Private Markets Update & Deal Flow

On May 20, SpaceX filed its S-1 with the SEC. The document runs 300 pages, but the sentence that matters is buried in the footnotes: Anthropic has signed to rent compute capacity across two SpaceX data centres for USD 1.25B per month, through May 2029. That is USD 15B a year. USD 45B over the life of the contract. For context, SpaceX's entire 2025 revenue was USD 18.7B. One customer. One signature. The AI division goes from the biggest liability in the prospectus to the headline number.

The matchmaking writes itself. SpaceX had quietly built Colossus 1 in Memphis — over 220,000 Nvidia GPUs, more than 300 megawatts of power — but Grok, the household-name AI product, never had the appetite to fill it. Across town, Anthropic had the opposite problem. Claude was scaling faster than AWS and Google could feed it. Excess supply, meet excess demand. The deal solves both companies' core constraints in a single signature, and it converts SpaceX's fixed-cost AI infrastructure from cash drain to cash machine at extraordinary margins.

The S-1 positions SpaceX as an integrated infrastructure company straddling three industries that rarely appear in one filing: space transportation, global broadband, and artificial intelligence. Starlink's numbers alone justify serious attention — roughly 9,600 satellites, over 10M subscribers across 164 countries, USD 11.4B in revenue last year. In February, SpaceX absorbed xAI, Musk's AI company, and the filing now describes AI as core strategy rather than a side bet. The prospectus even discloses that Anthropic has "expressed interest" in multiple gigawatts of orbital AI compute, with space-based data centres powered by solar energy planned from 2028. This USD 45B contract may be the first chapter of something considerably larger.

The financial picture still gives institutional investors pause. SpaceX lost USD 4.9B on USD 18.7B of revenue in 2025. Q1 2026 was sharper — a USD 4.3B loss on USD 4.7B of revenue. The Anthropic deal changes that calculus significantly. On governance, Musk retains approximately 85% of voting power through dual-class shares. His compensation package is even more extraordinary: a salary of USD 54,000 in 2025, supplemented by equity vesting only if SpaceX establishes a permanent human colony on Mars and hits valuations of USD 6.6T and USD 7.5T. Whether the market values SpaceX as an aerospace company, telecom, or AI platform — or all three — will define one of the most consequential listings in market history.

The signal: SpaceX is not simply an aerospace company seeking liquidity. It is the first platform to fuse orbital infrastructure, global broadband, and frontier AI compute under one cap table — with a controlling shareholder whose equity vests only if humanity becomes multiplanetary. The pre-IPO window at USD 1.35T closes June 11. Watch the multiple re-rate.

Google and Blackstone are launching a new AI cloud infrastructure company with an initial USD 5B equity cheque from Blackstone — the opening bid in what could grow to USD 25B of levered compute investment. The goal is straightforward: compete with CoreWeave using Google's own silicon. Until now, Nvidia has dominated the AI chip market almost without contest. Google wants to change that, and it is backing the ambition with one of the best-capitalised investors on earth. The new company plans to have 500 megawatts of AI computing capacity online by 2027 — enough to power a mid-size city — built around Google's Tensor Processing Units (TPUs). Google provides the chips, software, and cloud layer. Longtime Google executive Benjamin Treynor Sloss takes the CEO role. Blackstone takes majority ownership. Several data centres are already identified, some still under construction, and the first deployments are expected within 18 months. This is not Blackstone's first move in the AI infrastructure cycle — it already owns substantial data-centre assets and holds positions in Anthropic, OpenAI, and CoreWeave. The firm also launched a dedicated AI investment unit called BXN1 earlier this month to centralise that activity. What is new here is the directness of the challenge to Nvidia: a purpose-built company with Google hardware, Blackstone capital, and a mandate to provide the compute the AI industry needs at scale. The hyperscaler era is giving way to a new model. This week confirmed it.

The trend: Every major capital allocator is now racing to own a piece of AI infrastructure before the market sets a price. Blackstone is doing it with real estate and Google chips. SpaceX is doing it with rockets and GPU farms. The common thread: whoever controls the compute layer captures the margin. The window to enter at private-market prices is closing fast.

The war in Ukraine taught European militaries one uncomfortable lesson: if you do not own your satellite intelligence, someone else controls your battlefield. Germany has decided not to repeat that mistake. Berlin plans to invest EUR 35B in military space technology by 2030, and the first major test of that commitment is Spock 2 — a programme to build an AI-powered constellation of surveillance and target-identification satellites for the Bundeswehr.

AI drone maker Helsing has teamed up with satellite manufacturer OHB to chase the contract. The pitch is credibility on both sides: Helsing brings the AI targeting and mission intelligence software; OHB brings four decades of European satellite engineering. Co-CEO Gundbert Scherf argues Europe can "leapfrog" the US by building a defence space architecture from scratch rather than retrofitting Cold War infrastructure. Rheinmetall, Iceye, and Airbus are expected to bid against them. A contract decision is not expected before 2027.

The Spock 2 bid is one thread in a broader European defence-space tapestry. Helsing and OHB are simultaneously embedded in an alliance with Norwegian firm Kongsberg, rocket start-up Isar Aerospace, and sensor specialist Hensoldt. OHB is also developing a European alternative to Starlink for the German military. The direction of travel is unmistakable: Europe is building a parallel infrastructure stack for war-fighting, and it is spending at a pace that is just beginning to show up in contract awards. The companies positioned at the intersection of AI and orbital intelligence are the category-defining bets of this European defence cycle.

Worth watching: European defence-space is no longer a budget line — it is a strategic imperative backed by sovereign capital at scale. The procurement cycle is just beginning, the valuations of the emerging champions are still private-market prices, and the window to invest alongside the winners is open now.

Anthropic has built a model — Claude Mythos Preview — that is specifically designed to find holes in software. It has found rather a lot of them. According to the company, Mythos has already identified thousands of serious vulnerabilities across major operating systems and web browsers. Vulnerabilities that, if exploited before being patched, could have material consequences for economies, financial systems, and national security. The White House told Anthropic not to release it broadly. The company agreed, restricting access to around 40 organisations — Amazon, Microsoft, JPMorgan, and a small number of others cleared to use Mythos to harden their own defences. Everyone else waits. Bank of England Governor Andrew Bailey has reportedly asked Anthropic to brief the Financial Stability Board directly. His concern is not theoretical: if a model can find thousands of critical vulnerabilities in weeks, so can the people who would use them maliciously — and the gap between discovery and patch deployment in global financial infrastructure is measured in months, not days. The IMF has separately warned that AI-powered cyber attacks could become a "macro-financial shock" — particularly for emerging economies with weaker digital defences. Authorities in the UK have already urged financial firms to accelerate software updates. The broader problem is structural: the geopolitical tensions between major powers make meaningful international coordination on AI security close to impossible. The model exists. The capability is real. The question is simply who uses it first — and whether the defenders can keep up.

The investment angle: AI-augmented cybersecurity is transitioning from a nice-to-have to a mandatory spend category across global financial institutions. Companies that can provide defensive AI tooling at enterprise scale are entering a multi-year growth cycle driven by regulatory compulsion, not just demand. This is a theme we are tracking actively in our deal pipeline.