Ledgers, Lawsuits, & Liftoff

SpaceX's first public quarter, Kalshi's preemption shield cracks, and the AI buildout gets financed on credit.

πŸ‘‚πŸ½ You can now listen to The Markup

For those of you who don’t know how to read, we added the ability to listen to our newsletter. Just click the β€œListen online” link above.

Give it a try and let us know if you like it. Next week, you may even hear a familiar voice.

SpaceX dropped its first print as a public company and the AI unit did the damage. Anthropic is financing compute on private credit, $10B at a time. Kalshi lost twice in court in a single day. OpenAI crossed a billion users, cut prices 80%, and made frontier chat free. xAI got squeezed on both its product and its power. Apptronik became Google's reference humanoid.

A lot to unpack this week.

Let's get into it.

πŸ“ˆ Portfolio Pulse

πŸš€ SpaceX: First Print as a Public Company

SpaceX posted its first quarter since June's IPO: $7.8B in revenue, up 92%, beating estimates by nearly $1B. The catch was the AI unit, which lost $1.3B on $15.8B of capex, and shares fell about 7% after hours. Days later, Redwire booked the first commercial mission on the new Starfall reentry vehicle, up to 32 pharmaceutical-crystallization labs targeted for 2028. The near-term overhang is this week's first insider lockup expiry, freeing roughly $116B in pre-IPO shares into a short base already at 34% of float. The first public quarter gets repriced from both sides.

🧠 Anthropic: Compute on Credit

Anthropic signed a $10B, six-year compute deal with Nvidia-backed startup Volta, tapping a 133MW Norway data center running on Vera Rubin chips, while Blackstone shopped a second mega debt package to finance its Google-chip use. The same week, Anthropic sat in the White House meeting shaping the first US voluntary frontier-model testing framework. It is widening both its compute suppliers and its financing sources, and private credit is now the funding rail for the buildout. A safety-first lab writing the rules also sets the compliance baseline every rival will operate against. The capital stack now reaches well beyond equity.

πŸ“Š Kalshi: The Preemption Shield Cracks

Two courtroom losses in one day. A New York judge denied the CFTC's emergency TRO, letting the state's $36B suit proceed, and a Utah judge ruled the Commodity Exchange Act does not preempt state gambling law. Kalshi will appeal to the Tenth Circuit. The federal-preemption defense underpinning the entire event-contracts model is not holding in court, and every state win widens the map of jurisdictions that can move to block it. All of this lands as Kalshi reportedly raises near a $40B valuation, one week after taking $12.4B in World Cup volume. The rounds are getting bigger. So are the lawsuits.

πŸ€– OpenAI: A Billion Users, Eighty Percent Off

OpenAI crossed one billion active users and two million businesses, disclosed alongside an 80% price cut on GPT-5.6 Luna and 20% lower serving costs. Then it made frontier chat free and unlimited, defaulting Free users to Luna and moving Plus and Pro to an upgraded Sol. Behind it, SoftBank arranged a $10B margin loan against its OpenAI shares to fund the final tranche of its $30B commitment, lifting its stake toward $64B. Cutting price into a billion-user base while raising inference burn resets the floor rivals must clear. Make of that what you will.

⚑ xAI: Regulation Meets the Power Build

A Minnesota court let the state's AI nudification ban take effect over xAI's objection, and Mississippi regulators ordered the removal of all 69 temporary gas turbines at the Southaven data center, to be replaced by a permanent 1.2 GW plant. Regulation is now a direct input to both xAI's product surface and its power strategy, raising the cost and the timeline of the compute build-out. Colossus does not run without electrons, and the electrons just got more expensive.

🦿 Apptronik: Google's Reference Humanoid

Google DeepMind launched Gemini Robotics 2 and demonstrated it on Apptronik's Apollo 2 humanoid, which ran autonomous full-body tasks reasoning in real time. Being the reference hardware for a frontier physical-AI model deepens Apptronik's tie to Google DeepMind in the humanoid race. The model is the brain. Apptronik is building the body.

πŸ‘€ Watch List

πŸ“ˆ Polymarket: Chasing Kalshi's Valuation

Polymarket is in early talks to raise around $1B at a valuation above $20B, up from the $15B round it closed in April, with annualized revenue said to be well above $1B. The same week, it became the official prediction-market partner of the New York Yankees, putting its brand on stadium signage during televised games. A roughly 33% step-up in four months signals sustained institutional appetite and an escalating capital race with Kalshi. The category is repricing in real time.

βš›οΈ Valar Atomics: Sequoia Bets on Fission

Valar raised a $1B Series B led by Sequoia at a roughly $6B valuation, triple its April mark, plus a $200M credit facility, to scale from prototype SMRs to reactor fleets for AI data centers. A 3x step-up and $1.2B in fresh capital with a marquee lead vaults a founder-led fission startup into the front rank of AI-power SMR players. The power bottleneck is the whole game.

πŸ›‘οΈ Hadrian: A New Benchmark for the Factory Trade

Hadrian raised a $1.37B Series D at a roughly $7.87B valuation to scale automated defense, aerospace, and industrial factories, co-led by WCM, Washington Harbour, Valor, 137 Ventures, and Baillie Gifford, with JPMorganChase anchoring. A roughly 5x step-up to nearly $8B resets the comps for automated defense-industrial-base manufacturing and shows how hard growth capital is chasing the sector. The reindustrialization trade has a new benchmark.

πŸ’¬ Prompt

What can I help you with?

Write the headline about my industry that will make me feel stupid for not seeing it coming in 2030. Then work backwards and tell me what's already visible today that points straight to it.

πŸ“š Content

πŸ“ˆ The Compute Price Paradox

Dwarkesh makes the counterintuitive case that smarter models could 10x the price of compute: as models get sharper they burn far more tokens per task, so demand outruns the buildout and inference prices climb instead of falling. He points at Anthropic going from $9B to a projected $100B-plus in a year while capacity only triples. The whole Lambda, Groq, and Crusoe thesis lives or dies on which way this breaks. Watch the video.

πŸŽ™οΈ Gavin Baker Went Looking for a Bear Case

On a fresh Invest Like the Best, Atreides' Gavin Baker says he spent a week hunting for a single negative quantitative metric on AI and came up empty. His read: compute repricing as old contracts roll off backstops the buildout with cash flow rather than credit, and the data-center demand-collapse story is a myth. A clean counterweight to the prices-go-up-10x camp if you want both sides in one sitting. Listen here.

🎬 Karpathy Hands Opus 5 Ten Dollars and Middle-earth

Andrej Karpathy gave Anthropic's Opus 5 the opening paragraph of The Lord of the Rings, a roughly $10 token budget, and asked for a three.js render. The model ran for two hours, wrote 5,500 lines of code that procedurally build the scene, and took its own screenshots along the way to check its work. Janky but mindboggling, and a far better vibe test for frontier models than the old pelican-on-a-bicycle. Watch the clip.

🀝 Get in the Room

Private investing for (almost) everyone*

We are a curated network of exceptional angels, VCs, family offices, founders, and industry experts investing in pre-IPO companies. We back sharp operators and high-leverage deals. We move fast. We bring value. We don't miss around the edges. Deals don't wait. Neither do we.

*Vibes and values must be aligned. Must be an "accredited investor."

Stay sharp,

Β© 2026 LFG Ventures

All rights reserved.

The content presented in The Markup is for informational purposes only and does not constitute an offer or solicitation by LFG Ventures to buy or sell any securities. Nothing in this publication should be interpreted as investment advice, a recommendation, or an offer to provide brokerage services. Any offers or solicitations will be made only through official offering documents and subject to their specific terms, risks, and conditions.