AI news7
Anthropic triples run-rate to $30B, signs 3.5GW compute deal
Multi-gigawatt pact with Google and Broadcom locks in 3.5GW of TPU capacity from 2027, almost entirely US-based. Run-rate tripled since January, $1M+ enterprise customers doubled to 1,000+. Pentagon labeled Anthropic a supply-chain risk, rattling 100+ clients, but growth ahead of OpenAI.
therundown.ai ↗Why it matters
Compute is the new moat. Anthropic is betting hard on US infrastructure to serve the Claude demand that already outstrips supply — explains recent rate-limit pain.
Anthropic launches Managed Agents public beta
Pick task, tools, guardrails — Anthropic handles the backend plumbing that used to take engineering teams months. Agents run solo for hours without dropping state, multi-agent coordination in preview. $0.08/hr per session on top of API cost. Notion, Rakuten, Asana, Sentry onboard. Rakuten rolled out agents across five departments in one week each. Claude Cowork now generally available.
rundown.ai ↗Why it matters
The "agent-as-a-platform" play. Anthropic wants to be where agents live, not just the model they call. Changes the build-vs-buy math for every org considering agent infra.
Meta Superintelligence Labs ships first model — Muse Spark
Multimodal reasoning AI with multi-agent mode. MSL's debut product after Zuck's mega-hiring spree.
therundown.ai ↗Why it matters
Proof the lab is producing, not just recruiting. Opens up a fourth frontier model family competing directly with OpenAI/Anthropic/Google.
Oxford AI flags heart failure 5 years early from routine CT scans
86% accuracy across 72,000 patients. Reads invisible fat-texture changes around the heart. Highest-risk bucket: 1 in 4 develop heart failure within 5 years — 20x the low-risk group. NHS rollout underway, expanding to all chest CTs within months.
rdm.ox.ac.uk ↗Why it matters
AI is starting to ride on imaging infrastructure that already exists. No new scans, no new hardware — just smarter reading of what hospitals already collect.
Poke raises $10M at $300M valuation for personal-life AI agents
Text an agent to manage schedule, health, fitness, photos. Funded by notable investors; launching via SMS-first UX.
techcrunch.com ↗Why it matters
First venture-backed serious bet on personal-life agents (as opposed to work agents). Signals the category is moving from novelty to funded infrastructure.
Perplexity Computer hits $450M ARR, adds Plaid for personal finance
Connects 12,000+ financial institutions. Build custom tools for budgeting, spending, investment tracking. ARR up 50% month-over-month.
perplexity.ai ↗Why it matters
Perplexity is the clearest winner in "browser as agent surface." The product moat is turning into real recurring revenue, not just hype.
NASA uses AI to find 800 cosmic anomalies in Hubble archive
Scanned 100M images, surfaced previously undocumented phenomena. University of Michigan separately built brain-MRI reading AI at 97.5% accuracy in seconds.
science.nasa.gov ↗Why it matters
The unsexy AI wins — searching existing archives, reading existing scans — are the ones delivering real scientific leverage right now.
AI in the nonprofit sector6
Humanity AI: $500M five-year coalition launches
Ten foundations pooled: Ford, MacArthur, Mellon, Mozilla, Omidyar Network, Packard, Doris Duke, Lumina, Kapor, Siegel Family Endowment. Mission: ensure AI is shaped by and for people. Grantmaking from the pooled fund begins in 2026.
mellon.org ↗Why it matters
Largest philanthropic AI bet ever assembled. Changes who gets to shape "AI for good" narratives — pulls it out of Big Tech's hands.
OpenAI Foundation pledges $1B in grants over next year
Nonprofit arm of OpenAI commits to $1B of grantmaking and scaling philanthropic capacity. First major deployment of the nonprofit's funding firepower.
fortune.com ↗Why it matters
OpenAI's own nonprofit is now one of the largest AI-focused funders in the world. Could reshape which orgs get resourced to work on AI risk and access.
Stop Soldier Suicide's "Black Box Project" uses AI on device data
Families of suicide victims loan phones and laptops. AI plus forensic tools surface patterns — financial instability, disrupted sleep, digital withdrawal — that precede suicide. Goal: predictive risk modelling that respects privacy and consent.
philanthropy.com ↗Why it matters
A concrete, mission-bound AI application on one of the hardest problems in mental health. Not just an admin tool — actual clinical-grade pattern recognition.
International Rescue Committee crisis-response AI backed by $400K McGovern grant
IRC building AI that provides resource lists and tailored information to displaced populations — multilingual, context-aware, works offline where needed.
mcgovern.org ↗Why it matters
AI is getting to the edge where humanitarian response actually happens, not just the HQ spreadsheet. McGovern Foundation continues to be the most operationally-serious funder in this space.
Samaritas connects Detroit caseworkers with AI co-pilots
Michigan human-services nonprofit pairs tech startups with social-service staff. AI transcribes difficult family conversations, provides real-time prompts during court hearings and high-pressure settings. Goal: reduce cognitive load on frontline staff.
philanthropy.com ↗Why it matters
Frontline-worker AI is the most under-reported use case. Not replacing caseworkers — giving them real-time support in moments that matter most.
Virtuous 2026 Nonprofit AI Adoption Report: 92% adoption, 7% major impact
346 nonprofits surveyed. Near-universal AI use, but only 7% report major organisational capability improvement. Most orgs: "one person using ChatGPT to draft an appeal, rest of team buried in manual processes and disconnected systems."
virtuous.org ↗Why it matters
Adoption is not transformation. The gap between "we use AI" and "AI is rebuilding how we work" is the real story for 2026.