Three deeptech companies closed nine-figure-plus rounds in the same few weeks this August, and the striking thing isn’t the size of any single check. It’s who’s writing them and how the money is structured. A UK sovereign fund and Arm backed a semiconductor startup’s chip architecture bet. A revolving credit facility, not just equity, is funding a missile-manufacturing campus. Google and Toyota’s venture arm are betting directly on satellite production capacity, not just the data those satellites collect. None of this looks like the venture capital playbook from ten years ago. It looks a lot more like project finance.
A chip startup’s $3.3 billion bet on a different kind of memory
OLIX just raised a $312 million Series B at a $3.3 billion valuation, more than three times its roughly $1 billion mark from a $220 million Series A only six months earlier, in what’s being described as Europe’s largest-ever semiconductor round. The round was led by Fundomo, with participation from Arm, Hudson River Trading, Netflix co-founder Reed Hastings, and, notably, the UK government’s own Sovereign AI venture fund, according to Data Center Dynamics’ coverage of the raise . OLIX’s first chip, the DX-1, pairs on-chip memory built from SRAM (static random-access memory, faster than the alternative but harder to pack in large volumes) with a photonic interconnect, instead of the HBM (high-bandwidth memory) paired with copper wiring that most AI accelerators use, and the company is targeting more than 10,000 tokens per second per user in decode inference (the speed at which a large language model generates output text for a single user) on models with more than 100 billion parameters. Customer silicon is due in the second half of 2027.
That throughput number is worth sitting with for a second, because every public account of it traces back to OLIX’s own “Compute Manifesto.” No independent technical benchmark has surfaced yet to confirm it. That doesn’t mean the number is wrong. It means a UK sovereign fund just underwrote a specific, unverified architecture bet years ahead of the product shipping, the kind of claim that’s checkable in theory but hasn’t actually been checked by anyone outside the company yet , which is a meaningfully different kind of bet than a growth-stage investor writing into a company with revenue already flowing.
A missile maker’s revolving credit facility
Castelion closed a $1 billion Series C at a $13 billion valuation, co-led by JPMorganChase’s Strategic Investment Group, Andreessen Horowitz, and Carlyle-managed funds, with T. Rowe Price joining as a new investor, according to TechCrunch’s report on the round . But only $800 million of that billion is equity. The remaining $250 million arrived as a revolving credit facility. The money is going toward scaling production of Blackbeard, Castelion’s low-cost, mass-producible hypersonic strike missile, reportedly priced around $384,000 per unit, funding a new manufacturing campus in Sandoval County, New Mexico, and building on more than $500 million in US military contracts the company says it has landed over the past eighteen months.
A quarter of a mega-round showing up as debt against contracted government revenue, instead of pure equity, is the clearest signal in this entire brief that late-stage defense-tech investors are starting to underwrite production lines the way infrastructure lenders underwrite a toll road or a power plant: against a revenue stream that already exists, not against a story about where the company might be in five years. If you’re chasing a capital-intensive, government-contract-heavy business, that’s the lesson to take from Castelion’s cap table. Backlog and unit economics, documented well enough to satisfy a bank’s underwriting standards rather than just a VC’s pattern-matching, are what get a company access to debt alongside equity.
Big Tech buys into the factory, not just the data
Muon Space raised a $250 million Series C at a $1.5 billion valuation, pushing its total equity funding past $386 million, led by Eclipse with new investors Google, Salesforce Ventures, Wellington Management, and Toyota’s Woven Capital, according to SpaceNews’ report on the funding . The round funds a new 130,000-square-foot manufacturing facility in San Jose designed to produce up to 500 satellites a year by 2027, a tenfold jump from Muon’s current capacity, as it scales its “Mission Foundry” model of building custom satellite constellations for commercial, government, and sovereign customers.
What stands out here is that Google is investing directly in the factory that builds the satellites, not just a company that sells the data those satellites collect once they’re in orbit. That’s a step further up the supply chain than most hyperscaler space bets have gone so far, and it points toward Big Tech wanting a stake in orbital manufacturing capacity itself as satellite data and orbital compute increasingly feed their own AI businesses. For space-industry founders raising from strategics who are new to the sector, that shift also raises the diligence bar. The question used to be roughly “can you get to orbit.” Increasingly, it’s “what’s your actual production throughput, and can you defend those unit economics under real scrutiny.”
Three different rounds, one underlying shift
Put OLIX, Castelion, and Muon Space side by side and a pattern emerges that has nothing to do with sector. A sovereign wealth fund backing unverified chip architecture. A bank-grade credit facility layered onto a missile-manufacturing round. A search-engine company and an automaker’s venture arm underwriting satellite factory capacity. In each case, the capital isn’t just chasing a promising technology roadmap the way early-stage venture traditionally does. It’s underwriting a specific, physical production capability, tied to a real facility, a real contract base, or a real customer relationship, the way infrastructure and project finance has always worked: lend or invest against an asset and a revenue stream you can actually inspect, not against a narrative about where the market is heading.
For founders building anything capital-intensive, the practical takeaway is to start thinking about your own fundraise in those terms well before you’re big enough for a $250 million round. What’s the physical asset, contract backlog, or production capacity you can point an infrastructure-minded investor toward, rather than only a market-sizing slide? OLIX still has to answer that question with real, independent benchmarks before its H2 2027 shipping date arrives. Castelion and Muon Space already have facilities and contracts standing behind their numbers. That’s the difference between capital betting on a story and capital betting on a plant, the same distinction that’s separating this month’s most fundable robotics rounds from its most speculative ones , and it’s worth knowing which one you’re actually asking for.
What to watch next
Three concrete dates and claims are worth tracking as this plays out. Whether OLIX’s customer silicon actually ships on the stated H2 2027 timeline, and whether anyone outside the company benchmarks that greater-than-10,000-tokens-per-second claim. Whether Castelion draws down its $250 million revolving credit facility as Blackbeard production scales, and whether more government contracts land against its stated run rate. And whether Muon Space’s San Jose facility actually ramps toward 500 satellites a year by 2027, along with whether Google makes any further moves into the satellite-manufacturing stack beyond this one check. Each of those is a plant, a shipment, or a contract, not a valuation headline, which is exactly the kind of evidence this new breed of capital is supposed to be buying in the first place.
Clement Chen is a startup founder, advisor, and investor who writes about deeptech and building companies at clementchen.co .