The Investor

The Gating Event for AI Exits

SpaceX has filed its confidential IPO prospectus, internally dubbed 'Project Apex,' targeting a mid-June listing at $75 billion. The public S-1 is expected in May 2026, followed by investor site visits to SpaceX's Starship facility in Texas and data centers in Tennessee, then a formal roadshow in early June. This is the largest tech IPO since Arm's 2023 offering — and its success or failure determines whether Anthropic and OpenAI accelerate their IPO timelines into H2 2026 or stay frozen.

In an unprecedented move, SpaceX may refuse to name a lead left bank, instead listing underwriters alphabetically. Goldman Sachs's Kim Posnett and Morgan Stanley's Michael Grimes are both embedded at SpaceX's Hawthorne headquarters, fighting for pole position. The real prize isn't SpaceX's underwriting fees — it's control over the Anthropic and OpenAI IPO mandates that follow.

The Cursor Deal: $60B Reprices AI Developer Tools

Simultaneously, SpaceX announced an option to acquire Cursor for $60 billion with a $10 billion breakup fee — the most extraordinary M&A structure in AI to date. The option format suggests SpaceX wants post-IPO stock as acquisition currency while locking out competing bidders. For Cursor's investors (a16z, Thrive, Coatue, Accel), this is heads-I-win-tails-I-still-win: either the acquisition closes at $60B, or they pocket $10B in breakup fees.

The signal for your portfolio: AI developer tools have hit a valuation ceiling where only strategic acquirers with trillion-dollar ambitions can underwrite the price. Venture-backed IPOs at $60B for a coding tool strain credulity; a SpaceX acquisition funded by post-IPO equity makes the math work differently.

This creates an immediate repricing question for every AI developer tools company in your deal flow. But distinguish between strategic premium (SpaceX-specific) and market clearing price. The option structure — not the headline number — tells you this valuation requires a specific buyer, not a liquid public market.

The Banking War Signals What Comes Next

Whichever bank demonstrates it can execute a $75B listing in a fragile market earns the right to lead Anthropic and OpenAI. SpaceX's refusal to name a lead left is deliberate — keeping both banks hungry through the entire AI IPO cycle.

The Risk: Debt Disclosure

SpaceX's 'huge debt load' is the underreported risk. When the public S-1 drops in May, debt-to-equity ratios could compress the $75B target. If you hold SpaceX secondary positions, stress-test against a $50-60B outcome. The valuation trajectory — $400B→$800B (Dec 2025)→$1.25T (Feb 2026 post-xAI merger) — also carries governance risk around Musk's $6.6 trillion compensation target.

The Copilot Capitulation

GitHub VP of Product Joe Binder admitted that agentic coding workflows consume 'far more resources than the original plan structure was built to support' — and Microsoft's response was a full retreat. Individual signups are paused. Anthropic's Claude Opus 4.5 and 4.6 are removed from the platform. Opus 4.7 is gated behind Pro+ ($39/user/month). Session caps with weekly token ceilings are being introduced. The explicit admission: current pricing is unsustainable.

If GitHub — with Microsoft's Azure compute at cost and 100M+ developers — can't make AI coding tools profitable, the entire category of AI coding startups at Series A/B valuations of 60-100x ARR is mispriced. Uber's CTO separately demonstrated how Claude Code can 'blow up AI budgets.'

The Industry-Wide Repricing

This isn't isolated. Every major AI coding provider moved to usage-based or tiered pricing within the same month:

Simultaneously, Cursor is raising $2B+ at a $50B valuation while only 'slightly gross-margin positive' on a projected $6B ARR trajectory. That's a16z and Thrive writing enormous checks on a margin expansion thesis that hasn't been proven.

The Outcome-Based Pricing Fork

While token-based models break, a parallel development is reshaping the entire enterprise AI pricing stack. Adobe launched CX Enterprise with outcome-based pricing — customers pay only when AI agents deliver completed business outcomes, not per token consumed. Adobe President Anil Chakravarthy: 'Tokens don't equate to value.'

This is a leveraged bet on declining inference costs. If compute costs fall 50% over 18 months (consistent with current trajectories), Adobe's margins expand without repricing. Sierra (Brett Taylor's AI startup) is already pricing per completed task. Salesforce coined the 'Agentic Work Unit' in February 2026, building measurement infrastructure for outcomes.

The pricing model a company chooses is now a leading indicator of management conviction in their AI's actual efficacy. Outcome-based = 'our AI works well enough to bet margins on it.' Token-based = 'we're not confident enough to guarantee results.'

The Competitive Window

Copilot's freeze is a rare forced-churn event in a market with 50M+ developers. The one production-verified data point: Cloudflare achieved 93% AI coding tool adoption driving a 55% increase in merge requests (5,600→8,700/week). That's the most concrete enterprise productivity proof point published to date. Companies that convert displaced Copilot users in the next 60 days capture a once-in-a-cycle distribution windfall.

Not a Venture Investment — A Cloud Procurement Deal Structured as Equity

Amazon committed up to $25 billion in new investment into Anthropic ($5B upfront, $20B milestone-based), bringing total commitments to $33 billion. In return, Anthropic pledged to spend more than $100 billion on AWS over the next decade and run models on Amazon's Trainium custom chips. The Claude Platform is being embedded directly into AWS Bedrock. Amazon is buying future revenue at roughly 33 cents on the dollar.

Combined with Amazon's separate $50B commitment to OpenAI, Amazon has deployed $83B across both leading AI labs — hedging the model race while monetizing the infrastructure layer on both sides. This is not conviction about which lab wins. It's a cloud monetization masterclass.

The Flywheel That Forecloses Competition

The critical nuance: Anthropic is simultaneously buying Google's custom AI chips while contractually committed to Amazon's cloud. This dual-sourcing strategy gives Anthropic leverage against both hyperscalers — and signals that Nvidia's grip on frontier AI training is loosening at the very top of the market.

The Competitive Intelligence Gift: Google's Internal Panic

Multiple sources confirm Sergey Brin has personally assembled a DeepMind 'strike team' led by Sebastian Borgeaud (former head of pretraining), reporting directly to CTO Koray Kavukcuoglu and Brin himself. The catalyst: DeepMind's own researchers rate Claude's coding ability above Gemini's. Brin issued an internal memo demanding every Gemini engineer use internal agent tools, with usage tracked on a company leaderboard called 'Jetski.'

When your internal team concedes the competitor is better, and your co-founder creates an emergency team led by your former head of pretraining — that's not a product iteration, that's a wartime footing.

What This Means for Every Other AI Company

For competing AI labs: you either lock into a hyperscaler or you get outspent. For AI startups: if you're building on AWS, you're now competing for resources with Anthropic. If you're on Azure, you're competing with OpenAI. The multi-cloud AI orchestration market just went from nice-to-have to existential.

For Anthropic secondary positions: at $380B on $33B in Amazon commitments against $100B+ in AWS spend, the valuation is anchored to cloud economics. But Anthropic is also declining an $800B+ round — classic pre-IPO behavior. The company is building the revenue diversification story (Claude Design, enterprise productivity, not just API access) that public investors reward.

Meanwhile, Bezos's Project Prometheus — a 5-month-old AI physics simulation startup — is raising $10B at $38B valuation with JPMorgan and BlackRock participating. The market is bifurcating between 'celebrity AI' at $38B with zero revenue and 'normal venture' at reasonable multiples. The risk profiles are radically different.