Finance

When Anthropic settled the Bartz case for $1.5 billion — about $3,000 a book — the press called AI's copyright question resolved. A Pulitzer winner and five other authors read the same number, opted out, and are now arguing it was 50 times too low, across six labs at once. The real exposure isn't the six-author suit, which is pocket change. It's that the settlement re-priced the liability upward and left a tail that, unlike a class deal, can't be capped — right as these companies walk into IPOs that assume it's over.

Global startup funding hit a record $510 billion in the first half of 2026, and the celebration is missing the only detail that matters: OpenAI and Anthropic alone absorbed $217 billion of it, and every gain everyone is toasting — Menlo's $14 billion, Amazon's record quarter — is paper that has never turned into cash. The venture asset class has quietly become a leveraged bet on two IPOs, one of which its own CEO says he needs to survive and the other of which keeps getting pushed to 2027.

In one week, Microsoft, Amazon, Anthropic, and OpenAI all concluded the same thing: you can't sell frontier AI without shipping humans to install it. That's not a growth story — it's a margin confession, and they know it, which is why the humans are being tucked into joint ventures and subsidiaries where they won't touch the software-margin math being sold into trillion-dollar IPOs. The number that kills the whole 'Palantir playbook' framing: Palantir keeps services to under a fifth of revenue on purpose.

In a single week before its S-1, Anthropic launched a drug-discovery platform and bought a biotech for ~$400M, put Claude on Azure for the first time, signed California's entire government at 50% off, cut Sonnet 5 to a third the cost, shipped an AI teammate into Slack, and opened a Seoul office. One at a time it's press releases. Together it's a coordinated land-grab to lock in distribution on every front and defend a $965B valuation against defection and federal gating. The buried risk: cutting prices and buying channels during a token-rationing pullback, right as the S-1 lands, could spend the valuation it's meant to defend.

Visa wired its full network into ChatGPT and Mastercard built tokens that let an AI agent hold and spend a credential — the AI you talk to is becoming the AI that buys. The part shipped without a safeguard: if your agent overspends its scope, the Electronic Fund Transfer Act's 'unauthorized transfer' shield may not apply, and the consumer, not the bank, may eat it. No federal regulator has ruled, so the card networks are writing the consumer-protection rules themselves. Meanwhile the front door is being rebuilt to transact (Google's search box became an agent, -58% click-through, publisher antitrust suits) and the default-agent antitrust fight is next.

Tech cut ~153,000 jobs in 2026 while the four hyperscalers committed ~$725B to AI infrastructure, and the industry stapled the two into two narratives: the layoffs fund the AI splurge, and AI is doing the firing. Both collapse on contact with the numbers. Meta saved ~$2.4B firing 8,000 people while its free cash flow falls ~$35B under capex — the savings are a rounding error. And the biggest cut (Oracle's 30K) was legacy database staff, not AI-displaced knowledge workers, with ~275K AI roles sitting open. It's a skill-mix reshuffle dressed as an AI story.

Two of the biggest distributors of consumer AI just made opposite bets in the same season. OpenAI made its best model the free default for 500M; Meta capped free Meta AI and started charging — $7.99 and $19.99 Meta One tiers — for the chatbot behind 3.2B daily users. The consumer-AI business model is forking into ad-subsidized-free vs paywall, and Meta is the one player structurally conflicted: it's an ad company, so steering users into paid AI risks cannibalizing the engagement that funds its ads. And it's only a 3-market test — which tells you Meta isn't sure either.

A profitable 25-person agent startup whose token bill had outgrown salaries moved 100% of its model traffic off Claude to US-hosted DeepSeek, cut inference cost ~90%, and says the agents got better. It's not an outlier: Uber burned its whole 2026 AI budget in four months then capped engineers at $1,500/mo, and both labs rushed out spend-limit controls. The token economy just flipped from tokenmaxxing to rationing — and the contradiction Wall Street isn't pricing is record ARR while flagship customers defect. That's a net-revenue-retention problem wearing a growth story's clothes.

The cleanest way to read the AI boom is the AMD–OpenAI deal, and the cleanest way to read that deal is to notice what it doesn't contain: a purchase order. AMD gave OpenAI a warrant for up to 160 million shares at a penny — the '6GW' isn't a commitment, it's the threshold that vests the stock. Stack ten deals like it and the same three tricks repeat: vendors pay customers in their own equity, funders book profits on the rivals they fund, and labs rent compute from competitors who can pull the plug in 90 days. It isn't ten clever deals. It's one interlinked machine — and the question nobody asks is who the single point of failure is.

OpenAI confidentially filed its S-1 on June 8, 2026, targeting a Q4 2026 listing at $1 trillion or more. By June 25, two Tier-1 outlets — the Wall Street Journal and the New York Times, from separate source chains — reported that CFO Sarah Friar has privately suggested waiting until 2027, and that Sam Altman considers any valuation below $1T a 'non-starter.' Coverage frames this as a negotiation between the CFO and CEO over timing. It is not a negotiation. In Q1 2026, OpenAI posted a -122% non-GAAP operating margin — losing $1.22 for every dollar it earned — with $14 billion in projected full-year losses and $600 billion in locked infrastructure commitments through 2030. The $1T valuation requires public-market investors to bet on a profitability inflection that OpenAI's own internal models don't show until end of 2027. Meanwhile, Anthropic — $47 billion ARR, S-1 filed one week before OpenAI's, October 2026 target — arrives first at the 'pure AI lab IPO' narrative and absorbs the institutional capital before OpenAI comes to market. The date debate is real. The question it is hiding is whether OpenAI can survive public-market scrutiny long enough for its profitability inflection to materialize.

OpenAI's $122 billion Series G was the largest private funding round in history. The headline number — Amazon's $50 billion lead investment — has $35 billion contingent on an OpenAI IPO or a redacted 'AGI milestone' by December 31, 2028. If neither triggers, Amazon walks. Meanwhile, Amazon confirmed on record that it maintains no firewalls between that position and its simultaneously growing investment in Anthropic. In Q1 2026, more than half of Amazon's pre-tax income came from Anthropic stake appreciation. The company whose cloud pricing decisions determine both labs' operating costs is the largest investor in both.

On June 10, Visa announced that AI agents can now initiate purchases at any of its 175 million merchant locations — and Mastercard launched a competing agent payment system the same day with a fundamentally different architecture. The race to own AI commerce infrastructure is already underway. The legal question neither company has resolved: when an agent buys something the consumer didn't intend, does Visa's zero-liability guarantee apply? The Electronic Funds Transfer Act's 'unauthorized transfer' definition, written in 1978, has not been adjudicated for the scenario where a consumer authorized the agent but not the specific purchase.

Anthropic's $65B raise at $965B valuation got every headline right and missed the only number that matters. The $47B ARR is reported gross — Anthropic counts end-customer spend through AWS and Google as its own top-line. No outlet has published an estimate of net revenue. That number appears in the S-1. When it does, it will reset the valuation conversation. The press covered the milestone. Nobody covered the accounting.

OpenAI filed a confidential S-1 on June 8, targeting a September listing at $1T+. The filing forces public disclosure of numbers OpenAI has never had to share: gross margins by segment, compute costs, customer concentration, Sam Altman's equity, and revenue from the Microsoft relationship after the April renegotiation. At $25B ARR with $14B in projected 2026 losses and a $1T valuation target, investors are being asked to buy at 40x revenue for a company burning $1.22 per dollar earned. Seven weeks later, Anthropic files for October. The same banks sell both.

OpenAI's chief revenue officer publicly accused Anthropic of overstating its revenue by $8 billion. That dispute is unresolved. Anthropic then filed for an IPO, which means the S-1 will be the first document legally required to show its work. The number that defines this listing isn't the valuation — it's the accounting methodology.

Alphabet's stock fell $250 billion in a single session after losing the co-architect of Gemini and the AlphaFold Nobel laureate in 48 hours — but the market wasn't pricing two departures, it was pricing a question: can a company fund $180 billion in AI capex on $20.5 billion in annual free cash flow, without the people who built the thesis? The answer might be yes. The confidence interval just got a lot wider. And the culture that drove Noam Shazeer out twice is the part that doesn't fix itself.