Anthropic just leapfrogged OpenAI on revenue and rolled out a private model that finds software zero‑days faster than the security industry can patch them, dragging banks and regulators directly into AI model governance. AI compute and data centers are turning into politicized infrastructure bets — Texas subsidies, Maine bans, Iran threats on Gulf data centers — while Samsung, OpenAI, and others pour eye‑watering capex into chips and power.
At the same time, AI agents are driving real layoffs and youth backlash, so every dollar you put into this stack is now a compressed bet on cyber risk, energy, geopolitics, and labor all at once.
Key Events
/Anthropic reached a $30B revenue run rate and expanded a long‑term next‑gen TPU partnership with Google and Broadcom.
/Claude Mythos uncovered thousands of zero‑day vulnerabilities across major operating systems and browsers and triggered the $100M Project Glasswing defensive initiative.
/France ordered ministries to migrate government PCs from Windows to Linux and move health data to European clouds by 2026.
/Texas is paying over $1B a year in tax breaks to data centers while Samsung builds a major fab there for Tesla’s chips.
/Alibaba pivoted Qwen from open to closed models, stood up a 10,000‑chip AI cluster, and delivered $5M in annual savings to Shopify.
Report
Anthropic just overtook OpenAI on revenue while its Claude Mythos model quietly blew a hole in the cybersecurity market.
At the same time, AI compute and data centers are turning into politicized infrastructure bets, not boring cloud procurement.
mythos, glasswing, and the AI cyber shock
Anthropic’s new Claude Mythos model is finding thousands of high‑severity zero‑day vulnerabilities across major operating systems and browsers.
During testing it surfaced a 27‑year‑old TCP selective acknowledgment bug in OpenBSD and a 16‑year‑old flaw in FFmpeg that human experts had missed.
Mythos scores 93.9% on the SWE‑bench Verified coding benchmark. In Anthropic’s internal security suite it solved 100% of tests, outperforming earlier models like Opus 4.6.
Anthropic has decided Mythos is too dangerous to release, restricting access to select billion‑dollar companies, governments, and cybersecurity partners via the $100M Project Glasswing initiative.
Glasswing has already driven stock declines in incumbent cybersecurity vendors and triggered US and UK regulators to flag frontier models as financial‑stability and national‑security risks.
compute, chips, and data centers as political infrastructure
Texas is handing out over $1B a year in tax breaks to data centers even as its grid and water systems strain under new AI loads and drought conditions.
Maine is preparing a temporary ban on new data‑center construction, and analysts expect nearly half of US facilities planned for 2026 to be delayed or canceled amid AI‑driven overbuild and economic pressure.
OpenAI has paused its planned ‘Stargate’ UK data‑center project over electricity prices and regulation while projecting around $121B of compute spending and no profits until after 2030.
Amazon is committing $25B to new data centers in Mississippi while Microsoft is redesigning facilities in conflict zones after Iranian threats on Middle Eastern sites, putting hyperscale AI infra squarely in the path of geopolitics.
Samsung is booking record $38B quarterly profit from AI chips, has raised DRAM prices by about 30%, and is building a major Texas fab for Tesla’s next‑gen chips and the 1‑terawatt‑per‑year Terafab project.
digital sovereignty and fractured AI regulation
France has ordered ministries to migrate government PCs from Windows to Linux, present exit plans by autumn, and move roughly 80,000 social‑security staff and national health data onto European clouds and sovereign tools by 2026.
Red Hat has laid off its entire engineering team in China amid US–China tech tensions, and European officials are openly talking about forking US‑controlled software stacks to regain control.
Japan has loosened privacy rules to position itself as the “easiest country to develop AI,” setting up a deliberately permissive alternative to EU data‑sovereignty demands.
In the US, Illinois legislation backed by OpenAI aims to limit when AI labs can be held liable, while xAI is suing Colorado over its AI anti‑discrimination law, signaling diverging state‑level approaches to AI liability and fairness.
Domestic‑security agencies are adding another layer, with the NSA pushing router firmware reboots, the FCC banning some foreign routers, and the UK testing strict phone age and identity checks amid already high public distrust.
china’s AI stack as cost disrupter and political risk
Alibaba has launched a new data center built around 10,000 AI chips and is pivoting its Qwen models from open to closed as it chases direct monetization.
Shopify reports saving $5M a year after switching to Qwen for some workloads, showing that Chinese models can already undercut Western labs on price‑performance for enterprise use.
The latest Qwen3.6 Plus model offers native vision input and ranks just behind GLM‑5.1 on code and reasoning benchmarks, while Alibaba’s HappyHorse‑1.0 currently tops video‑generation rankings.
Cheap open‑weight models from China and elsewhere have also replicated many of Mythos’s vulnerability findings at a fraction of the token cost, indicating that AI‑enabled cybersecurity capability is not confined to a few frontier labs.
At the same time, tools like the Obliteratus toolkit have already stripped safety filters from Qwen 1.5B, illustrating how quickly guardrails on closed Chinese models can be removed once they are deployed.
agents, layoffs, and the new labor politics of AI
AI‑linked layoffs in tech hit nearly 80,000 roles in Q1 2026, with about half of those cuts explicitly attributed to automation and AI initiatives.
Oracle alone laid off 30,000 employees, is rehiring 8,000 as contractors at roughly 60% of prior pay without benefits, and has framed the restructuring as funding AI data centers and commitments to OpenAI.
AI agents are already managing full Meta Ads accounts for a month at a time, a 9B‑parameter data‑analyst agent can complete 89% of workflows without humans, and charts show agents significantly displacing human roles over the past year.
Yet builders estimate that roughly 90% of agents being developed are unlikely to generate revenue, even as companies like Salesforce cut thousands of roles while struggling with AI reliability.
On the workforce side, Gen Z employees are sabotaging AI rollouts out of job‑loss fears, white‑collar workers report that mandated tools don’t actually help, and younger cohorts increasingly describe AI as a threat rather than an opportunity.
What This Means
AI is consolidating into a small set of commercially powerful but politically exposed suppliers, while the infrastructure, regulatory, and labor environments around them fragment and harden.
Capital allocation into this space is increasingly less about picking individual models and more about choosing which geopolitical blocs, regulatory regimes, and workforce narratives you are willing to be tied to.
On Watch
/California’s 2027 ban on hosting most open‑weight image and video models may push open‑source AI infrastructure and research out of the state or force new legal workarounds.
/Unitree’s R1 humanoid robot launching at $5,900 on Alibaba’s AliExpress could turn capable biped robots into a consumer‑ and SMB‑grade product rather than an enterprise‑only capex item.
/Milla Jovovich’s open‑source AI memory system hitting a perfect LongMemEval score and over 30,000 GitHub stars in two days hints at rapid commoditization of long‑context infrastructure.
Interesting
/Anthropic's annual recurring revenue surged from $9 billion to $30 billion in just three months, showcasing rapid growth.
/The collaboration between OpenAI, Anthropic, and Google aims to combat model copying from Chinese competitors, indicating a strategic response to global competitive pressures.
/The Pentagon's blacklisting of Anthropic has led the company to form a consortium, indicating a strategic pivot in response to the military's actions.
/A single ungoverned automated agent caused $5.4-10 billion in losses during the 2024 CrowdStrike outage, emphasizing the need for governance of AI machine identities.
/Meta plans to pay its top AI executives nearly a billion each in bonuses, indicating a significant investment in AI leadership.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/Anthropic reached a $30B revenue run rate and expanded a long‑term next‑gen TPU partnership with Google and Broadcom.
/Claude Mythos uncovered thousands of zero‑day vulnerabilities across major operating systems and browsers and triggered the $100M Project Glasswing defensive initiative.
/France ordered ministries to migrate government PCs from Windows to Linux and move health data to European clouds by 2026.
/Texas is paying over $1B a year in tax breaks to data centers while Samsung builds a major fab there for Tesla’s chips.
/Alibaba pivoted Qwen from open to closed models, stood up a 10,000‑chip AI cluster, and delivered $5M in annual savings to Shopify.
On Watch
/California’s 2027 ban on hosting most open‑weight image and video models may push open‑source AI infrastructure and research out of the state or force new legal workarounds.
/Unitree’s R1 humanoid robot launching at $5,900 on Alibaba’s AliExpress could turn capable biped robots into a consumer‑ and SMB‑grade product rather than an enterprise‑only capex item.
/Milla Jovovich’s open‑source AI memory system hitting a perfect LongMemEval score and over 30,000 GitHub stars in two days hints at rapid commoditization of long‑context infrastructure.
Interesting
/Anthropic's annual recurring revenue surged from $9 billion to $30 billion in just three months, showcasing rapid growth.
/The collaboration between OpenAI, Anthropic, and Google aims to combat model copying from Chinese competitors, indicating a strategic response to global competitive pressures.
/The Pentagon's blacklisting of Anthropic has led the company to form a consortium, indicating a strategic pivot in response to the military's actions.
/A single ungoverned automated agent caused $5.4-10 billion in losses during the 2024 CrowdStrike outage, emphasizing the need for governance of AI machine identities.
/Meta plans to pay its top AI executives nearly a billion each in bonuses, indicating a significant investment in AI leadership.