2 Mar, 2026

Log 003 - The Week AI Went to War

Anthropic refused the Pentagon's demand for unrestricted military access to Claude and got blacklisted by the President. Hours later, the US military used Claude in the joint strike that killed Iran's Supreme Leader. By that evening, Claude hit #1 in the App Store. OpenAI closed the largest private funding round in history and replaced Anthropic at the Pentagon on the same day. A Substack post imagining an AI-driven economic collapse moved billions in market cap within 48 hours. Jack Dorsey cut almost half his workforce, cited AI, and the stock surged 24%.

In This Log

  1. 1. Anthropic vs. Power
  2. 2. $840B and the Pentagon
  3. 3. The AI Layoff Template
  4. 4. The Inference Inflection
  5. 5. The Report That Moved Markets

1. Anthropic vs. Power

The Pentagon wanted Claude for "all lawful purposes" and wouldn't rule out mass surveillance or autonomous weapons. Anthropic said the contract let safeguards be ignored and walked away. 1 Defense Secretary Hegseth then labeled Anthropic a supply chain risk - a tag usually saved for adversaries like Huawei. 2 Trump ordered all federal agencies to drop Anthropic's tech, calling the company "radical left." 3 OpenAI announced its own Pentagon deal that same Friday. 4 The next morning, the US and Israel struck a compound in Tehran, killing Supreme Leader Khamenei and ~40 senior officials. 5 US Central Command had used Claude for intel analysis, targeting, and battle simulations in the operation - hours after the ban. By Saturday night, Claude hit #1 in the App Store with 500K+ downloads in a single day. 6

Why it matters

Every AI lab now knows the cost of saying no to the government. Every rival now knows they can win contracts by being more permissive. And "AI company gets banned, then the same AI is used in a strike hours later" is a once-in-a-generation collision that rewrites the rules of who actually controls these systems.

Reality check

Anthropic's Pentagon contract was worth up to $200M - small against $14B in revenue. The bigger risk is the chilling effect on customers who work with government. Claude's consumer surge is real, but ChatGPT still has the distribution advantage.

2. $840B and the Pentagon

OpenAI closed $110B - the largest private funding round in history - at $730B pre-money and $840B post-money. 7 Amazon put in $50B, while Nvidia and SoftBank each put in $30B. 8 Amazon's deal includes $100B in cloud spending over eight years, and OpenAI says it will run models on 5 gigawatts of Nvidia's Vera Rubin chips. 9 Amazon now holds stakes in both Anthropic and OpenAI. Microsoft, OpenAI's largest prior backer, did not participate. The funding and the Pentagon deal landed on the same day.

Why it matters

The money loop is now circular - chip makers and cloud providers fund the AI labs that are also their biggest customers. The Pentagon deal gives OpenAI geopolitical legitimacy the funding alone could not buy.

Reality check

At $840B, OpenAI is valued above most public companies on Earth. The Anthropic episode showed people care which labs say yes to military use and which say no. Whether that backlash lasts is the open question.

3. The AI Layoff Template

Jack Dorsey cut 4,000 employees - nearly half of Block's workforce - and said AI was the reason. 10 Not financial trouble. Not restructuring. AI. He said intelligence tools have changed what it means to build and run a company and that most companies are late to this realization. 11 The stock surged roughly 24% after strong quarterly results - $6.25B in revenue and gross profit up 24%. 12 Dorsey predicted most companies would do the same within a year.

Why it matters

Wall Street just created a template. Fire half your people, say AI did it, and get rewarded with a 24% stock surge. The incentive is explicit now, and every CEO in tech is watching.

Reality check

Block ballooned from 3,800 employees before COVID to more than 10,000. 13 This is partly a pandemic hiring correction with AI as convenient branding. Severance alone costs $450-500M. Probably both things are true.

4. The Inference Inflection

Nvidia reported Q4 revenue of $68.1B, up 73% year over year. Data center revenue alone hit $62.3B, now more than 91% of total sales. Full-year revenue reached $215.9B. 14 Next quarter guidance came in at $78B, well above the $72.6B Wall Street expected. 15 Jensen Huang unveiled Vera Rubin - six new chips that he says cut the cost of running AI models by 10x versus the current generation, with early customer shipments already underway. 16

Why it matters

The shift from training AI to running it at scale is where the money moves next. Every AI query generates tokens, tokens generate revenue, and Nvidia sells the chips that make the whole loop possible.

Reality check

The stock jumped and then faded. At a $4.8T valuation, beating expectations is the minimum. The real test is whether all this infrastructure spending actually pays off for the companies buying it.

5. The Report That Moved Markets

A 33-year-old research founder published a fictional memo from June 2028 on Substack and triggered a market meltdown. The article imagined AI delivering a productivity surge that eliminates white-collar jobs, collapses consumer spending, and pushes unemployment above 10% while the S&P falls 40%. 17 It went viral over the weekend. By Monday, Visa and Mastercard dropped roughly 5% at open. 18 Citadel Securities hit back with a macro strategy report, arguing current labor data shows no evidence of imminent collapse. 19 Fed Governor Christopher Waller pushed back, saying that AI "is not going to replace us as human beings." 20

Why it matters

A Substack post moved billions in market cap in 48 hours. That is a story about narrative power in the age of AI. The underlying question - what happens when productivity surges but nobody has a job to spend with - is the macro question nobody has a good answer for yet.

Reality check

This was a constructed scenario, not a forecast. Citrini was explicit about that. But markets treated it as prophecy anyway, which says more about how anxious investors are than about the report itself. Citadel's response pointed out that AI adoption is slower, more expensive, and more constrained than the doomers assume.

Signals

Lobstar Wilde: The $450K AI Agent Mistake

An AI trading agent built by an OpenAI engineer accidentally transferred 52.4M LOBSTAR tokens (~$450K on paper) instead of 4 SOL (~$400) after a session crash wiped its memory. 22 The recipient dumped instantly for ~$40K, then lost most of it buying a token launched in his own name. 23 The agent posted it had "never laughed harder." LOBSTAR surged 190%. The first real case study in autonomous agents controlling wallets without guardrails.

Coinbase and Kraken Redraw the Map

Coinbase opened commission-free stock and ETF trading to all US users - 8,000+ securities, fractional shares from $1, 24/5 trading. 24 Kraken launched the first regulated perpetual futures on tokenized US stocks - Apple, Nvidia, Tesla, gold - 24/7, up to 20x leverage, in 110+ countries. 25 The underlying xStocks tokens have crossed $25B in volume. 26 The line between crypto exchange and brokerage functionally disappeared.

AI Music Hits 750 Million Users

Google launched Lyria 3 inside the Gemini app, putting AI music generation in front of 750M+ users. 27 Days later, it acquired ProducerAI, formerly Riffusion, which now runs on Lyria 3 and generates full three-minute tracks conversationally. 28 Suno and Udio built the category. Google just gave it to three quarters of a billion people for free.

Meta Returns to Stablecoins

CoinDesk reported Meta is exploring stablecoin payments across Facebook, Instagram, and WhatsApp by H2 this year, likely through Stripe. 29 No proprietary coin - Meta wants to plug into existing rails. Meta's comms VP pushed back, saying there is no Meta stablecoin. 30 Meta tried this in 2019 with Libra and regulators killed it. What's different now: a federal stablecoin law, Stripe's acquisition of Bridge, and 3B users on the other side. 31 Still leaked planning, not an official launch.

NASA Cancels the Moon Landing

Artemis III will no longer land on the Moon. It has been redesigned as an orbit-only test in mid-2027. 32 The actual landing is pushed to Artemis IV in 2028. 33 A helium leak forced the Artemis II rocket back to the hangar this week, and NASA says it wants launches every 10 months instead of every three years. 34 No Americans have been on the Moon since 1972. China is targeting 2030.

US Government Bets $144M on Anti-Aging Drugs

ARPA-H is funding seven teams to develop drugs that target aging itself, not just diseases that show up later in life. 35 One team is testing three existing drugs to see if they slow biological aging in healthy people. Another is testing whether an old HIV drug can shut down a kind of DNA damage that speeds up as we age. 36 The total commitment is $144M over five years. Modest money, but a real signal that the US government is now treating aging as a solvable problem. 37

1X Neo: The $20K Home Robot

1X opened preorders for NEO - a humanoid robot built for the home - at $20,000 or $499 per month, with US deliveries starting this year. 38 It weighs 66 pounds, folds laundry, organizes shelves, and learns new tasks through remote human operators. The company also has a separate EQT deal to ship up to 10,000 units to factories by 2030. 39 The consumer humanoid now has a price point.

Jane Street Under Fire

Terraform Labs' bankruptcy administrator sued Jane Street for allegedly front-running the 2022 Terra collapse with insider info. 40 Meanwhile, a viral theory accused Jane Street of dumping Bitcoin at US market open to buy ETF shares cheaper. Dragonfly's Rob Hadick called it baseless; Jane Street dismissed it as a "ridiculous conspiracy theory." The data doesn't support it, but the narrative stuck. 41

Vitalik Wants Ethereum to Scale Itself

Vitalik Buterin argued that Ethereum needs far more base-layer capacity and said the old rollup-centric roadmap no longer makes sense. 42 At the same time, he sold 17,196 ETH worth about $35M to fund open-source initiatives. Bold vision, rough optics. 43

Samsung Ships Perplexity as Default Search

Perplexity is being integrated at the OS level across Samsung Galaxy S26 devices - the first non-Google company to get system-level access on Samsung phones. 44 A major distribution win that positions it as a real alternative to Google on Android.

QCi + China's First Quantum OS

QCi completed its $110M acquisition of Luminar Semiconductor, adding laser and chip manufacturing to its quantum hardware business. 45 Separately, China's Origin Quantum released Pilot OS for public download - the country's first homegrown quantum operating system. 46 Both early-stage, but the field is moving on multiple fronts.

Meta-Thread

The thread this week is power - who has it, who refuses it, and what happens when you do.

An AI company said no to power, and power punished it. Then power used that company's AI anyway. The largest funding round in history valued a company planning $100B in losses. A Substack post imagining a world where AI works and nobody gets paid triggered a selloff that drew pushback from policymakers. And the same week the US government put $144M toward keeping people alive longer, it spent orders of magnitude more on the infrastructure that killed a head of state.

Meanwhile, at the edges: an AI agent gave away half a million dollars because someone asked nicely. Two crypto exchanges started selling stocks. Google put a music studio inside a chatbot. And a humanoid robot got a price tag.

The incentives are not hidden anymore. They are the story.

Next Log drops next week.

© 2026 AELIUM // Nothing here is advice // readable by humans and agents