The Omniscient Bulletin · 2026-06-30
The Omniscient Bulletin — June 30, 2026
Brussels delays its toughest AI rules as the bills for the boom, in compute and memory, come due
The AI build-out ran into its limits today. In Brussels, the European Union gave final approval to push back its toughest AI rules by more than a year, easing a rulebook industry had called too heavy. The strain showed on the supply side too: Google conceded it could not sell Meta all the Gemini capacity it wanted, and a US lawsuit accused the three big memory makers of choking supply to profit from AI. Two stories cut the other way, as Palantir and Nvidia pitched 'sovereign' AI to Washington and a new study found the firms spending most on AI are hiring, not firing.
Policy
EU gives final approval to delay its high-risk AI rules and ban deepfake nudes
The Council of the European Union gave final approval to its 'digital omnibus' on artificial intelligence, the last step before the law is published. It delays the AI Act's high-risk rules, once due in August, to December 2027 for standalone systems and August 2028 for AI built into products, and pushes the deadline for national regulatory sandboxes to August 2027. It also adds a new prohibition, effective this December, on AI that generates non-consensual intimate images or child sexual abuse material, one of the few places the omnibus tightens rather than loosens the rules.
Compute
Google rations Meta's access to Gemini after running short on compute
Google has limited how much of its Gemini models Meta can buy, after Meta sought more computing capacity than Google could supply, the Financial Times reported. Google told Meta around March that it could not meet the full request, delaying some of Meta's internal AI projects; other Google Cloud customers were hit to a lesser degree. Meta has since pressed staff to use AI tokens more sparingly. The episode is a rare admission of scarcity from a hyperscaler: Google's cloud backlog nearly doubled last quarter, and Sundar Pichai has said a shortage of computing power is capping its growth.
Industry
Palantir and Nvidia build a sovereign-AI engine to run open models for US agencies
Palantir and Nvidia unveiled a joint system that lets US government agencies run Nvidia's open Nemotron models on their own infrastructure, train them on their own data, and keep full ownership of the resulting weights, including in air-gapped networks cut off from the open internet. It pairs Nvidia's enterprise software and chips with Palantir's Sovereign AI Operating System, the stack built on its Foundry and Apollo products. The pitch is 'sovereign' AI for commerce, energy, and health agencies that cannot send sensitive data to a commercial cloud. Palantir shares rose on the news.
Compute
Chipmakers sued in the US over claims they choked memory supply to chase AI
A class action filed in California accuses the three companies that make almost all of the world's DRAM, Samsung, SK Hynix, and Micron, of deliberately starving the market to drive up prices. The 17 plaintiffs say the firms have throttled supply of commodity memory since 2022 under the cover of shifting capacity to the high-bandwidth memory that AI chips need, helping push DRAM prices up about 700 percent in four years. The three control more than 90 percent of DRAM sales. The suit seeks to halt the alleged coordinated cutbacks and to win triple damages.
Industry
Chamath Palihapitiya takes over an AI coding startup with a fresh $135M
Chamath Palihapitiya stepped in as chief executive of 8090, an enterprise AI coding startup, which raised a $135 million Series A led by Salesforce Ventures. The investor left his board seat to run the company full time, saying 'there was no decision to make except to be all in.' Founded in early 2024, 8090 sells a 'Software Factory' aimed at corporate engineering teams that want production-grade software with audit trails, rather than what it calls vibe-coded prototypes. Other backers include the venture firms of Jeffrey Katzenberg, David Sacks, and David Friedberg.
Labor
A new study says heavy AI adopters are hiring more, not less
A study from the finance startup Ramp and the labor-data firm Revelio Labs offers a counterpoint to the AI-takes-jobs story: companies that spend heavily on AI, about $30 per employee a month in their first three months, grew their headcount 10.2 percent, with entry-level roles up 12 percent. The authors are careful to note it does not prove AI creates jobs, since the sample of roughly 22,000 firms skews toward venture-backed tech companies that may be hiring anyway. It runs against Goldman Sachs figures showing the economy has shed about 16,000 jobs a month to automation.