AI economics snapped this quarter: big companies are torching budgets on premium tools while Chinese players push model prices toward zero, and Anthropic is being valued like a near‑monopoly at the top of the stack. All of it runs on hardware dangerously concentrated in Taiwan, with defense, regulators, and workers starting to clamp down on how and where AI gets used.
The only real question is how much of that concentration and political risk you’re willing to hold to stay on the frontier.
Key Events
/Anthropic raised $65B in Series H at a $965B valuation, surpassing OpenAI as the most valuable AI startup.
/Uber exhausted its entire 2026 AI budget in four months, largely on Anthropic’s Claude Code tools.
/Xiaomi permanently cut MiMo‑v2.5 API prices by up to 99% and is offering free access to new frontier AI models.
/Nvidia plans to ramp annual investment in Taiwan to about $150B, reinforcing TSMC as its core AI manufacturing hub.
/Microsoft secured a $9.7B contract with the Pentagon to modernize defense IT and AI and reduce license sprawl.
Report
AI went from 'free magic' to a line item that can blow a hole in your P&L in about six months. At the same time, ultra‑cheap Chinese models and a near‑trillion‑dollar Anthropic on the premium end are pulling AI pricing in opposite directions.
ai unit economics are breaking in two directions
Uber burned its entire 2026 AI budget in four months on Anthropic’s Claude Code and its leadership is now saying AI spend is getting “harder to justify.” Microsoft, Uber, and others report that AI coding tools are already more expensive than hiring humans, with internal Microsoft data showing AI usage can cost more than staff.
An unnamed company accidentally ran up a $500M AI tools bill in a single month because employees had no usage caps. A fully AI‑generated film that screened at Cannes spent $400k of its $500k production budget purely on AI compute.
In sharp contrast, Xiaomi has permanently cut prices for its MiMo‑v2.5 API by up to 99% and is giving free access to new frontier models, using ultra‑cheap inference as a customer acquisition play.
anthropic as near‑trillion‑dollar fulcrum
Anthropic just raised $65B at a $965B valuation, overtaking OpenAI as the most valuable AI startup after tripling its value in three months.
Reported annual revenue has already climbed from $9B to $47B in a single year. Claude Opus 4.8 now tops the GDPval‑AA benchmark for real‑world agent work with a score of 1890, 121 points above the next model.
At the same time, Anthropic is finalizing a classified NSA surveillance contract and marketing itself as the enterprise‑ and compliance‑first lab, in contrast to OpenAI’s consumer tilt.
Its Claude Opus 4.8 API is priced at $5 per million input tokens and $25 per million output tokens, and one of its largest customers has already blown through an entire year’s AI budget using its coding tools, while prospective users voice worries about sending it sensitive data.
taiwan as the fragile center of the stack
Nvidia now plans to spend about $150B per year in Taiwan, a step change from a few years ago that locks TSMC in as the geographic core of its AI roadmap.
Authorities recently busted a smuggling ring moving Nvidia GPUs out of Taiwan, while TSMC employees angry over proposed bonus cuts are openly talking about strikes and unionization.
Commentary in the US and elsewhere is increasingly questioning deep reliance on Taiwanese manufacturing given rising tensions and the risk that a Chinese move could trigger sanctions and choke off global chip supply.
On the human‑capital side, China is now restricting overseas travel for top AI talent at firms like Alibaba and DeepSeek to curb technology leakage, tightening an already‑constrained part of the ecosystem.
defense and space as durable ai buyers
Microsoft landed a $9.7B Pentagon deal to consolidate and modernize defense IT and AI, explicitly pitched as a way to cut costs and license sprawl across the Department of Defense.
SpaceX is simultaneously building core military infrastructure: the US Space Force picked Starlink to run a sensor‑to‑shooter targeting network, and NASA signed a $20B agreement for a permanent Moon base.
Starlink has already flown 50 launches in 2026 and is being installed across major US airlines’ fleets, turning a nominally civilian network into dual‑use infrastructure that US military drones have reportedly tapped against company rules.
On the model side, Anthropic is finalizing a classified NSA surveillance tools contract, while ChatGPT is seeing user churn specifically over concerns about Pentagon ties and privacy.
Taken together with reports that AI is rapidly accelerating new weapons development in China, this positions defense and security as one of the few AI demand pools with both budget and political tailwinds.
regulators and workers are starting to bite back
Spain has flat‑out blocked prediction markets Polymarket and Kalshi on licensing grounds, Illinois now forces AI labs through independent audits, and California is mandating gun‑blocking software in 3D printers and banning AI chatbot toys for kids.
The EU AI Act adds binding cybersecurity obligations for autonomous agents, while Pope Leo XIV’s 42,300‑word encyclical calls for disarmament of AI and warns that opaque, monopolistic algorithms risk dehumanizing society.
In the labor market, surveys show 99% of executives expect AI‑driven layoffs within two years, even as evidence mounts that AI hiring tools are disproportionately rejecting Black and Asian candidates.
The first US ride‑share union has formed, TSMC workers are angry enough over bonuses to talk about strikes, and California’s governor has signed an order to shield workers from AI job losses.
US law enforcement is now warning about rising anti‑tech extremism as resentment over AI, layoffs, and surveillance spreads.
What This Means
AI is bifurcating into a costly, concentrated premium stack and an ultra‑cheap commodity layer, all sitting on a geopolitically fragile hardware base and increasingly entangled with defense and politics. The live decision is how much concentration and political risk to accept in exchange for frontier capabilities versus pushing more of your stack into cheaper, less dependent territory.
On Watch
/Apple is lining up a 2027 launch of smart glasses and camera‑equipped AirPods while Chinese police already use smart glasses, hinting at a looming shift in mainstream interfaces.
/Tesla is building a Texas factory targeting production of up to 10 million Optimus humanoid robots per year as EngineAI and others claim tens of thousands of humanoids annually, signaling a serious bid to industrialize general‑purpose robots.
/IBM’s $2BAnderon quantum chip foundry and a $5B IBM/Red Hat open‑source AI commitment point toward an emerging alternative compute and software stack outside the Nvidia‑plus‑hyperscaler axis.
Interesting
/GPU rental prices have doubled in the last four months, reflecting the high demand for AI compute resources.
/Power bills near data centers can exceed 250% of normal rates, raising concerns for local economies.
/Larry Fink's proposal to confiscate savings and pensions for AI infrastructure funding has sparked debate about the future of personal finance.
/The AI industry is expected to see many startups fail within two years due to high operational costs, favoring established players like OpenAI and Anthropic.
/China's development of a futures market for AI tokens aims to rival US exchanges, indicating a competitive shift in the global AI landscape.
We processed 10,000+ comments and posts to generate this report.
AI-generated content. Verify critical information independently.
/Anthropic raised $65B in Series H at a $965B valuation, surpassing OpenAI as the most valuable AI startup.
/Uber exhausted its entire 2026 AI budget in four months, largely on Anthropic’s Claude Code tools.
/Xiaomi permanently cut MiMo‑v2.5 API prices by up to 99% and is offering free access to new frontier AI models.
/Nvidia plans to ramp annual investment in Taiwan to about $150B, reinforcing TSMC as its core AI manufacturing hub.
/Microsoft secured a $9.7B contract with the Pentagon to modernize defense IT and AI and reduce license sprawl.
On Watch
/Apple is lining up a 2027 launch of smart glasses and camera‑equipped AirPods while Chinese police already use smart glasses, hinting at a looming shift in mainstream interfaces.
/Tesla is building a Texas factory targeting production of up to 10 million Optimus humanoid robots per year as EngineAI and others claim tens of thousands of humanoids annually, signaling a serious bid to industrialize general‑purpose robots.
/IBM’s $2BAnderon quantum chip foundry and a $5B IBM/Red Hat open‑source AI commitment point toward an emerging alternative compute and software stack outside the Nvidia‑plus‑hyperscaler axis.
Interesting
/GPU rental prices have doubled in the last four months, reflecting the high demand for AI compute resources.
/Power bills near data centers can exceed 250% of normal rates, raising concerns for local economies.
/Larry Fink's proposal to confiscate savings and pensions for AI infrastructure funding has sparked debate about the future of personal finance.
/The AI industry is expected to see many startups fail within two years due to high operational costs, favoring established players like OpenAI and Anthropic.
/China's development of a futures market for AI tokens aims to rival US exchanges, indicating a competitive shift in the global AI landscape.