AI just split into two camps: state‑aligned stacks like OpenAI/xAI/Palantir that are all‑in on defense work, and ethics‑branded stacks like Anthropic that are getting blacklisted at home and rewarded abroad. At the same time, $110B of capital locked OpenAI into quasi‑utility status even as users uninstall ChatGPT and push Claude to #1, while regulators quietly wire surveillance and identity checks into the OS and network layers.
The money, the state, and the stack are converging; where you plug into that system is now a political and financial choice, not a purely technical one.
Key Events
/OpenAI raised $110B in a single funding round from Amazon, Nvidia, and SoftBank.
/OpenAI agreed to deploy its models on the U.S. Department of War’s classified network and signed a roughly $200M Pentagon contract.
/Anthropic was designated a U.S. 'supply chain risk' and banned from federal systems after refusing Pentagon demands on surveillance and autonomous weapons.
/The 'Cancel ChatGPT' backlash drove a 295% spike in ChatGPT app uninstalls after OpenAI’s Pentagon deal.
/Anthropic’s Claude app hit #1 on the U.S. App Store as users defected from ChatGPT.
Report
Power around AI just concentrated in a single stack while the U.S. government started playing Huawei games with its own vendors. The real risk this quarter isn’t 'is AI real'—it’s which side of that split your dependencies sit on.
state‑aligned vs ethics‑aligned ai
Anthropic flatly refused a major Pentagon contract that required dropping safeguards against mass surveillance and autonomous weapons, and in response was labeled a U.S. 'supply chain risk' and banned from federal systems.
The company’s custom Claude model for the Pentagon, reportedly 1–2 generations ahead of the consumer version, is already running on a classified cloud yet is now politically radioactive.
OpenAI went the other way, signing a deal to deploy its models on the Department of War’s classified networks and then inking a roughly $200M Pentagon contract within hours of Anthropic’s ban.
In parallel, xAI agreed to Pentagon demands for essentially unrestricted military use of its models, while Palantir’s CEO is openly talking about AI being nationalized, putting a clear 'state‑aligned' camp on one side of the board.
capital gravity around openai
Against that backdrop, OpenAI closed a $110B round, the largest private tech raise on record. That deal values the company around $730B pre‑money, putting it in mega‑cap territory before an IPO.
Amazon alone is wiring in $50B, reportedly contingent on either an IPO or an AGI milestone within a year, while Nvidia and SoftBank are in for additional tens of billions combined.
Hyperscaler AI capex is forecast to hit $770B in 2026. OpenAI’s own models project roughly $157B in cumulative cash burn through 2028, so this isn’t growth capital, it’s utility‑scale infrastructure finance.
Nvidia’s CEO is already calling his big OpenAI commitment 'maybe the last' such check, underscoring how dependent the rest of the ecosystem is on this one stack working economically.
user backlash and claude’s run
After the Pentagon deal landed, 'Cancel ChatGPT' drove a 295% spike in ChatGPT app uninstalls and a wave of Plus subscription cancellations.
One activist campaign claims 2.5M signatures to 'QuitGPT', and ChatGPT’s one‑star reviews jumped 775%, showing how fast brand sentiment can flip when users think their data is feeding the surveillance state.
Anthropic rode that wave: Claude’s app jumped from #129 to #1 on the U.S. App Store as users defected from ChatGPT in support of its stance.
The company is nearing a $20B revenue run rate, up roughly $5B in a few weeks. Around 80% of its enterprise demand is now outside the U.S., so the 'ethics‑first' posture is monetizing strongest in markets least impressed by American defense partnerships.
regulation and surveillance moving up the stack
California just made age verification an OS‑level responsibility: every operating system, including Linux and SteamOS, will have to collect birth date or age bracket at account setup starting in 2027.
Brazil has passed a similar law taking effect in March, and Colorado and New York are exploring related rules, exactly the shift Mark Zuckerberg has lobbied for to move verification duty from apps to OSes and app stores.
At the same time, OpenAI’s Department of War contract bans domestic mass surveillance in name but lets the government use its models for 'all lawful purposes', while the Pentagon is already buying bulk 'anonymized' geolocation data on Americans and insisting that isn’t spying.
Consumer and workplace products are normalizing the same pattern: Meta has sold about 7M AI glasses amid reports of intimate videos being reviewed by human moderators, and Burger King is putting AI headsets on staff to score 'friendliness' at the drive‑through.
labor, margins, and fragile infrastructure
Jack Dorsey’s Block cut about 4,000 staff, taking headcount to just under 6,000. That pushed gross profit per employee toward $2M, comparable to Nvidia, and the stock jumped about 24% after the announcement.
A European bank, Morgan Stanley, and Mizuho are all tying layoffs or workload cuts for thousands of roles directly to AI investment or automation.
Yet 56% of CEOs report zero financial return from AI by 2026, and Goldman Sachs estimates AI added 'basically zero' to U.S. GDP last year, so the cost‑cut story is outrunning hard productivity data.
Meanwhile, the physical layer is wobbling: Iranian drones hit Amazon data centers in the UAE, and a separate data‑center fire there disrupted global AI services in the middle of a regional shooting war.
DDR5 memory is being scalped for AI data‑center builds, smartphone shipments are forecast to drop sharply on memory shortages, and entry‑level PCs are expected to vanish by 2028 as component prices stay elevated.
What This Means
AI is no longer a generic 'tech theme' but a polarized political and capital stack where your vendors, regions, and workforce plans quietly declare which side of a state‑vs‑society fight you’ve joined. The gap between the money pouring into that stack and the measurable productivity coming out of it is widening fast.
On Watch
/Alibaba’s Qwen 3.5 stack is closing the quality gap with frontier models for local use (e.g., a 122B‑parameter model performing near Sonnet 4.5) just as key staff depart and signals point to a pivot away from fully open releases.
/Cisco’s SD‑WAN zero‑day (CVE‑2026‑20127), already exploited since at least 2023, lets attackers remotely tamper with network configs on federal and global networks, making AI‑augmented offense against critical infra a realistic near‑term threat.
/Google ending its 30% app‑store fee and welcoming third‑party app stores cracks open mobile distribution economics right as AI agents and app‑like frontends proliferate, potentially reshaping who owns the customer relationship.
Interesting
/China's electricity generation now surpasses that of the US and EU combined by 40%, indicating a shift in global energy dynamics.
/Nvidia and Apple are offering six-figure salaries to attract talent in HBM and NAND sectors amid an AI memory boom.
/The alarming statistic that 86% of LLM apps are vulnerable to prompt injection underscores the urgent need for improved security measures.
/Reports indicate that AI models recommended nuclear strikes in 95% of war game scenarios, raising alarms about AI's role in military decision-making.
/Legal concerns about AI training data liability are becoming a major barrier to adoption proposals in the tech industry.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/OpenAI raised $110B in a single funding round from Amazon, Nvidia, and SoftBank.
/OpenAI agreed to deploy its models on the U.S. Department of War’s classified network and signed a roughly $200M Pentagon contract.
/Anthropic was designated a U.S. 'supply chain risk' and banned from federal systems after refusing Pentagon demands on surveillance and autonomous weapons.
/The 'Cancel ChatGPT' backlash drove a 295% spike in ChatGPT app uninstalls after OpenAI’s Pentagon deal.
/Anthropic’s Claude app hit #1 on the U.S. App Store as users defected from ChatGPT.
On Watch
/Alibaba’s Qwen 3.5 stack is closing the quality gap with frontier models for local use (e.g., a 122B‑parameter model performing near Sonnet 4.5) just as key staff depart and signals point to a pivot away from fully open releases.
/Cisco’s SD‑WAN zero‑day (CVE‑2026‑20127), already exploited since at least 2023, lets attackers remotely tamper with network configs on federal and global networks, making AI‑augmented offense against critical infra a realistic near‑term threat.
/Google ending its 30% app‑store fee and welcoming third‑party app stores cracks open mobile distribution economics right as AI agents and app‑like frontends proliferate, potentially reshaping who owns the customer relationship.
Interesting
/China's electricity generation now surpasses that of the US and EU combined by 40%, indicating a shift in global energy dynamics.
/Nvidia and Apple are offering six-figure salaries to attract talent in HBM and NAND sectors amid an AI memory boom.
/The alarming statistic that 86% of LLM apps are vulnerable to prompt injection underscores the urgent need for improved security measures.
/Reports indicate that AI models recommended nuclear strikes in 95% of war game scenarios, raising alarms about AI's role in military decision-making.
/Legal concerns about AI training data liability are becoming a major barrier to adoption proposals in the tech industry.