AI news5
Big Tech spent $130B on AI in a single quarter and Google literally ran out of cloud capacity
Microsoft, Alphabet, Meta, and Amazon reported Q1 2026 earnings on Wednesday with combined capex of roughly $130B in the quarter alone, nearly 2x Q1 2025. Sundar Pichai said Google Cloud's revenue would have been higher if Google could have built infrastructure fast enough to meet demand. Microsoft's AI business hit a $37B run rate (up 123% YoY) with 20M paid M365 Copilot seats; Google Cloud grew 63% to $20B with a $462B backlog; AWS Trainium passed a $20B run rate with OpenAI committing 2GW and Anthropic up to 5GW; Meta raised 2026 capex guidance to $125-145B and Google raised 2026 capex to up to $190B.
theneuron.ai ↗Why it matters
Hyperscaler capacity is now the binding constraint on enterprise AI rollouts globally, including for nonprofits running Copilot, Gemini, or Claude in production, and procurement timelines should assume waitlists for serious workloads through 2026.
Anthropic's revenue per active user is now triple Microsoft's and seven times OpenAI's, per Counterpoint
Counterpoint Research data published this week pegs Anthropic's average monthly revenue per active user at $16.20, far ahead of Microsoft's $5.00, OpenAI's $2.20, and Google's $1.10. The figure crystallises Anthropic's enterprise-and-developer mix versus OpenAI's mass-consumer base, and aligns with Anthropic crossing $1T in valuation this week. Independent analyst LinkedIn coverage flagged the gap as Anthropic capturing premium use cases while competitors chase scale.
theregister.com ↗Why it matters
For organisations choosing a primary frontier vendor, ARPU difference of this size signals Anthropic is the deeper-workflow bet and OpenAI is the broader-distribution bet, and procurement decisions should reflect which one matches the actual user base.
Microsoft 365 Copilot becomes multi-model and auto-routes between OpenAI and Anthropic per task
Microsoft confirmed this week that M365 Copilot now routes traffic between OpenAI's GPT and Anthropic's Claude automatically, deciding which model to call for each task. Accenture has signed up 740,000 paid Copilot seats internally. The change comes as Microsoft renegotiated its OpenAI deal to non-exclusive, with revenue share through 2030 and licensing through 2032, and as both labs released updated frontier models in the past two weeks.
theverge.com ↗Why it matters
Procurement for nonprofits standardised on Microsoft 365 just got materially better without a vendor change, and the cross-vendor routing pattern is likely to spread across major productivity suites in the next 12 months.
Mayo Clinic AI detects pancreatic cancer up to 3 years before diagnosis on routine CT scans, doubling early-detection rates
Mayo Clinic published a landmark validation study on Tuesday showing its AI model can detect pancreatic cancer up to three years before clinical diagnosis from standard abdominal CT scans, identifying subtle changes specialists miss. The model doubles the early-detection rate for one of the deadliest cancers, where late-stage detection is the dominant survival problem. The validation study was conducted on routine images already in patient records.
newsnetwork.mayoclinic.org ↗Why it matters
This is one of the cleanest examples to date of frontier AI clinically validated on existing health data, and it is the template every disease-focused nonprofit should now apply to its own imaging or longitudinal-data archives.
Wired publishes OpenAI's Codex system prompt and the "never talk about goblins" line goes viral
Wired published OpenAI's Codex system prompt on Wednesday, including the explicit directive: "Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant." Anthropic's Amanda Askell publicly mused that she is not sure Claude's analogous behaviour is even a problem. The leak is the rare moment a frontier lab's actual production prompt enters the public record without a CFAA fight.
wired.com ↗Why it matters
Production system prompts are now part of the public discourse on how labs steer models, and any organisation deploying frontier AI in customer-facing surfaces should expect their own prompts to leak and design accordingly.
AI in the nonprofit sector4
Aligning Science Across Parkinson's and the Michael J. Fox Foundation commit $261M to expand the Collaborative Research Network
Aligning Science Across Parkinson's (ASAP) and The Michael J. Fox Foundation announced $261M in new grant funding on Tuesday to expand the global Collaborative Research Network, which maps the biological underpinnings of Parkinson's to enable personalized treatments. UTHealth Houston received a $7.8M three-year slice to study brain processes underlying the disease. The Foundation also held its annual Central Park event on 25 April, raising nearly $2M from 3,000+ attendees.
michaeljfox.org ↗Why it matters
This is the largest single 2026 disease-research foundation commitment to date, and the data-sharing requirements built into ASAP's grant terms remain the strongest precedent for open nonprofit research-data infrastructure in the AI era.
Texas Children's Hospital wins $6.7M NIH grant to publish one of the largest open Alzheimer's drug-discovery datasets
A Texas Children's Hospital researcher was awarded a $6.7M NIH grant this week to build and publicly release one of the largest open Alzheimer's drug-discovery datasets, spanning early brain development to disease onset. The dataset will be made freely available to researchers worldwide, lowering the cost base for any AI-assisted drug-discovery effort targeting Alzheimer's. The award lands the same week Chan Zuckerberg Biohub committed $500M to a Virtual Biology Initiative for open AI cell models.
firstwordpharma.com ↗Why it matters
Two open-data Alzheimer-and-cell-biology infrastructure commitments in one week sets the pattern for federally and philanthropically funded nonprofits to lead the open-data layer that AI drug discovery actually needs.
Food Allergy Fund Summit signals a new era of AI-led detection and drug repurposing for food allergies
The Food Allergy Fund 2026 Summit on Tuesday spotlighted AI-driven detection methods and repurposed-drug pipelines as the central direction for food-allergy research. Speakers framed the convergence of AI imaging, biomarker discovery, and existing-drug screening as a structural shift in a research area that has historically been chronically under-funded. The Fund itself is a small disease-specific philanthropy that has been catalytic in shifting research investment toward earlier intervention.
biospace.com ↗Why it matters
The pattern, where a small disease-focused nonprofit uses AI to leapfrog larger research budgets, is replicable, and any rare-disease or under-funded condition foundation should now look at AI detection plus repurposed-drug screening as the highest-leverage 2026 strategy.
Chronicle of Philanthropy: foundations created the nonprofit jargon crisis and AI is making it harder to fix
A Chronicle of Philanthropy opinion piece this week argues that grantmakers themselves trained nonprofits to write in jargon through every touchpoint in the grant-making process, from letters of inquiry to progress reports, and that funders have the power to stop. The piece lands as nonprofits increasingly use AI to draft grant applications in the very language funders rewarded historically. The op-ed explicitly puts the fix on funders, not nonprofits.
philanthropy.com ↗Why it matters
AI is amplifying whatever style and framing funders have rewarded, so the jargon problem will get worse fast unless funders simplify their own templates, and this is the cleanest near-term lever for sector-wide AI-quality improvement.