AI news5
Musk vs OpenAI $130B trial opens, four weeks of testimony begin with "stole a charity" framing
The federal jury trial in Musk's lawsuit against OpenAI opened Tuesday in California with Musk taking the stand and his lawyer telling jurors Sam Altman "stole a charity." The suit seeks $130B in damages, ouster of Altman and Greg Brockman from the board, and unwinding of OpenAI's for-profit conversion. OpenAI's counsel framed the case as "sour grapes" from a co-founder who left and lost; Microsoft's lawyers separately said they "knew nothing" of the 2023 Altman firing. Hundreds of pages of internal emails are set to enter the public record across the four-week trial.
therundown.ai ↗Why it matters
A verdict against OpenAI's restructuring would set US precedent for how nonprofit assets can convert to for-profit equity, and would directly shape every charity considering an AI commercial spin-off.
Anthropic crosses $1T valuation and overtakes OpenAI as the most valuable AI company
Anthropic's latest funding round closed at a $1 trillion valuation on Tuesday, the same day OpenAI reportedly missed its Q1 revenue and user targets and watched Oracle and chip stocks fall on the news. The Atlantic published a piece literally titled "Anthropic's Little Brother." Anthropic also paired the milestone with shipping Claude for Creative Work, native connectors for Blender, Adobe Creative Cloud, Autodesk Fusion, Ableton, Splice, SketchUp, Resolume, and Canva.
theneurondaily.com ↗Why it matters
The "smartest model" race has stalled into a feature-parity loop, and the competition has shifted to whichever lab is deepest inside the tools paying customers already use. Anthropic's nonprofit-friendly track record now sits behind the most valuable AI company in the world.
OpenAI lands on AWS Bedrock with GPT-5.5, Codex, and Managed Agents one day after the Microsoft exclusivity reset
OpenAI announced that GPT-5.5, Codex, and Managed Agents are now available through Amazon Bedrock, the first concrete proof point of the new non-exclusive Microsoft contract. The deployment lands inside the same AWS console enterprise customers already use for Anthropic and Mistral. AWS gets parity selection, OpenAI gets distribution outside Azure for the first time, and Microsoft keeps revenue share through 2030 and licensing through 2032.
openai.com ↗Why it matters
For organisations standardised on AWS, the decision tree for which frontier model to use just collapsed into one console, and procurement, security, and billing barriers between OpenAI and Anthropic effectively disappear.
NVIDIA open-sources Nemotron 3 Nano Omni, a vision-audio-text model running 9x faster than rival multimodal open models
NVIDIA released Nemotron 3 Nano Omni this week, a small open multimodal model that handles vision, audio, and text and reports 9x speed gains over comparable open multimodal models. The model targets edge and on-device workloads, weights are openly available, and pricing is unconstrained. Xiaomi simultaneously open-sourced MiMo-V2.5-Pro, a 1M-context model that ties Moonshot Kimi K2.6 on Artificial Analysis benchmarks for agentic tasks.
rundown.ai ↗Why it matters
A second wave of small, fast, openly-licensed multimodal models lowers the floor for any nonprofit running constituent-facing voice-or-image AI without an OpenAI or Anthropic budget.
Wired permit review: planned US gas-powered AI data centres would emit 129M tons of CO2 per year, more than Switzerland or Norway
A permit review of OpenAI, Meta, Microsoft, and xAI's planned gas-powered AI data centres projects 129+ million tons of greenhouse gases per year, roughly 1.9% of total US emissions and more than the entire national output of Norway or Switzerland. The figures arrive as Nvidia's own executives say the cost of AI compute has now passed the cost of employees at most companies. Reuters and The Register also flagged growing AI vendor lock-in: workflows, RAG layers, and compliance wrappers are making model migration prohibitively expensive even before the bills arrive.
wired.com ↗Why it matters
Climate-aligned and impact funders are about to have a much harder time defending unrestricted AI infrastructure grants, and procurement decisions in the social sector now carry an explicit emissions footprint.
AI in the nonprofit sector4
Chan Zuckerberg Biohub commits $500M to a Virtual Biology Initiative building open AI cell models for global research
Biohub, the nonprofit founded by Mark Zuckerberg and Priscilla Chan, committed $500M on Tuesday to the Virtual Biology Initiative, a multi-year programme to build the open data foundation needed for AI to simulate human cells. Biohub describes the work as the first large-scale scientific initiative combining frontier AI with frontier biology to cure or prevent disease, with imaging as a $400M sub-focus. All cell models generated will be released free to the global scientific community via biohub.org. Time Magazine ran the announcement under the headline "If AI Can Model Cells, Science Can Deliver Cures."
prnewswire.com ↗Why it matters
Open public AI biology infrastructure at this scale changes the cost base for every health, disease, and ageing nonprofit running predictive science, and sets a precedent that Big-Tech-aligned philanthropy can fund pre-competitive datasets rather than proprietary tools.
Devex: nonprofits are not failing at AI, the way they are funded is
Devex published a sharp commentary this week arguing that the nonprofit AI adoption gap is mostly a funding-design problem, not a capability problem. Two-thirds of nonprofits experienced funding cuts in 2025 while service demand rose, but most AI funding still goes to tools rather than to staff time, integration, governance, or evaluation. The piece sits alongside a parallel Devex opinion that "most AI funding for nonprofits goes to the tools, almost none goes to what makes those tools actually work."
devex.com ↗Why it matters
For sector funders this is the clearest articulation yet that restricted tool-only AI grants will continue to underperform the headline 92%-adoption / 7%-impact split, and it lands as the AI for Humanity Round 2 survey opens for the second annual cycle.
NonprofitPro: AI adoption divide is widening between nonprofits with integration capacity and those running pilots without it
NonprofitPro published an analysis this week arguing the social sector is splitting into two AI cohorts: organisations that have invested in data infrastructure, governance, and staff training, and a much larger group running disconnected AI pilots without backbone support. The piece flags the same 92%-adoption / 7%-impact pattern from the Virtuous + Fundraising.AI benchmark, framed as a structural divide rather than a maturity curve. Suggested fixes centre on integration with CRM and case-management systems, repeating the constraint TFSF Ventures highlighted in its newly published six-layer deployment methodology for nonprofits this week.
nonprofitpro.com ↗Why it matters
The "AI haves and have-nots" framing gives boards and funders a sharper diagnostic than the headline adoption stat alone, and it points capacity-building dollars at integration and governance rather than tool licences.
Vantage Data Centers, Oracle, and OpenAI commit $225,000 to relaunch the Port Washington-Saukville Education Foundation
Vantage Data Centers announced on 27 April that it is partnering with Oracle and OpenAI to relaunch the Port Washington-Saukville Education Foundation in Wisconsin with a $225,000 commitment to support student programs in the local school district. The foundation had become dormant before the new infrastructure tenancy, and the trio is funding student-facing AI literacy and STEM programming alongside it. Vantage frames the relaunch as the community-investment leg of its data-centre build in PWSSD.
vantage-dc.com ↗Why it matters
This is one of the cleanest examples yet of a hyperscaler-adjacent nonprofit lifecycle, where the data-centre tenant funds local education infrastructure as part of the deal, and it is replicable in every town hosting frontier compute.