AI news4
OpenAI projects $278B in negative cash flow through 2030 as a $1.2T round looms
An internal OpenAI presentation from July, reported by the Financial Times and confirmed by Bloomberg, projects $278 billion in negative free cash flow from 2026 to 2030, against roughly $856 billion of compute and infrastructure spending and about $840 billion of cumulative revenue. The company's record $122 billion March round is now expected to run out around 2028, two years before its own models project positive cash flow, and investors have approached OpenAI about a new round valuing it above $1.2 trillion, up from $852 billion in March.
techtimes.com ↗Why it matters
OpenAI is now telling investors its last record raise bought runway to the next raise rather than to self-sufficiency, which reframes frontier AI funding as a permanent obligation rather than a series of milestones.
Nvidia-backed Nscale files for a $35B US listing
British data centre group Nscale has filed to go public in the US, targeting a valuation of up to $35 billion, just two years after being spun out of a crypto mining group. The Nvidia-backed AI cloud provider, whose board includes former UK deputy prime minister Nick Clegg and former Meta executive Sheryl Sandberg, reported a $1 billion net loss on $141 million of revenue in the first half of the year, mostly from non-cash charges.
chipbriefing.substack.com ↗Why it matters
A loss-making AI cloud provider testing public markets at a $35 billion valuation is the first real read on whether investors will keep funding compute infrastructure at current prices.
China's CXMT puts its fifth-generation memory platform into mass production
Chinese DRAM maker CXMT says its fifth-generation technology platform has entered mass production, a step toward competing with Samsung, SK Hynix and Micron in the global memory market. The platform is designed to make more powerful memory at lower cost and with less power, and CXMT is also eyeing the NAND flash market alongside domestic peer YMTC.
chipbriefing.substack.com ↗Why it matters
Memory is the second major input cost after logic chips for AI training, and a credible domestic Chinese supplier changes the pricing and export-control picture for every large model builder.
Anthropic and Accenture to embed independent safety evaluators inside the lab
Anthropic and consulting firm Accenture announced a partnership to create a team of embedded evaluators that would sit inside Anthropic alongside employees, observe model-training decisions in real time, and provide independent safety assessments, with each company reportedly committing at least US$1 billion over five years to the effort. The arrangement follows Anthropic's call for outside verification of frontier models and its backing for independent evaluators.
winzheng.com ↗Why it matters
Placing outside evaluators inside a lab is the most concrete test yet of whether third-party safety oversight can work in practice, rather than as a pledge made to regulators.
AI in the nonprofit sector3
Raspberry Pi Foundation brings Experience AI to Australia and New Zealand with a $1.2M Google.org grant
The Raspberry Pi Foundation is expanding its Experience AI programme into Australia and Aotearoa New Zealand, supported by a $1.2 million Google.org grant, to train 5,000 educators over four years and reach about 150,000 students by 2028. In Australia the University of Adelaide's Computer Science Education Research Group is the implementation partner, while the New Zealand rollout is led by Maori-led organisation Tonui Collab Charitable Trust, adapting the curriculum to local context rather than importing it unchanged. Experience AI won the UNESCO King Hamad Bin Isa Al-Khalifa Prize in 2025.
mechanism.me ↗Why it matters
A funder-backed, locally adapted AI curriculum with named delivery partners in both countries gives Australian and NZ schools a ready path into AI literacy without each building its own.
Mercy Health joins a new National Diagnostic AI Consortium as a founding member
Mercy Health has been named a founding member of a new National Diagnostic AI Consortium focused on diagnostic artificial intelligence, a multi-organisation effort that could affect procurement, clinical workflows and data governance inside member health systems. The public announcement confirms membership but has not yet published a charter, funding figures, timelines or named pilots.
zeeshank9.com ↗Why it matters
Health systems are joining diagnostic AI consortia before the governance terms are public, which is exactly the stage where data-sharing and liability clauses get set by default rather than by design.
Tech To The Rescue and Google.org open the next AI Impact Scaling cohort, Matching Day 23 September
Fifteen social impact organisations referred through Google.org's Skills-Based Volunteering Fellowship and Tech To The Rescue's network are spending twelve months inside an AI enablement programme, working with dedicated pro bono tech teams on data readiness, cybersecurity and responsible AI. The cohort joins a peer community of more than 50 organisations, and meets on 23 September for Matching Day to pair with tech teams from a network of more than 2,000 companies. Organisations enter with a working, evidenced solution rather than a pilot.
techtotherescue.org ↗Why it matters
A structured pro bono build with data-readiness and security support is the scarce part of nonprofit AI adoption, and the requirement to arrive already delivering results keeps the programme from becoming a pilot farm.