing compute and energy demand now sits, and it is what chip export rules increasingly target alongside training. costs are pushing firms toward cheaper Chinese models from Moonshot, DeepSeek and Alibaba.
Read at TechCrunch ↗
Nvidia's physical AI line, the chips and software it sells to builders of robots, cars and robotaxis, generates about $10 billion in annual revenue, per The Wall Street Journal. That is a small slice of the $303 billion Nvidia took in over the four quarters ended in July. Chief Executive Jensen Huang said the line will grow tenfold within a decade. U.S. rules block the company's most powerful AI chips from China while leaving its robotics and automotive chips unrestricted. China is the world's largest industrial robot market. Nvidia pairs those chips with developer software and has released open-weightOpen-weight modelAn AI model whose trained parameters are published so anyone can download, run and modify it, as opposed to one reachable only through the developer's API. The main policy tension is that open weights spread capability widely and cannot be recalled once released. robot models named GR00T and Cosmos. Counterpoint Research analyst Brady Wang said China is central to Nvidia's physical AI road map.
Read at LiveMint ↗