Hey Upper East Siders (and Sand Hill Road dwellers)...

Gossip Girl here, your one and only source into the scandalous lives of Sand Hill Road's elite. And darlings, today's tea is scalding. We're talking Pentagon standoffs, half-billion dollar chip bets, and a certain safety-first AI darling having a full identity crisis. Grab your oat milk lattes and settle in.

💣 The Main Event: Anthropic's Very Bad, No Good Week

Spotted: Anthropic having an absolute meltdown of a week — and not the graceful kind.

First, CEO Dario Amodei got summoned to the Pentagon on Monday where Defense Secretary Pete Hegseth essentially told him: open up Claude for full military use by Friday at 5pm, or face the consequences. We're talking canceled contracts, supply chain blacklisting, and potentially invoking the Defense Production Act. Yes, the one they use for wartime. Over an AI chatbot. (Axios)

Sources say the meeting was "cordial" — which in DC speak means someone definitely raised their voice. Amodei held his red lines: no mass surveillance of US citizens, no fully autonomous weapons without human oversight. The Pentagon's response? They started calling up Lockheed Martin and other defense primes to assess their "reliance" on Claude. You didn't hear it from me, but that's the first step toward a formal blacklisting. (Gizmodo)

Then — in what the company swears is totally unrelated timing — Anthropic dropped its signature safety pledge. You know, the one that said they'd "pause scaling" if safety measures couldn't keep up? Gone. Chief Science Officer Jared Kaplan told Time the policy "was not in keeping with the current state of the AI race." Translation: we can't afford principles when OpenAI is breathing down our neck. (Business Insider)

And because this week apparently wasn't spicy enough, a hacker used Claude to breach multiple Mexican government agencies, stealing sensitive tax and voter data. Gambit Security uncovered the breach. The irony of the "safety-first" AI company's model being jailbroken while the Pentagon demands less safety? Chef's kiss. (Bloomberg)

Congress is now being urged to investigate. Bipartisan letters are flying. The LA Times is calling it a "battle of the AI brands." This is giving season finale energy, and it's only Wednesday.

💰 Spotted: The Freshest Deals on the Block

MatX — $500M Series B Word on the street: this Nvidia challenger just raised a cool half-billion led by Jane Street and Situational Awareness (that's Leopold Aschenbrenner's fund, for those keeping score). The chip startup, founded by two former Google TPU engineers, claims their processors will be 10x better at training LLMs than Nvidia's GPUs. Stripe co-founders Patrick and John Collison and Spark Capital also joined. Bold claim + bold backers = we're watching. (TechCrunch)

Gambit Security — $61M Seed + Series A The Israeli cyber resiliency platform that just exposed the Claude-Mexico hack raised a combined $61M from Spark Capital, Kleiner Perkins, and Cyberstarts. Talk about perfect timing for a press cycle. (Axios)

Comp — $17.25M Series A São Paulo-based AI HR platform led by Khosla Ventures, with Keith Rabois taking a board seat. Co-founders Christophe Gerlach and Pedro Bobrow serve 100+ companies including Nubank and QuintoAndar. An AI-native HR platform that combines autonomous agents with embedded HR professionals? The future of "you're hired" just got automated. (TechCrunch)

Anthropic acquires Vercept — $50M raised In acquisition news, Anthropic snapped up computer-use AI startup Vercept. CEO Kiana Ehsani announced the deal on LinkedIn. Fun twist: co-founder Matt Deitke had already been poached by Meta's Superintelligence Lab last year with a jaw-dropping $250M salary. One co-founder goes to Meta, the other sells to Anthropic. That's what we call hedging your bets, darlings. (TechCrunch)

RLWRLD — $26M Seed 2 Physical AI startup building robot foundation models for actual factories and warehouses (not pristine lab conditions). CEO Junghee Ryu is training models with real-world industrial data. Because apparently robots need to learn that factory floors are messy. Relatable. (The Robot Report)

Cicada — $13.5M Series A NYC-based electronic trading platform led by Citi, with Brazil's B3 stock exchange venture arm joining. Elly also raised $8M from Sorenson Capital for its AI-native hiring platform. And London's Callosum pulled in $10.25M from Plural to challenge entrenched AI compute models with multi-chip infrastructure. The long tail of deals is thriving.

Ascend Capital Partners closed its second healthcare-focused PE fund at $791M. And Seraphim Space raised a $100M fund backed by the British Business Bank for seed and Series A space tech investments, bringing their total AUM to $550M. The final frontier gets more VC love.

🗣️ What the VCs Are Saying

Troy Kirwin (Investor @ a16z, Building @speedrun) dropped a viral thread predicting "In 2026, Venture Capital will eat Private Equity." His thesis: as AI eats labor costs in traditionally PE-owned service businesses, the VC and PE worlds are colliding through AI rollups, PE portfolio companies becoming startup channel partners, and AI-native platforms acquiring legacy businesses. 1.3K likes and counting. The Patagonia vests and the PE suits are merging, and honestly? We saw this coming. (Post)

Katherine Boyle (GP, a16z) got quoted making waves about Elon Musk's real legacy: "I think Elon's most important contribution to this country is training two generations of engineers to work with their hands again." The post riffing on her take went viral — arguing that SpaceX and Tesla didn't just build products, they rebuilt America's industrial knowledge base after decades of decay. Say what you will about the man, but Boyle isn't wrong about the talent pipeline. (Post)

Peter (1kx, @pet3rpan_) had a sharp take on the commodification of capital: "The required cost of capital to launch a software company will trend towards zero." He argues early-stage VC must now compete on distribution, signal, governance, and differentiated services — not just checks. 202 likes, 15K views, and a lot of GPs nervously adjusting their pitch decks. (Post)

🔥 The Bigger Picture

Nvidia printed $68.1B in Q4 revenue — up 73% YoY — and previewed its next-gen Vera Rubin AI system shipping in H2 2026. Jensen Huang also said the OpenAI investment deal is "close." Meanwhile, Meta inked a $100B+ deal with AMD for AI chips, proving Jensen doesn't get to be the only GPU supplier at the cool kids' table. (WSJ)

Cato Networks crossed $350M ARR, with CEO Shlomo Kramer saying AI is driving adoption. And the software sector got a reprieve when Anthropic's enterprise agent partnerships with Slack, Intuit, DocuSign, and others quelled "Softwaremageddon" fears. Wedbush says the disruption risk is "overblown." For now. (CNBC)

Oh, and Trump told Big Tech to build their own power plants for AI data centers. Amazon, Google, Meta, Microsoft, Oracle, xAI, and OpenAI will reportedly sign a pledge on March 4th. Because apparently the AI boom needs its own energy policy now.

🫖 The Quick Sips

• Foodforecast raised €8M Series A — AI-driven forecasting for ultra-fresh food. Because even sandwiches deserve optimization.
• Stanhope AI raised $8M for adaptive AI in robotics and defence. London's defense tech scene isn't sleeping.
• BeyondMath pulled in a $10M seed extension from Cambridge Innovation Capital for foundational AI models.
• Healthtech VC hit $15.3B in 2025 (up 26% YoY) per PitchBook, with AI reshaping every clinical and administrative workflow.
• TechCrunch Founder Summit 2026 is June 9 in Boston. 1,000+ founders and investors. Mark your calendars.

Until next time, remember: in the world of venture capital, there are no secrets — only stories that haven't been leaked yet.

You know you love me,

XOXO — Gossip Girl 💋