The Omniscient Bulletin · 2026-07-07
The Omniscient Bulletin — July 7, 2026
The buildout's bills came due: Microsoft paid in jobs, Anthropic in a 20-year lease, and ransomware learned to run itself
The bills for the AI buildout all seemed to arrive on Monday. Microsoft cut 4,800 jobs while planning 190 billion dollars in spending, most of it on AI. The United Nations opened its first global AI summit with a call to ban killer robots. Anthropic rented an entire Kentucky campus for 20 years, SK Hynix launched the biggest American listing ever by a foreign company to feed the memory boom, and Nvidia's next big rack reportedly slipped a year. Security researchers documented the first ransomware attack run by an AI agent. And OpenAI shipped voice models built for the least glamorous frontier of all: the customer-service call.
Labor
Microsoft cuts 4,800 jobs as its AI spending bill comes due
Microsoft said Monday it is cutting about 4,800 roles, some 2.1 percent of its workforce, with 1,600 people laid off immediately. Gaming takes the hardest hit: Xbox sheds roughly 3,200 jobs over the next year, and four studios are being cut loose, with Compulsion Games and Double Fine going independent. Chief people officer Amy Coleman insisted the eliminated roles 'are not being replaced by AI,' but the cuts land as Microsoft projects roughly 190 billion dollars in capital spending this year, most of it AI infrastructure. Shares slipped 1.4 percent Monday.
Policy
The UN's first global AI summit opens with a call to ban killer robots
The Global Dialogue on AI Governance, the UN's first government-level AI forum, opened Monday in Geneva. Secretary-General Antonio Guterres called lethal autonomous weapons 'morally repugnant' and said the technology 'must be banned by international law.' He set out four priorities: safety baselines for frontier systems, human-rights red lines, AI capacity for developing countries, and disclosure of the industry's environmental footprint. More than 20 countries backed the capacity push on day one. The line aimed at Silicon Valley: 'We cannot vibe-code the future of humanity.'
Compute
Anthropic signs a 20-year, $19 billion lease on a Kentucky data center campus
Anthropic is taking the entire Justified Data campus in Hawesville, Kentucky, a former aluminum smelter site that TeraWulf bought in February for 200 million dollars. The 20-year lease covers up to 401 megawatts of computing load and is expected to bring TeraWulf about 19 billion dollars in revenue, with first capacity in late 2027 and the full buildout by early 2028. TeraWulf shares jumped, and it sold its stake in a Texas project to partner Fluidstack the same day. For a lab that has spent a month rationing its most powerful model, the message is simple: more compute is on the way.
Data Center Dynamics / SEC filing→
Industry
SK Hynix kicks off a $28 billion US listing, the biggest ever by a foreign company
The Korean memory maker launched its Nasdaq offering Monday, selling 17.79 million shares as American depositary receipts, with final pricing Thursday and trading set for Friday. At about 28 billion dollars it would be the largest first-time US share sale by a foreign company, ahead of Alibaba and Saudi Aramco, and second on record only to SpaceX's debut last month. Proceeds fund new Korean fabs and ASML's most advanced chipmaking machines. SK Hynix holds 56.4 percent of the market for the high-bandwidth memory that AI accelerators depend on, and its offering is a bet the boom holds.
Compute
Nvidia's next-generation AI rack reportedly slips a year to 2028
Research firm SemiAnalysis reported that Kyber, the liquid-cooled rack Nvidia promised for its Rubin Ultra chips in 2027, has slipped to 2028 because its circuit-board midplane is proving too hard to manufacture. The 600-kilowatt design, unveiled at GTC 2025, was pitched at four times the performance of today's Blackwell systems; a stopgap that bolted two current racks back to back was scrapped after big cloud customers balked. Nvidia's response: 'Our roadmap is intact.' If the report holds, the annual upgrade rhythm the whole AI buildout is priced on just broke.
SemiAnalysis / Data Center Dynamics→
Research
Researchers document the first ransomware attack run end to end by an AI agent
Security firm Sysdig says an operation it calls JadePuffer is the first known ransomware attack driven end to end by a large language model. The agent broke in through a known Langflow flaw, moved across the network, escalated privileges, encrypted more than 1,300 configuration records, and wrote its own ransom note, fixing one failed login in 31 seconds while narrating its reasoning in code comments. A human still picked the victim and set up the infrastructure. The model is unknown; a Microsoft researcher suspects an open-weight system stripped of its safety training.
Frontier
OpenAI ships new voice models tuned for the phone lines
OpenAI released gpt-realtime-2.1 and a cheaper mini version for its Realtime API, the speech-to-speech line behind phone agents and voice assistants. The models are better at reading back codes and account numbers, handling background noise, and recovering when callers interrupt, and developers can now dial reasoning effort up or down. OpenAI says worst-case latency across its voice models is down at least 25 percent through better caching. No frontier fireworks, just the unglamorous work of making AI hold up on a customer-service call, where most people will actually meet it.