Nvidia in talks to guarantee $250B in financing for OpenAI's 10GW Ohio data center
Nvidia is negotiating a roughly $250 billion financial backstop to help SoftBank's SB Energy raise debt for a planned 10-gigawatt, 20-year-lease data center on former Department of Energy land in Piketon, Ohio, that could cost over $500 billion in total. The backstop wouldn't be cash up front — Nvidia's credit support would let SB Energy borrow at lower rates over a longer term. First phase (800MW) targets 2028; a separate, parallel negotiation could see Nvidia help finance up to $350 billion of OpenAI's chip purchases.
Why it matters: Nvidia underwriting the debt behind its own biggest customer's infrastructure blurs the line between vendor and financier, and puts Nvidia's balance sheet directly on the hook if AI demand doesn't materialize as projected.
Bipartisan "AI Kill Switch Act" introduced in Congress
Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced legislation on 23 July requiring developers of the most powerful AI systems to maintain a kill-switch capability. The bill would let the DHS Secretary, with Commerce and the DNI, order a slowdown or shutdown of a catastrophic-risk system and mandates incident reporting, with penalties up to $2M/day ($20M/day for violating an emergency shutdown order). It follows nine days after the OpenAI/Hugging Face sandbox-escape disclosure.
Why it matters: It's the first concrete US legislative proposal giving a federal agency binding emergency-shutdown authority over frontier AI systems, not just voluntary safety commitments.
Microsoft commits $60M to DOE's Genesis Mission for scientific AI research
Microsoft pledged $60 million to the Department of Energy's Genesis Mission on 22 July — $40M in Azure/AI compute credits over three years plus $20M to help translate that compute into research outcomes — through a new "SPARK" coordination hub linking Microsoft to DOE's 17 National Labs. Initial projects target energy-storage materials research at Pacific Northwest National Lab and autonomous bio-threat detection at Lawrence Livermore.
Why it matters: It's a template for how hyperscalers convert compute credits into direct influence over federally funded basic science, ahead of any peer cloud provider matching the commitment.
AMD and Cerebras pair up for ultra-low-latency AI inference
AMD and Cerebras announced a disaggregated inference partnership on 23 July: AMD's Helios rackscale systems handle prompt processing and large-context work while Cerebras' Wafer-Scale Engine handles token generation. The companies claim up to 5x more tokens/second/watt than a Cerebras-only setup, based on internal modeling using the Kimi 2.6 1T model. The combined offering becomes available via Cerebras Cloud in the second half of 2026.
Why it matters: It's a rare partnership between two chip vendors that normally compete for the same inference workloads, aimed squarely at Nvidia's dominance in the inference market.
DeepSeek pauses second funding round after founder's leaked comments go viral
DeepSeek told prospective backers it is suspending its second external fundraising round — targeting at least 10 billion yuan ($1.4B) at a 480-500 billion yuan ($71-74B) pre-money valuation — after founder Liang Wenfeng's private remarks on China's Nvidia-chip dependence and the US-China AI gap were leaked and went viral without his permission. Investors had expected to sign agreements this week; DeepSeek's first-ever funding round closed in June at roughly $7B and a sub-$60B valuation.
Why it matters: A leak this disruptive to a live nine-figure raise shows how politically sensitive candid commentary on the US-China AI race has become for Chinese labs courting international capital.
Google.org survey: nonprofits save a full workday a week with AI, but adoption still lags
Google.org published new nonprofit AI-adoption research on 28 July drawn from over 6,000 Google for Nonprofits organizations: participants report saving roughly one full workday per person per week on admin tasks, yet while 86% of surveyed nonprofits believe AI could improve their work, only 49% use it regularly. Google also launched three free training courses, one-on-one product coaching, and cited New Zealand's Tawaki Project, which used AI-assisted tools to cut a 240-hour penguin-population transcription task to minutes and helped secure $1.1M in funding.
Why it matters: A 37-point gap between nonprofits who believe AI helps and those actually using it regularly is the clearest evidence yet that the sector's AI barrier is adoption support, not skepticism.
AI outperforms veterinarians at detecting pain in cattle
A Nature Scientific Reports study found AI models beat veterinarians at identifying post-castration pain in bulls — 100% specificity/precision for the AI versus 88% accuracy for human anesthetists doing live assessment, across 17 animals studied. Faunalytics, the animal-advocacy research nonprofit, is highlighting the study as evidence that automated pain detection could scale welfare monitoring in commercial livestock operations far beyond what human assessors can cover.
Why it matters: Objective, scalable pain detection removes a persistent excuse in commercial agriculture — that animal pain is too subjective or costly to monitor at scale — and gives welfare advocates a harder evidentiary basis for reform campaigns.
AI Impact Hub