[AI WEEKLY RUNDOWN] OpenAI Cuts Off Cursor, Nvidia Buys Hugging Face ($12.9B), & Medical AI Hits Clinical Reality (Aug 25–31, 2026)

🔍 Keywords PULSE Medical AI, Nvidia Hugging Face, OpenAI Cursor Cutoff

Summary:

In this weekly briefing, we analyze “The Physical Intelligence Convergence, Sovereign Compute, and the Clinical Evidence Gap.” We deconstruct OpenAI severing ties with Cursor post-SpaceX acquisition. We evaluate Nvidia’s massive $12.9 billion acquisition of Hugging Face and its $700 billion sales forecast. In healthcare, we examine Macau’s PULSE AI transforming routine blood draws into precision metabolomics, UCLH conducting live AI brain surgery, and a PLOS Digital Health report exposing that 99.7% of FDA-cleared medical AI devices lack patient-outcome evidence. Finally, we cover South Korea’s free citizen AI initiative, Andreessen Horowitz’s $1.1B hardware fund, and the federal ruling blocking the Pentagon’s Anthropic ban.

Itemized Important Topics:

  • OpenAI Terminates Cursor Partnership: OpenAI announces it will cut off Cursor’s access to its models by November 12, citing trust issues following SpaceX’s acquisition of Cursor parent Anysphere.

  • Nvidia Acquires Hugging Face for $12.9 Billion: Nvidia agrees to acquire open-source repository Hugging Face at an 80x ARR multiple ($150M ARR), securing control over the dominant distribution hub for open-weights models.

  • PULSE Framework Decodes Blood Metabolomics: Published in Nature Computational Science, Macau University’s PULSE AI reads 61 routine blood markers to generate 251 expensive metabolomic biomarkers, matching costly specialized laboratory panels.

  • PLOS Study Exposes FDA Medical AI Gap: An analysis of 1,357 FDA-cleared AI medical devices reveals that only 3 devices were ever evaluated in clinical trials measuring real patient outcomes like mortality or hospitalization.

  • Live AI Assist Used in Brain Surgery: Surgeons at UCLH complete a pituitary tumor removal assisted by a real-time computer vision AI that monitored camera feeds to flag hidden arteries and optic nerves.

  • Anthropic Launches Model Hardware Standard (MHS): Anthropic introduces an open protocol letting LLMs like Claude control laboratory microscopes, liquid handlers, and industrial robot arms without custom code.

  • South Korea Provides Free AI to Every Citizen: South Korea selects SK Telecom, KT, and Kakao to build free, unlimited basic AI services for all citizens, backed by 512 government-supplied Nvidia B200 GPUs.

  • Andreessen Horowitz Raises $1.1B “Machine Age” Fund: a16z unveils a $1.1 billion fund dedicated exclusively to AI hardware infrastructure, targeting semiconductors, robotics, storage, and data center energy.

  • 100+ Tech Firms Warn of Imminent AI Cyberattacks: OpenAI, Anthropic, Google, and Microsoft publish a joint call to action warning that agentic cyberattacks will soon become significantly harder to stop.

  • Federal Judge Overturns Pentagon’s Anthropic Blacklist: U.S. District Judge Rita Lin blocks the Department of Defense from blacklisting Anthropic over its refusal to allow Claude to be used for mass surveillance or autonomous weapons.

  • OpenAI Reveals Custom “Jalapeño” Chip: OpenAI showcases its Broadcom-built 700W inference chip, outperforming Nvidia’s 1,200W Blackwell in energy efficiency on DeepSeek R1 workloads.

🎧 Listen/WATCH ADS-FREE: https://podcasts.apple.com/us/channel/djamgamind/id6760446113

Visit our Research Hub at https://djamgamind.com/pdfs

Before we start, I need your vote:

My free bedtime-story app DjamgaMind Kids (600+ ad-free stories on Black history, African & Caribbean folktales) is a Top 10 finalist in Melamoon, a national pitch competition for Black founders in Canada. Next round is a public vote: Top 5 go to the Toronto Grand Finale in October. One vote, ~20 seconds, one per person, closes Sept 22:

Visit https://djamgamind.com/pitch to vote

Profile: https://melamoon.ca/top-10-finalists-vancouver/djamgamind.

Thank you!

#Melamoon2026 #DjamgaMind

OpenAI cuts off Cursor after SpaceX acquisition LINK

  • OpenAI said late Friday it will end its partnership with Cursor and cut the coding tool’s access to its models on November 12, roughly three months after SpaceX bought Cursor’s parent company Anysphere.

  • OpenAI blamed the split on trust rather than technology, writing that it cannot be sure SpaceX will follow its terms of service given past contract violations by X and other Musk-owned companies.

  • The cutoff pushes Cursor to rely more on Anthropic’s Claude and its own in-house Composer model, leaving open the question of whether Anthropic will make the same move to drop the tool.

South Korea gives every citizen free AI LINK

  • South Korea has picked three groups led by SK Telecom, KT and Kakao to build free, unlimited basic AI services for every citizen, with a full launch planned by the end of 2026.

  • The government will hand the consortia a combined 512 Nvidia B200 chips during 2026, and starting in 2027 will cover part of the running costs while the companies must add their own money and find revenue.

  • Rules set in July require at least half of each service’s model use to come from independent Korean foundation models, with 30% from other domestic developers, though the total cost and real-world capacity remain unsettled.

Meta tests robots for data center jobs LINK

  • Meta is trying out robots to help run the data centers behind its AI, with machines that can swap networking cables, reboot servers, move racks, and check equipment, according to a WIRED report.

  • The company is testing hardware from Watney Robotics, Kinova, and ABB, but the machines still need people watching over them and struggle with tangled cabling, battery life, obstacles, and visual inspections.

  • One worker estimated a working cable-swapping robot could take over up to 80% of some jobs, though that was the employee’s guess, not a Meta forecast, and the robots remain slower than humans.

The rise of AI agents ‘for normal people’

Edited by Max Lockie, Editor at LinkedIn News

A pair of AI startups are the buzz of Silicon Valley, thanks to their ability to make AI agents easier to use outside of programming environments. Town uses customizable characters to handle tasks, with a userbase that ranges from investors to plumbers. Instinct — which just skyrocketed to a a $2.5 billion valuation — operates through texts and calls to make purchases and handle errands. Both firms position AI agents as personal assistants rather than developer tools, though their access to personal data has raised privacy concerns.

Andreessen Horowitz raises $1.1B fund for AI buildout

Edited by Andrew Barker, Editor at LinkedIn News

On Friday, Silicon Valley venture capital giant Andreessen Horowitz unveiled its new $1.1 billion “Machine Age” fund, geared toward investments in the ever-accelerating artificial intelligence buildout. As the firm’s first fund dedicated to hardware infrastructure, the venture will focus on AI chips, data storage, networking and robotics. Announcing the fund on LinkedIn, Andreessen Hororwitz General Partner David George noted: “Every time computing shifts, the infrastructure underneath gets rebuilt. What’s different now is the size of it. ... We’ve never seen a compute transition this big move this fast.”

X finds 200000 fake energy accounts LINK

  • X says it found a network of 200,000 fake accounts tied to China, with 200 of them posting cartoons and text designed to sway American debate over AI data centers and energy policy.

  • The 200 accounts claimed data centers push up household electricity prices and strain local grids, sharing AI-made cartoons that showed operators as greedy businessmen while ordinary families paid the bill.

  • X said the accounts used OpenAI tools to create the images and talking points, matching a June OpenAI report about China-linked ChatGPT users generating anti-data center content and other political material.

Anthropic makes first move into physical AI LINK

  • Anthropic stepped into physical AI on Thursday with the Model Hardware Standard, a research preview that links large language models like Claude to real equipment such as factory robotic arms and lab microscopes.

  • MHS works with any model, not just Claude, and builds on Anthropic’s Model Context Protocol from 2024, letting companies connect AI to gear in minutes or hours instead of the usual weeks of custom work.

  • Anthropic built MHS with the HHMI Janelia Research Campus in Virginia, giving early access to labs and makers including Genentech, Carnegie Mellon, QuEra, Universal Robots, Amazon Web Services, Doosan Robotics, and Danaher.

Meta adds kill switch to smart glasses LINK

  • Meta is trying to fix the “pervert glasses” reputation of its camera-equipped Ray-Ban AI Glasses, with a new patent describing a “mute or lock” switch that can shut off the device’s sensors.

  • The patent, filed in February and granted last week, covers “secure hardware-enforced sensor privacy systems” for augmented reality devices, with drawings that closely match Meta’s current glasses, suggesting a response to its recent privacy scandals.

  • The move follows widespread misuse by creeps, pranksters, and bullies, plus a report that Nairobi-based contractors reviewed intimate footage; owners can already defeat the recording LED with a simple sticker.

OpenAI and 100+ firms warn AI cyberattacks are imminent LINK

  • More than 100 companies, including OpenAI, Anthropic, Google, Microsoft and AWS, signed an open letter warning that AI-powered cyberattacks will soon become far more common and harder to stop as models grow more capable.

  • The letter, titled “A call for collective action on cyber defense,” pushes organizations, governments and AI firms to treat security as a top priority and admit that current defenses won’t hold against these coming threats.

  • It asks frontier AI companies to build monitoring and security tools, make AI agents traceable and accountable, and share methods for spotting attacks, though the warning comes after chatbots were already used to hack governments.

Judge blocks Pentagon’s Anthropic ban LINK

  • A federal judge overturned the Pentagon’s decision to blacklist Anthropic as a supply-chain risk, ruling that the government violated the AI company’s constitutional rights in their fight over military use of its models.

  • U.S. District Judge Rita Lin wrote in her 59-page ruling that “national security is not a blank check” for retaliation, noting the Pentagon kept working with Anthropic even after labeling it a risk.

  • The clash began this year when Anthropic refused to let the military use Claude for mass surveillance of Americans or fully autonomous weapons; the government is expected to appeal, and a separate designation is being fought in the D.C. Circuit.

Nvidia paused its revenue-sharing deals LINK

  • Nvidia has paused its program that gave cloud providers credit support and revenue sharing to build AI data centers, worried that the deals could draw antitrust scrutiny, The Wall Street Journal reported.

  • Internal talks at Nvidia questioned how far the chip designer could go in backing its customers and shaping their operations, so it froze the effort last week, though the program was not canceled and could return later.

  • The July initiative included partners like Sharon AI, deploying 40,000 Blackwell GPUs, and Firmus, planning up to 170,000 GPUs in Indonesia; Nvidia’s earnings also revealed $108 billion in guarantees and credit deals for AI buildouts.

Nvidia is buying Hugging Face for $12.9B LINK

  • Nvidia has agreed to buy Hugging Face, the open-source AI model hosting platform, for about $13 billion, according to The Information, which first reported the talks before confirming the deal.

  • The price is roughly 80 times Hugging Face’s $150 million in annual recurring revenue and nearly double the $7 billion Nvidia reportedly offered back in January 2026, after the startup doubled its customer base during the year.

  • The deal lands as Chinese open-weight models like Z.ai’s GLM-5.3-Flash, whose weights are distributed on Hugging Face, gain attention, sharpening the debate over Western open AI following the Hot Chips conference.

Nvidia’s sales forecast eases AI bubble fears LINK

  • Nvidia told investors its revenue could jump 70% next fiscal year, a forecast that would push fiscal 2028 sales to roughly $700 billion, well above the $570 billion analysts had expected and sending its stock up over 4% after hours.

  • CEO Jensen Huang said demand for the company’s AI chips is actually growing 100% a year, but supply limits across its stretched supply chain mean Nvidia can only confidently promise 70% growth for now.

  • The forecast followed a strong second quarter, with revenue of $96.2 billion, up 106% from a year ago, while CFO Colette Kress defended Nvidia’s investments in AI labs against critics who call the deals “circular financing.”

Bill Gates proposes ‘Human Reserved’ jobs LINK

  • Bill Gates has floated the idea of “Human Reserved” jobs, occupations society would deliberately keep for people even when AI could handle them, naming childcare and jury service as roles humans are clearly better suited to.

  • He says at most 40% of jobs could realistically be protected, either permanently or temporarily for workers who cannot retrain, with education and healthcare kept selectively rather than shielded across the board.

  • To back this up, Gates wants to tax AI tokens and robots instead of banning automation, calling it a bigger change to the tax system than any in his lifetime while listing objections he cannot answer.

Data center backlash reshapes the midterms LINK

  • Growing anger over data centers across the U.S. has upended the 2026 midterm elections, forcing politicians from both parties to distance themselves from projects while President Trump defends an industry expected to draw over $700 billion this year.

  • The land-hungry projects have joined conservative ranchers and left-leaning environmental groups worried about water and farmland, as seen with Project Jupiter, a $165 billion Oracle and OpenAI complex planned near New Mexico’s drought-stricken border.

  • Governors are toughening rules, with Texas’s Greg Abbott ordering steps to shield residents from higher electric bills and vowing to strip the state’s billion-dollar-plus yearly tax break for data centers.

Zuckerberg’s plan to replace thousands of Meta workers with AI fell apart LINK

  • Mark Zuckerberg’s plan to swap thousands of Meta employees for AI, known as Project OT, collapsed after the CEO scrapped a second round of layoffs planned for November, according to documents seen by Reuters.

  • The scheme aimed to shrink teams by up to 60%, running two waves of cuts in May and November, with small human “talent-dense” groups overseeing virtual AI workers and new “AI systems” helping decide promotions.

  • The effort backfired as heavier AI coding caused more outages and possible data leaks, spiking to 405 incidents and forcing staff to spend up to 70% of their time fixing problems, though Meta still cut 10% of its workforce.

Anthropic’s top AI struggles to win users LINK

  • Anthropic’s most powerful AI model, Opus 5, is failing to draw many users since its late-July launch, with cheaper options from the company pulling far more spending, according to billing data.

  • The Ramp AI index, built from card data across 70,000 companies, shows Opus 5 captured just 3.5% of Anthropic model spend in July, while the older Opus 4.8 led with 28% and Fable 5 sat at 8%.

  • Anthropic’s annualized revenue reached $65 billion in July, up from $47 billion in May, and the company told investors it now has 6,000 customers each paying $100,000 or more a year.

OpenAI chip beats Nvidia’s Blackwell LINK

  • OpenAI has revealed “Jalapeño,” an inference chip built with Broadcom for running AI models, that beat Nvidia’s Blackwell along with AMD and Google chips on power efficiency across several top open source models.

  • The chip went from team hiring to tape-out in about 16 months, and hit over 700 tokens per second per user on DeepSeek R1, without speculative decoding or splitting prefill and decode work across separate chips.

  • The current A0 version already leads on tokens per megawatt, while a newer B0 chip in the fab adds roughly 25% more efficiency, delivering 13.4 PFLOPs on TSMC’s N3P at just 700W, and ships with HBM4 memory.

Robots beat Usain Bolt’s 100m record LINK

  • A humanoid robot called Tiangong Ultra, built by Beijing’s X-Humanoid, ran the 100-meter sprint in 9.39 seconds at the World Humanoid Robot Games, beating Usain Bolt’s human record of 9.58 seconds from 2009.

  • A second Chinese robot named Lightning, made by phone maker Honor, also topped Bolt’s mark with 9.47 seconds in the final, though both machines lost control afterward, with Tiangong veering off track and Lightning carried away.

  • The five-day Beijing event drew over 2,000 robots from 666 teams across 16 countries for 51 events, from soccer to warehouse logistics, with more than 40% of contests requiring the robots to operate on their own.

Most Americans oppose nearby data centers LINK

  • Three-quarters of Americans say they don’t want a data center built near them, according to a new poll from Heatmap, with 64% strongly opposed and only 15% in favor of a nearby facility.

  • The turn against data centers spans age, gender, income, party, and the rural-urban split, with the facilities running 43 points underwater with Republicans, 65 with independents, and 75 with Democrats.

  • Opposition has swung 33 points in a year across four surveys using the same wording, rising from 42% last August to 75% now, and a July Redfin poll found similar results, with 53% opposed.

Anonymous AI coding model wows devs LINK

  • A mysterious AI coding model called Ox Alpha has shown up on OpenRouter as an anonymous, free-to-use release, and developers testing it have been impressed by its performance on programming and agent tasks.

  • The model offers a context window of just over a million tokens, and the open-source agent OpenCode said it would be free for a week with its provider handling up to 100 trillion tokens a day.

  • Nobody knows who made it, with guesses ranging from China’s Z.ai to Microsoft’s MAI family, and OpenRouter warns that prompts and completions are kept by the unnamed provider, a problem for European firms under data protection law.

Nvidia hikes AI server prices over 15% LINK

  • Nvidia has warned its biggest customers that servers built around its AI chips will cost more than 15% extra, with the higher prices hitting Grace Blackwell and Vera Rubin systems shipping early next year.

  • The exact increase depends on the chip generation and memory setup, and it stems from soaring DRAM prices, with contract makers building servers for Microsoft, Google, and Oracle already telling their customers about the coming jumps.

  • A 15% rise on rack-scale systems costing several million dollars each adds hundreds of thousands per rack, and with TSMC unable to meet demand, buyers have little room to push back against Nvidia’s roughly 75% margins.

Twitch sued for training AI on streamers LINK

  • A Connecticut streamer named Warren Pandiscia has filed a class action lawsuit against Twitch and Amazon, saying the companies used his streams and videos to train AI models without getting his permission or a license.

  • The complaint claims Twitch and Amazon broke their contract by treating streamers as “free training stock” for separate commercial AI products, and that creators lost money or property because of the practices.

  • Earlier this month Twitch added a setting to opt out of AI training, but it’s on by default, and product chief Mike Minton said on a Patch Notes stream that “if it was opt-in, nobody would opt in.

What Else Happened in AI this last week of August 2026?

OpenAI, Anthropic, AWS, Microsoft, and 100+ organizations left warnings that critical infrastructure has only months to prepare for cheaper, more accessible AI-enabled cyberattacks. — Axios

Advent and Stripe dropped their pursuit of PayPal after offering more than $50 billion, ending takeover talks that could have reshaped one of online payments’ oldest names. — Bloomberg

A federal judge blocked the Pentagon’s blacklisting of Anthropic and ordered agencies to stop enforcing it, after the company refused to permit unrestricted military use of Claude. — CNBC

Hugging Face unveiled Microduck, a $399 open-source robot developers can train in simulation and deploy in the real world, making physical AI experimentation cheaper, cuter, and more accessible. — TechCrunch

Anthropic opened a research preview of its Model Hardware Standard, giving AI agents a common interface for safely operating lab and manufacturing equipment while reducing the custom engineering needed to connect each device. — Anthropic

Nvidia reportedly agreed to acquire Hugging Face for $12.9 billion, bringing a leading open-source AI platform under the chipmaker’s control and extending its reach from infrastructure into models and developer software. — CNBC

Meta agreed to pay up to $16.7 billion to settle claims from 29 states that it concealed child mental health harms. In response, the company has now added teen usage limits, nighttime blocks, stricter age checks, and parental controls. — CNBC

Anthropic will reportedly spend $45 billion renting AI computing capacity from Nscale, a deal that would give the infrastructure startup a major anchor customer while underscoring frontier AI’s soaring capital demands. — Reuters

In medicine and biotechnology, AI is moving from text summaries to physical execution and deep biological synthesis: Macau University's PULSE framework can now infer 251 high-cost metabolomic biomarkers directly from 61 routine blood markers, UK surgeons at UCLH completed the world's first live-video AI-guided brain surgery, and Anthropic launched its Model Hardware Standard (MHS) to connect Claude directly to laboratory equipment. However, a landmark PLOS Digital Health study revealed a glaring reality: out of 1,357 FDA-cleared AI medical devices, only 3 were actually evaluated on patient-centered health outcomes.

🔗 RESOURCES

Related articles

← All articles