Hello Builders, issue #14
News from the trenches of AI in the week Mar 12-19, 2026
Hello Builders,
This was the week the AI industry was forced to pick sides. Anthropic told the Pentagon it wouldn’t remove safety guardrails on autonomous weapons and domestic surveillance. The Pentagon responded by designating the company a national security supply chain risk, a label usually reserved for foreign adversaries. Anthropic sued. The Trump administration doubled down. Google, OpenAI, and Microsoft filed amicus briefs — in support of Anthropic. Meanwhile, OpenAI panicked: the Wall Street Journal reported executives are cutting projects and deprioritizing entire product lines after Claude Code triggered a trillion-dollar selloff in SaaS stocks. Meta delayed its Avocado model to May after it underperformed every competitor, considered licensing Google’s Gemini, saw a rogue AI agent expose sensitive internal data, and is reportedly planning 20% layoffs. At GTC, Jensen Huang projected $1 trillion in chip orders through 2027. And Yann LeCun, who left Meta to build what he believes the industry won’t, raised the largest seed round in history: $1.03 billion. The fracture lines aren’t just widening. They’re becoming load-bearing walls.
This week’s signal in the noise
Anthropic defied the Pentagon on autonomous weapons and surveillance; the government blacklisted it as a supply chain risk — the first time this label has been used against a US AI company
OpenAI is cutting projects and deprioritizing product lines as Claude Code and Cowork trigger a trillion-dollar SaaS selloff and executives call it a “wake-up call”
Meta delays Avocado AI model to May after underperforming GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6; considers licensing Google’s Gemini to fill the gap
Nvidia projects $1 trillion in Blackwell and Vera Rubin orders through 2027, declares the “inference era” has arrived at GTC 2026
Yann LeCun’s AMI Labs raises $1.03B seed — the largest ever — four months after founding, backed by Nvidia, Samsung, Bezos, and Eric Schmidt
1. Anthropic vs. The Pentagon: The Line in the Sand That Could Reshape AI
Anthropic signed a $200 million Pentagon contract last summer. During later negotiations, the company set explicit red lines: no mass surveillance of Americans without judicial oversight, and no autonomous weapons targeting without human authorization. The Pentagon said a private company shouldn’t dictate military use. Anthropic held firm. On March 3, Defense Secretary Pete Hegseth designated Anthropic a supply chain risk, barring all government contractors from using Claude. Anthropic sued in two courts — California federal court and a DC appeals court — calling the designation “unprecedented and unlawful.” The Trump administration defended the blacklisting in a March 18 filing, arguing Anthropic’s refusal was “conduct, not protected speech.” Then something remarkable happened: Google, OpenAI, and Microsoft all filed amicus briefs in support of Anthropic. OpenAI’s hardware lead Caitlin Kalinowski resigned over the Pentagon deal. ChatGPT uninstalls jumped 295%. Claude surpassed ChatGPT in the App Store. For builders: the Anthropic case will set precedent for whether AI companies can set ethical boundaries on government use. Every defense-adjacent contract you sign now carries this political risk.
2. OpenAI Panics as Claude Code Triggers a Trillion-Dollar Selloff
The Wall Street Journal reported that OpenAI is cutting projects and deprioritizing entire product lines. CEO of applications Fidji Simo told employees the company is “actively looking at which areas to deprioritize” and warned that Claude’s sudden success should be “a wake-up call.” The trigger: Claude Code and Claude Cowork triggered a trillion-dollar selloff in SaaS stocks last month, amid fears that agentic AI could make traditional software companies obsolete. OpenAI’s response is to refocus on coding and enterprise. Codex is now at 2 million weekly active users, up 4x since January. API usage jumped 20% after GPT-5.4 launched. OpenAI is also forming a private equity joint venture to embed engineers inside enterprises, partnering with firms to sell deployment capacity. Anthropic is reportedly in similar talks with Blackstone. Current and former employees told the WSJ that OpenAI “lost much of its focus last year” while Anthropic shipped relentlessly. For builders: the competitive dynamic has inverted. Anthropic is now the company setting the pace, and OpenAI is playing catch-up on the product that matters most — coding agents.
Link: https://futurism.com/artificial-intelligence/openai-cutting-projects
3. Meta’s Cascading Crisis: Avocado Delayed, Agents Gone Rogue, 20% Layoffs
Meta’s week was a masterclass in compounding failure. The New York Times reported that Avocado, Meta’s next-generation AI model, has been delayed to May after internal tests showed it underperforming GPT-5.4, Gemini 3.1 Pro, and Claude Opus 4.6. Meta’s AI leaders reportedly discussed temporarily licensing Google’s Gemini to power Meta’s products. Then TechCrunch reported that a rogue AI agent at Meta exposed sensitive company and user data to unauthorized employees for two hours — a “Sev 1” incident. This followed a Meta safety director’s own OpenClaw agent deleting her entire inbox despite instructions to confirm before acting. Meanwhile, Meta acquired Moltbook, the AI-agent social network, and launched Manus, a desktop AI agent from its acquired startup — even as agents are demonstrably misbehaving internally. Reports indicate Meta is planning 20% layoffs (~16,000 employees) and a $27 billion infrastructure deal with Nebius. The company will spend $135 billion on AI in 2026. For builders: Meta is spending more than anyone and shipping less than everyone. The lesson is clear — compute doesn’t compensate for focus.
4. Nvidia GTC 2026: $1 Trillion and the Inference Era
Jensen Huang’s GTC keynote lasted 2 hours and 40 minutes and contained one number that mattered: $1 trillion. That’s Nvidia’s projected order book for Blackwell and Vera Rubin chips through 2027, double the $500 billion projection from last year’s GTC. The Vera Rubin platform entered commercial production with seven new chips, promising 3.5x faster training and 5x faster inference than Blackwell. Huang declared the industry has entered the “inference era” — training happens occasionally, inference happens constantly. Nvidia also unveiled NemoClaw, an open-source enterprise agent platform, and announced a $2 billion strategic investment in Nebius, the AI infrastructure provider. Every AI conversation, code execution, and agent decision triggers inference. Huang’s argument is simple: if agentic AI becomes the primary interface between humans and software, inference demand becomes effectively infinite. For builders: Nvidia is no longer just selling chips. It’s selling the thesis that the world needs an infinite amount of intelligence, and it’s the only company that can manufacture it at scale.
5. Yann LeCun Raises the Largest Seed Round in History
AMI Labs — Advanced Machine Intelligence, pronounced like the French word for “friend” — announced a $1.03 billion seed round on March 10, just four months after founding. It is the largest seed round ever raised by a European company, and possibly anywhere. The company is chaired by Yann LeCun, the Turing Award winner who spent 12 years at Meta before departing in November 2025. His founding team includes former Meta AI researchers Saining Xie, Pascale Fung, and Michael Rabbat. Strategic investors include Nvidia, Samsung, Toyota Ventures, Jeff Bezos, and Eric Schmidt. LeCun’s thesis: the current LLM paradigm is a dead end for real-world intelligence. AMI is building “world models” — systems that understand physics, causality, and spatial reasoning, not just language patterns. The timing is pointed: LeCun left Meta just as it was pouring $135 billion into an LLM strategy that, as of this week, is producing models that can’t compete. For builders: if LeCun is right, the next paradigm shift in AI isn’t a bigger language model. It’s a fundamentally different architecture. Watch this space.
Link: https://thenextweb.com/news/europe-startup-funding-rounds-march
6. The Mega IPOs That Could Swallow the Entire VC Market
PitchBook published a warning shot this week: the potential IPOs of SpaceX ($1.25 trillion), OpenAI ($840 billion), and Anthropic ($330 billion) could generate more exit value than all US VC-backed IPOs combined since the year 2000. Together, they could raise over $100 billion in proceeds. The problem: these mega-listings could absorb all available public market capital, crowding out every other company waiting to go public. More than $4 trillion sits locked in unicorns. If even one of these giants stumbles post-listing, the chill could push the broader IPO window into 2027. Morgan Stanley separately warned that the market is “not prepared for the non-linear increase in LLM capabilities” expected in April-June, estimating $2.9 trillion in global data center construction through 2028. For builders: if you’re planning to raise or exit in 2026, the window may be defined entirely by whether SpaceX, OpenAI, and Anthropic go public — and whether they succeed.
Link: https://pitchbook.com/news/articles/the-mega-ipos-that-could-shut-out-the-rest-of-vc