The easy AI money phase is ending: token bills, GPU costs, and political pushback are now showing up in real P&Ls and cancelled projects. Anthropic and Nvidia are sprinting ahead on premium models and chips just as cheaper open and China-adjacent options emerge, forcing an explicit choice between cost, control, and geopolitical risk.
Any big AI bet you make this year is de facto a bet on whose economics and regulators you’re willing to live with.
Key Events
/Anthropic raised $65B in Series H at a post-money valuation of about $965B.
/Uber burned through its entire 2026 AI budget in four months as AI costs outpaced human labor.
/Nvidia plans to invest roughly $150B per year in Taiwan while authorities there raided 12 sites in a crackdown on Nvidia AI chip smuggling.
/A small Michigan town blocked OpenAI and Oracle from building a $16B data center facility.
/DeepSeek made its 75% V4 Pro API price cut permanent and is seeking $10.29B to fund open-source AI models.
Report
The core AI story this quarter is that the unit economics are uglier than advertised, just as infrastructure and politics are getting more complicated.
Where you place chips now is less about who has the smartest model and more about whose costs, supply chains, and regulators you trust.
aI economics are breaking the hype narrative
Uber burned its entire 2026 AI budget in four months, and its COO says it is getting harder to justify AI spend because many projects are not producing useful features.
At Microsoft, internal Claude Code licenses are being cancelled and some engineers are banned from external AI tools after token-based billing made AI work more expensive than human employees.
One unnamed firm reportedly racked up about $500M in costs from unregulated use of Anthropic tools, underscoring how quickly experimentation can turn into a balance-sheet problem.
GPU rental prices have doubled in the last four months, and memory now accounts for nearly two-thirds of AI chip component costs, so infra inflation is colliding with these runaway token bills.
Meanwhile roughly 80% of AI projects are said to be failing to meet expectations, and many executives report AI labor often costs more than human workers instead of less.
premium models vs discounted and open rivals
Anthropic raised $65B in Series H at a post-money valuation of about $965B and is expected to turn a profit this quarter on $10.9B in Q2 revenue.
Its annualized revenue jumped from $87M in January 2024 to $44B by late April 2026, making it the most valuable AI startup and putting talk of a path to a $2T valuation into circulation.
Claude Opus 4.8 now leads benchmarks like GDPval-AA and the Artificial Analysis Intelligence Index for agentic real-world tasks, surpassing GPT-5.5 and other frontier models.
On the other side of the barbell, DeepSeek made a 75% promotional price cut on its V4 Pro API permanent and is raising $10.29B for a large open-source AI effort, while Xiaomi cut some MiMo-v2.5 API prices by up to 99%.
DeepSeek’s V4 Pro is about 11.5x cheaper than OpenAI’s GPT-5.5 for comparable workloads, and users are already questioning whether high-priced token plans at labs like Anthropic and OpenAI are sustainable without subsidies and quotas.
chips, memory, and taiwan–china risk
Nvidia plans to invest around $150B per year in Taiwan and calls it the “epicentre” of the AI revolution, while Taiwanese authorities just raided 12 locations in a first major crackdown on Nvidia AI chip smuggling.
TSMC’s market cap has exceeded $2T and helped push Taiwan into the world’s fifth-largest stock market, with AI demand driving a 9.5% GDP growth forecast for 2026.
China is ramping production of DRAM and NAND, pushing into DDR5 and aiming to capture mainstream memory markets, which is expected to drive down prices even as China-linked supply becomes more central to AI workloads.
Huawei has built a 122TB SSD using packaging to bypass US 3D NAND sanctions and supplied 2PB of flash storage for Norway’s LLM training, while projecting transistor densities comparable to 1.4nm processes by 2031.
US and European buyers are wary of Chinese-origin AI technologies, but China’s AI startup funding still tripled to $16B in Q1 with strong pushes into LLMs and humanoid robotics, tightening the bind between cost, capacity, and geopolitics.
regulation, sovereignty, and social license
A small Michigan town blocked OpenAI and Oracle from building a $16B data center, and the Netherlands stopped a US takeover of DigiD provider Solvinity on cybersecurity and sovereignty grounds while seizing 800 servers from a hostile hosting firm.
Illinois now requires independent audits of AI models for safety, and Spain has blocked prediction markets Polymarket and Kalshi for operating without gambling licenses.
Pennsylvania sued an AI company whose chatbots allegedly misrepresented themselves as licensed doctors, while YouTube will automatically detect and label AI-generated videos more prominently.
Lawmakers are floating sharper tools including Senator Warren’s proposed AI tax, MPs demanding an AI “kill switch”, California’s executive order to address AI job displacement, and Delaware rulings granting corporations voting rights in some local elections.
Pope Leo XIV’s Magnifica Humanitas encyclical explicitly warns that opaque AI controlled by a few firms risks dehumanization and calls for AI “disarmament”, giving moral cover to stricter oversight.
labor, talent, and the ai layoff narrative
Surveys show 99% of CEOs and executives expect AI-driven layoffs within two years, and tech layoffs have already surpassed 100,000 in 2026 with over 142,000 people cut at firms like Meta, LinkedIn and Cisco.
Many of those layoffs are described as reallocating budgets to fund AI initiatives, even as an MIT report and Goldman Sachs say AI is not the primary driver of most recent tech job cuts.
China has explicitly told companies not to use AI as a justification for layoffs, reflecting concern about youth unemployment and social stability.
Despite apocalyptic rhetoric about white-collar work, US software-developer employment is still rising and software-engineer job postings are climbing, driven by more code and GitHub activity.
At the same time FAANG hiring has slowed, their resume value is being questioned, California’s share of local big-tech employment is falling, and solo founders now account for 63% of new C-corps, shifting tech talent into more fluid and fragmented channels.
What This Means
AI is now as much about margins, supply chains, and political legitimacy as model quality, so every large AI bet is implicitly a macro risk position.
On Watch
/Mount is now insuring autonomous AI agents against workflow risks, hinting at a new insurance product class around AI-driven operations.
/Millions of AI agents are exposed by a vulnerability in an open-source package, with 245,000 OpenClaw instances on the public internet and over 30,000 actively compromised.
/Starlink’s V3 satellites aim to match the bandwidth of all current services within a year while the EU is already exploring ways to limit its expansion.
Interesting
/The CLOUD Act complicates international data transactions, leading to security concerns among nations like the Netherlands.
/Andrej Karpathy's transition from OpenAI to Anthropic has sparked discussions about the internal dynamics at OpenAI and its competitive stance.
/Microsoft's relationship with OpenAI is reportedly strained, leading them to consider developing their own AI models.
/The Chinese government is actively preventing the overseas travel of AI talent to safeguard its technological advancements.
/Concerns are rising over China's influence on anti-AI sentiment in the West, with allegations of funding political entities to shape narratives.
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 post-money valuation of about $965B.
/Uber burned through its entire 2026 AI budget in four months as AI costs outpaced human labor.
/Nvidia plans to invest roughly $150B per year in Taiwan while authorities there raided 12 sites in a crackdown on Nvidia AI chip smuggling.
/A small Michigan town blocked OpenAI and Oracle from building a $16B data center facility.
/DeepSeek made its 75% V4 Pro API price cut permanent and is seeking $10.29B to fund open-source AI models.
On Watch
/Mount is now insuring autonomous AI agents against workflow risks, hinting at a new insurance product class around AI-driven operations.
/Millions of AI agents are exposed by a vulnerability in an open-source package, with 245,000 OpenClaw instances on the public internet and over 30,000 actively compromised.
/Starlink’s V3 satellites aim to match the bandwidth of all current services within a year while the EU is already exploring ways to limit its expansion.
Interesting
/The CLOUD Act complicates international data transactions, leading to security concerns among nations like the Netherlands.
/Andrej Karpathy's transition from OpenAI to Anthropic has sparked discussions about the internal dynamics at OpenAI and its competitive stance.
/Microsoft's relationship with OpenAI is reportedly strained, leading them to consider developing their own AI models.
/The Chinese government is actively preventing the overseas travel of AI talent to safeguard its technological advancements.
/Concerns are rising over China's influence on anti-AI sentiment in the West, with allegations of funding political entities to shape narratives.