China controls roughly 70% of rare earth production and about 90% of processing. That single fact sits at the bottom of Bessemer's five-frontier defense roadmap, listed as Frontier 5, and it's the one that quietly breaks the whole optimistic frame.
But start where Bessemer starts, because the framing is earned.
"Defense tech has advanced more in the past twenty-four months than in the previous three decades," the firm wrote this week, and the claim is defensible. Autonomy has moved from concept to combat. AI has slotted into real Department of War workflows. Procurement reform is speeding deployment, geopolitical threats are driving budgets, and rising budgets are pulling in the AI talent that once refused to touch defense.
The Neoprimes are real too. Anduril and Saronic are scaling across domains that were experimental in early 2024. Bessemer's takeaway — "national security now runs on startup innovation" — is not hype. It's mostly description.
Bessemer's takeaway — 'national security now runs on startup innovation' — is not hype. It's mostly description.
The reindustrialization story read as a rising tide: capital rediscovering atoms after a decade in pure software. This week the tide meets a wall, and the wall is the venture model itself.
Because here's the question the roadmap glosses: of these five frontiers, which can a venture check actually fix?
That's not a mission question. It's a business-model question, and the answer splits the five cleanly in two.
Three of the frontiers are software-shaped. Autonomy moving from concept to combat. AI permeating DoW workflows. Edge and network resilience — mesh networks and GPS alternatives that keep contested environments survivable. These are frontiers where the plumbing has matured, the models have caught up, and a well-timed check compounds into a company. Anduril and Saronic exist because venture math works on them: software margins, defensible IP, deployment cycles that a fund's ten-year horizon can actually hold.
Two of the frontiers are atoms-shaped. And atoms don't care about your fund's IRR.
Frontier 3 is advanced manufacturing, and Bessemer names the wound directly: global conflicts have exposed a structural weakness — the West can't surge munitions and energetics fast enough. That's not a software gap. That's a physical capacity gap, measured in propellant lines, forging presses, and years of permitting.
Frontier 5 is energy and materials independence, and it's the starkest of all. There is no leadership in AI, defense, or clean energy, Bessemer writes, without first securing the materials that make it possible. A prerequisite for sovereignty.
Fine. But sovereignty against a ~90% processing monopoly is not a thing venture buys.
Consider the arithmetic. Set venture-scale capital against the capital intensity of rare-earth separation — and against China's reported control of about 90% of rare-earth processing.
A venture check isn't the wrong size for that problem. It's a rounding error against it.
This is the tell. When Bessemer says startups are "stepping up with low-cost, scalable production" on the munitions surge, the phrase does a lot of quiet work. Energetics manufacturing is not scalable in the software sense. You cannot A/B-test a propellant plant into existence, and no additive-manufacturing startup closes a gap that's fundamentally about physical throughput and a supply chain that runs through Chinese refineries.
Asking venture to solve energetics surge or magnet supply chains is asking capital to do the job of industrial policy. Those are problems for loan guarantees, offtake agreements, strategic stockpiles, and multi-decade government commitment — the DPA-and-Ex-Im-Bank toolkit, not the term-sheet toolkit.
The mismatch isn't a founder-quality problem. It's a structural one. Venture returns come from capital-light compounding; rare-earth processing is capital-heavy grinding against a subsidized monopoly. No amount of talent inflow changes the shape of that curve.
Which is exactly why one manufacturing bet stands out as genuinely fundable — and it's the one that refuses to be a manufacturing bet at all.
Isembard, backed by USV, is reportedly betting the future of manufacturing looks nothing like the vertically-integrated primes of the last century. Instead, per USV, it's a software-driven network stitching hundreds of small machine shops into a decentralized production model.
Read that structure carefully. Isembard isn't raising a fortune to build a plant. It's not trying to out-capitalize China on processing tonnage. It's turning manufacturing into a routing problem — matching demand to distributed, already-existing capacity through software.
That's the move that fits the venture model, because it inverts the capital intensity. The machine shops already exist; the presses are already bought; the physical CapEx is somebody else's sunk cost. What Isembard sells is the layer on top — the coordination, the network — and that layer has software economics.
It scales the way autonomy and AI workflows scale, not the way a magnet refinery scales. The atoms stay distributed and someone else's; the value accrues to the routing intelligence. That's why it's the manufacturing thesis venture can actually underwrite.
The contrast is the whole argument. Isembard works because it's a software company wearing manufacturing's clothes. A rare-earth processing startup fails as a venture bet because it's an industrial-policy problem wearing a startup's clothes.
So Bessemer's Neoprime thesis is really two theses in one coat.
The first — autonomy, AI workflows, edge resilience — is a genuine venture frontier, and the Neoprimes prove it. Money will flow there, companies will compound there, and the "twenty-four months beats three decades" claim will keep looking prescient.
The second — surge munitions capacity, rare-earth independence — is a national mobilization problem that got dressed up in a startup pitch because the startup pitch is what's fashionable this cycle. The Information notes the AI boom is pushing Silicon Valley back toward factories, drones, and shipyards. True. But enthusiasm for atoms doesn't rewrite the economics of atoms.
The danger in one roadmap is that it launders a policy failure into a market opportunity. If "national security runs on startup innovation" becomes the whole story, the frontiers that need Washington's balance sheet get handed instead to funds structurally incapable of closing them — and everyone claps at the seed round while the ~90% processing gap stays exactly where it is.
The reindustrialization turn is real. So is the money. The discipline this week is telling which frontiers a check can fix and which ones need something the venture model was never built to supply.
The tell is never the mission. It's always the business model. On four of five frontiers, watch the software. On the fifth, watch who's actually writing the checks — and notice it isn't a fund.