Deals, economics, and the business of AI.
Weeks after its ~$7.4bn June round, DeepSeek is preparing to raise as much as 50 billion yuan at a ~$74 billion valuation and has begun early work toward an IPO on Shanghai's STAR Market. The company that made frontier AI cheap is now raising like a frontier lab — and doing it entirely onshore.
A Samsung Health consent notice read like an ultimatum — train our AI on your medical records or lose them. Samsung says the warning was misleading and opting out deletes nothing you rely on. The panic was an overreaction; the pattern that caused it isn't going away.
Google's beloved research tool gets a code-running 'secure cloud computer' and cross-app sync — but the real story is the brand consolidation, the same week the rumour mill has Gemini 3.5 Pro finally arriving. The model link is our reading; Google's post names no model.
The EU's General Court upheld a refusal to register 'OPENAI' as a trademark, ruling the name merely describes 'openly accessible AI.' It's the latest time the descriptiveness rule has caught the company — after the US refused 'GPT' in 2024.
The viral $23bn figure is real, computed by PJM's independent market monitor — and six months old, narrower than the headline, and an order-of-magnitude estimate. The actual news came the same day, quietly: a third straight capacity auction at the price cap.
Standard-form exclusions for generative AI went live in US business liability policies in January. Now the organisation that writes those standard forms says it is looking at agentic AI too. For companies deploying AI, the quiet story of 2026 is that the ordinary insurance safety net is being cut away — just as a new, specialised market moves in to sell it back.
Moonshot AI's Kimi K2.7 Code is now generally available in Copilot's model picker — hosted by GitHub on Azure, so your code need never touch Moonshot's servers. In the week of the Fable 5 compact, the open-weight lane keeps flowing through the export-control chess match.
While the US gated GPT-5.6, DeepSeek open-sourced a technique that speeds up serving its models 60–85% per user — over its own prior method, in its own system. The number needs a footnote; the strategy is the story.
Two AI-and-jobs headlines this week looked like opposites — one firm shedding people because of AI, another rehiring them because of it. Read together, they're the clearest map yet of what AI is actually doing to work.
Built with Broadcom and designed in a claimed-record nine months, the inference chip is OpenAI's move to own the hardware under its models — and ease its dependence on Nvidia. The performance numbers, for now, are OpenAI's own.
In a regulated filing, Oracle said AI is reducing its workforce — rare corporate candour. But the figure is a full year's cumulative cuts, the language is a risk disclosure, and AI is doing the cutting two ways at once.
Insilico Medicine and SK Biopharmaceuticals just announced an 'up to $2.5bn' AI drug-discovery deal. The real upfront sum is about $18m. Here's the genuine science, and how to read the billion-dollar headlines that follow these deals.
Zhipu doesn't charge for its GLM models — yet the market briefly valued it above $128bn. How an open-weight lab actually makes money, why it gives the models away, and what that trillion-HKD number is really pricing.
In sixteen days in May, CME, the New York Stock Exchange's owner, and a third challenger all moved to launch futures markets for renting GPUs. Here's what compute futures are, the three rival bets, and the serious case that it won't work.
John Jumper — who led AlphaFold and shared the 2024 Nobel Prize in Chemistry — is leaving Google DeepMind after ~9 years to join Anthropic, announced via his own X post and confirmed by Bloomberg and Reuters.
Noam Shazeer — co-author of the 2017 transformer paper and a co-lead of Google's Gemini — is leaving Google to join OpenAI, announced this week.
Per the FT, Cambridge startup CuspAI is reportedly raising ~$400m at a ~$2.6bn valuation, backed by Bezos Expeditions and Kleiner Perkins — but term sheets are only signed, the deal hasn't closed.
SpaceX (Nasdaq: SPCX) signed a merger agreement on 16 June 2026 to acquire Anysphere, the maker of the AI code editor Cursor, for $60 billion in an all-stock deal — disclosed the same day in an SEC filing.
PwC's 2026 Global AI Jobs Barometer (1bn+ job ads, 27 countries): AI skills now carry a 62% average wage premium (up from 57%), and AI-skill jobs are growing ~69% a year vs 9% for the market overall.
KPMG pulled its report "Redefining excellence in the age of agentic AI" (published Oct 2025) after named organisations said its claims about their AI use were untrue — GPTZero traced the errors to AI hallucinations.
According to The Information, Amazon CEO Andy Jassy was among the tech leaders who raised concerns to senior Trump administration officials about security risks in Anthropic's most advanced model before the US ordered it disabled.
Jeff Bezos's AI startup Prometheus raised $12bn in June 2026 at a $41bn valuation — extraordinary for a company that only launched in November 2025 with $6.2bn.
In a 12 June internal memo (first reported by Reuters/Bloomberg), Zuckerberg admitted: 'we've made mistakes and will almost certainly make more' in Meta's AI restructuring.
OpenAI is acquiring Ona — formerly the cloud-dev-environment maker Gitpod — to give Codex a secure place to run long, autonomous tasks. Terms undisclosed; subject to regulatory approval.
Apollo Global Management and Blackstone have assembled a roughly $36 billion private credit deal — reportedly the largest on record — to finance Anthropic's compute build-out, coming together this week (early June 2026).
OpenAI and Oracle announced (late 10 June) that Oracle Cloud Infrastructure (OCI) customers will be able to apply their eligible Oracle Universal Credits toward OpenAI's frontier models and Codex.
Visa and OpenAI announced a partnership (10 June, at the Visa Payments Forum) letting AI agents in ChatGPT make real purchases on a user's behalf using tokenized Visa credentials.
TensorWave raised a $350M Series B at a $1.55B valuation, led by AMD and hedge fund Magnetar Capital — nearly 4x its ~$400M valuation a year ago.
Per a report from The Information, OpenAI is in advanced talks to lease a proposed 10-gigawatt data-centre campus on federal (Department of Energy) land in southern Ohio.
Reliance Industries and Meta announced (10 June) a 168MW AI-enabled data centre in Jamnagar, Gujarat, to be delivered within ~2 years with an option to scale.
PhysicsX closed an oversubscribed $300M Series C at roughly a $2.4B valuation, led by Temasek — more than double its valuation from under a year ago.
PointFive raised a $60M Series B led by Accel (with Index Ventures, Salesforce Ventures and others) at a ~$500M valuation — $96M raised to date.
China is drafting a five-year blueprint to spend ~2 trillion yuan (~$295bn) on a network of interconnected national AI computing hubs, led by the National Development and Reform Commission.
OpenAI confirmed on 8 June it has confidentially filed a draft registration (an S-1) with the SEC — the standard first step toward a public listing.
DeepSeek is closing its first-ever external raise — ~50bn yuan (~$7.4bn) — at a $52–59bn valuation, reportedly within weeks.
Flourish raised $500m at a $2.5bn valuation; Jeff Bezos put in ~$100m, alongside Alphabet's GV, Lux Capital and Catalio.
Moonshot AI — maker of the Kimi chatbot — is in early talks to raise up to ~$2bn at a ~$30bn valuation.
Buy the model, build the workflow: foundation models are now a commodity input, but the orchestration, data plumbing, and evals around them are where durable advantage lives.
Hourly AI implementation work is a commodity race to the bottom; margins compress as the tooling that makes you fast also makes you replaceable.
Time-saved estimates are the weakest form of AI ROI — they're easy to inflate and rarely convert to cash unless headcount or capacity actually changes.
AI-ops is emerging as a distinct discipline: the continuous work of monitoring, evaluating, securing and updating AI systems after they ship.
The demo-to-production gap is mostly organisational, not technical: ownership, integration, and accountability gaps kill more pilots than model quality does.
AI pricing is climbing a ladder: raw token/compute pricing at the bottom, per-seat and usage in the middle, outcome-based pricing at the top.
Open-weight models have closed enough of the capability gap that the choice is now about control, cost-at-scale, and data residency — not raw quality alone.
