Newsletter · 2026-09-05 · Ashutosh Agarwal

The Weapon Is No Longer the Jet, It Is the Code Flying It - Aerospace & Defense Weekly - Week of September 5, 2026

Aerospace and Defense Weekly for the week of September 5, 2026: four operators explained how autonomy is displacing the airframe, physical-AI startups raised nearly $50 billion in six months with defense at the front of the queue, and the primes answered with a surprise long-range missile.

Aerospace & Defense Weekly

Week of September 5, 2026: The Weapon Is No Longer the Jet, It Is the Code Flying It

Last week we complained that no builders showed up: plenty of clever commentary, but no one who actually makes the machines. This week was the opposite, and it was a treat. Four genuine operators sat down and talked shop: the CEO of a European defense company whose AI just out-dogfought human fighter pilots over the Baltic Sea; a Ukrainian founder mass-producing the software that turns a $500 drone into a precision weapon; a former Navy fighter pilot now delivering drones to the Pentagon under a radically new kind of contract; and a Marine-turned-entrepreneur rolling up the unglamorous middle of the defense industry one family business at a time.

Put their conversations next to the money story, physical-machine startups just raised nearly $50 billion in six months with aerospace and defense the top two categories, and a single picture snaps into focus. The cash is real and enormous. But the thing the cash is chasing has changed. For thirty years, "defense" meant a small number of exquisite, billion-dollar platforms. Increasingly it means cheap hardware built by the thousand, with all the cleverness moved into software and autonomy. This week the people actually building that future explained, in plain terms and with real numbers, how it works and why it's happening now.

One habit we keep, because it changes how much a claim is worth: we tell you whether a speaker is an operator, someone who builds the thing, runs the company, or buys it for the government, or a pundit, analyst, or former insider talking from the outside. Both are useful. They are not the same kind of evidence. A founder telling you his factory makes 3,000 drones a month is worth more than an investor guessing that it might.

1. The money: $50 billion in six months, and defense is at the front of the line

Start with the number that frames everything else. On This Week in Startups (September 2), the panel, all venture investors with money at stake rather than operators, walked through a chart that stopped the conversation. "Physical AI companies now have raised almost $50 billion in venture funding," said host Jason Calacanis, "in the first half of 2026, up 80% from 2025. Six months of this year outraised 2022, 2023, and 2024 combined." "Physical AI" is the industry's phrase for startups building actual machines that sense and act in the real world, drones, robots, rockets, as opposed to pure software. And, Calacanis added, "aerospace, defense are the top two categories, obviously." One version of the chart put it at "450 deals and $45 billion in physical AI in the first half of 26. Back in 23, it was not even five."

The most interesting explanation for why came from Caitlin, a venture investor who spent more than a decade running people teams before moving into investing. She judges bubbles, she said, by watching where the best engineers want to work. "10 years ago, the smartest engineer I met, they wanted to work on ads... Not because they wanted to work on ads, but because that's where the money was." Today, "that engineer, she wants to build rocket engines in Moses Lake, Washington... Because she wants to feel connected to humanity, to a badass mission, to something that is meaningful." When talent that used to optimize click-through rates starts moving to hard physical problems, she argued, that's a real signal, not just a funding fad.

The panel also flagged a shift in how the U.S. government itself is buying software. The Pentagon's internal portal for its roughly 3 million military users already ran Google's Gemini; it is now adding two more, OpenAI's ChatGPT for "administration, document management, logistics, policy work on unclassified data," and Elon Musk's Grok "for government via Starshield" (the SpaceX secure-communications service) for "faster ops analysis, supply chain and acquisition work." And Caitlin put her finger on the trap now facing the startups that won all these contracts: "the bottleneck used to be winning the contract, but now it's really just surviving one." A 30-person company that lands a federal deal, she warned, then has to actually deliver against it at a scale agencies used to expect only from 100-person-plus vendors, and prove the revenue will recur.

Why it matters: the demand debate is over; this is the clearest single data point yet on just how much capital is flooding in, and that defense sits at the very front of the queue. But the same conversation carries the warning that should temper any euphoria: a signed government contract is the start of the hard part, not the end, and a lot of these newly-funded names will struggle to deliver at scale. Keep both halves in mind.

2. "That turns out to be bullshit": a defense CEO on the week the AI beat the pilots

If you read one thing this week, make it the conversation with Torsten Reil, co-founder and CEO of the European defense company Helsing, on David Senra (August 26). Reil is a genuine operator, running a company of roughly 1,300 people, and what he described is the sharpest illustration all week of the "software is the weapon" thesis.

Helsing set out two years ago, he said, to build "a superhuman fighter pilot": an AI, with no human in the plane, for beyond-visual-range air combat, the most advanced kind. He called that kind of dogfighting "essentially chess in the sky." The conventional wisdom, he said, was that this was the one thing a machine could never do: "we'd been told by pilots, active pilots that this is the kind of last bastion of air combat... nothing will ever replace the intuition that a human has in the sky." His verdict: "And that turns out to be bullshit. The AI is much, much better."

How they did it sounds like a video-game studio more than a weapons lab, which is roughly what it is, since Reil's background is in AI-driven game physics. Helsing built "an extremely high performance simulator that runs 20, 25,000 times real-time speed" and trained a neural network with reinforcement learning, the same broad approach DeepMind used to master Go. Then they proved it in the real world: "last year we flew a real fighter jet... the Saab Gripen over the Baltic Sea. And when I say we, our AI did and took full control of the plane, in air combat against another Gripen, which was human controlled." Three scenarios, all against a human pilot. They named the system Centaur, and Reil says it now works "in swarms" of fighter jets, not just one-on-one.

The strategic conclusion he drew is the part investors should sit with: "crewed fighter jets were going to be very difficult to deploy in the future because you'll end up facing these swarms of autonomous fighter jets with superhuman capabilities." And on the pace of it all, a line worth taping to a monitor: "back in the old days... everything was always five years away... It's now the exact opposite. When you think something is five years away, it's actually usually now two years away."

Rather than partner with a traditional prime, too slow in his view, Helsing bought its own aircraft manufacturer outside Munich, runway and hangars included, and is building its own jet, the CA-1 Europa, with Centaur as the core. It flies for the first time next year. That is a startup deciding to become an aircraft OEM from scratch, on the theory that it can "leapfrog everyone in this category."

The rest of the interview filled in a genuine industrial story, not just a demo:

• Mass production is the whole game. Helsing's HX2 is a mass-manufactured strike drone with a 2 to 4 kilogram explosive payload and a range of about 100 kilometers. In July and August of last year the company went "from dozens of HX2 per month to a baseline of 1,000," and says a single small factory "can now surge to 3,000 a month." Reil's design philosophy is to build hardware that is "good enough" and cheap, and move the sophistication into software: simpler sensors and looser manufacturing tolerances, with AI recovering the accuracy. "You have to think it through as one unified system."
• The bottleneck isn't the factory, it's the contracts. Strikingly, Reil said manufacturing speed isn't the constraint today: "the bottleneck right now is the big programs need to come. So the first ones are coming. We just closed a one billion framework contract in Germany, but they now need to come everywhere." Once the orders land, "we can crank up manufacturing accordingly."
• A record-fast win. Helsing won a German program of record to upgrade the Eurofighter's electronic-warfare systems, the gear that detects and interprets enemy anti-aircraft radars, beating larger consortia alongside its partner Saab. Reil claims Helsing went "from founding to program of record" in "two years and four months," which he believes is the fastest ever.
• The budget behind it all. Germany's defense budget, he noted, is "going from 52 billion euros last year to originally I think 153 billion in 2029... I think it's gone up again to over 180 billion." Because troop numbers won't rise much, "most of the money is going to go on to new equipment." He also flagged the political reality behind it: Germany is "a pacifist society" whose post-war identity was "never again," now forced by events to rethink, and, he argued, one of the countries with "the biggest growing defense budget" and "one of the most thriving new defense ecosystems in the world."

Why it matters: this is the operator version of the argument the pundits keep gesturing at. Autonomy isn't a slide in a pitch deck; it's a neural network that already beat human pilots in a real jet, attached to a company producing thousands of strike drones a month and winning big European programs faster than anyone thought possible. If Reil is even roughly right that crewed fighters become "very difficult to deploy," that is a long-run question mark over the single most valuable product lines at the traditional primes, and a reason the software layer, not just the airframe, is where value is migrating. His "one billion contract, but they need to come everywhere" line is also the cleanest statement of the sector's real gating factor: Western capacity to build is arriving faster than Western willingness to order.

3. How a $500 drone kills a $5 million tank: the math of autonomy

The other must-listen operator conversation came from Yaroslav Azhnyuk, founder of the Ukrainian company The Fourth Law, on Eye On A.I. (August 31). Where Reil talked jets, Azhnyuk talked about the cheap frontline drones being used by the million in Ukraine, and made the single most memorable economic argument of the week.

His provocation, and the episode's title: "within 5 to 10 years it will become unethical to use weapons without AI... Because those that are less lethal and non-AI weapons will be more likely to cause collateral damage, not less." His reframing of the whole debate: "There's lots of conversation about AI and defense. What we should be really talking about, I think, is software and defense."

Then the numbers, which are worth walking through slowly because they explain why every military on earth is suddenly interested in this. A basic frontline FPV drone, the small first-person-view quadcopter that has come to define this war, costs about $100. But its hit rate is low: "what is its effectiveness? Well, it's between 10 and 20%," meaning only one in five to one in ten actually strikes its target. Add an autonomy module, Azhnyuk's product, and that effectiveness rises "two to four times even with level one autonomy," the simplest kind, where the pilot locks a target and the drone finishes the job on its own. His summary of the leverage involved: "you can take out a $5 million tank with a drone costing $500. And if you can add autonomy to that drone for the cost of a couple hundred dollars on top... autonomy itself makes every drone usable by a hundred times as many people," because you no longer need a trained pilot for the final approach. He stacked the multipliers into a deliberately dizzying figure: "10 million, 100 million times more" cost-effective than the systems being replaced.

The Fourth Law's strategy is telling. Rather than build the whole drone, it sells autonomy modules that snap onto other people's hardware. Azhnyuk called it "the Android strategy," many teams, many modules, one ecosystem, and contrasted it with the "Apple strategy" of full vertical integration he attributes to former Google CEO Eric Schmidt's defense venture (which he named Swift), whose "narrow" interceptor drones he said are "far ahead of Anduril or Alsa" and have started "selling into Poland and Romania." He was candid that Schmidt's operation is "our main competitor slash peer."

A few more grounded details from an operator actually in the fight:

• He laid out a five-level scale of drone autonomy modeled on self-driving cars (level one: locks and hits a target you designate; level five: takes off, finds, strikes, and lands entirely on its own), plus an eight-dimension framework for a whole autonomous battlefield. The key constraint is cost: to be used on a $100 drone "manufactured in millions," the autonomy has to be cheap. "If the system that allows the visual autonomy costs even a couple thousand dollars, it will be difficult to use."
• A hard supply-chain fact: "there is no semiconductor manufacturing in Ukraine." Circuit-board assembly, yes; but the actual chips and camera sensors must be imported from foreign fabs. The Fourth Law has plans to build a thermal-sensor factory, but it's not built yet.
• His big-picture call: "the world will be rearming for the next couple of decades, driven by supply chain instability, advances in drones and robotics, and AI," and "Ukraine and Ukrainians will play a critical role in the rearmament of the West." His warning on timing: a 2026 ceasefire "is not after the war. After the war is when the global arms race is going to extend."

Why it matters: this is the clearest cost accounting anyone gave all week for why cheap-mass-plus-software beats exquisite platforms in a war of attrition, and why the value is in the autonomy layer, not the airframe. The module-versus-integrated-stack debate (Android vs. Apple) is the real strategic question for defense-tech investors: does the winner own the whole drone, or the software that makes everyone's drones smart? And the Ukraine chip point is a quiet reminder that even the scrappiest drone war still runs, ultimately, on semiconductors made somewhere else.

4. Washington's new trick for buying drones fast: pay up front

If Sections 2 and 3 are about what to build, this one is about how the Pentagon is learning to buy it. On Emerging Tech Horizons (September 2), Paul Fermo, president of Robinson Unmanned and a former Navy F-18 fighter pilot who also served on the Joint Staff writing requirements, explained the Pentagon's "Drone Dominance" program and the contracting idea underneath it that could reshape how the whole industry gets funded.

The idea is the advanced market commitment. Traditionally, Fermo explained, the drone business ran on uncertainty: "the acquisition of unmanned systems has been uncertain, episodic, volume limited," which meant companies couldn't justify investing in factories and tooling. Drone Dominance flips that. In his words, the government said: "we're going to buy this many systems. We're going to pay this much for them. We're going to guarantee you. What you need to do is show up and perform." Because the order is guaranteed on the back end, companies can "invest in tooling, industrial capacity and inventory" up front. Robinson was one of the winners on the program's "Gauntlet One" selection and is "actively delivering systems to the government right now."

Fermo's careful optimism is worth quoting, because it comes with the scar tissue of someone who has watched Pentagon fads come and go. The host, a self-described "grumpy ex-Hill staffer," asked point blank whether this would actually stick or was just another senior-leader initiative that never survives into the real budget. Fermo's answer: he's optimistic, and there's early evidence, because the Defense Innovation Unit (DIU) "just came out yesterday with a new solicitation for something very similar," borrowing lessons and capabilities from Drone Dominance. In other words, the model is already spreading.

He also made the dual-use case concrete. Robinson builds everything from a 249-gram nano drone up to uncrewed versions of the Robinson R44 and R66 helicopters; the R66 "turbine truck," built with Sikorsky, is part of the Marine Corps' MARVEL contested-logistics program. The same aircraft, he noted, can resupply offshore oil rigs, ferry medical organs, or evacuate battlefield casualties: "you're removing humans from an area or circumstance that may put them at risk." His view of who wins: not the company "that has the most impressive vehicle" but the one that "combines reliable aircraft, autonomy, payloads, communication systems, data exploitation, and scalable manufacturing," the same all-of-it-together argument Reil made.

Why it matters: the advanced market commitment is the boring-but-crucial plumbing that makes the whole "cheap mass" doctrine financeable. It's the government solving the exact problem Reil named: orders need to actually show up so factories can scale. If this contracting model genuinely spreads from Drone Dominance to DIU and into the regular service budgets, it changes the risk profile of every drone manufacturer, because guaranteed volume is what turns a prototype shop into an industrial business. Watch whether it survives contact with the normal Pentagon budget process. Fermo is hopeful, but even he framed it as an open question.

5. Europe's checkbook: 5% of GDP, and pension funds told to "get off their high horses"

For the macro backdrop to all of this, Jones Day Talks (September 3) hosted Alex Chalk, the former UK Lord Chancellor and Minister of State for Defence Procurement, now a lawyer at Jones Day. He's an insider-turned-adviser, a commentator with real hands-on experience buying weapons but no longer an operator, and he laid out the money and the mechanics of Europe's rearmament.

The headline figures: at the 2025 NATO summit in The Hague, allies committed to spending 5% of GDP on defense, and combined allied defense spending already "increased by nearly 20% in real terms compared to 2024." What changed? "It is principally Russia," Chalk said flatly; without the full-scale invasion, "I don't think this would have had the level of priority." He described Britain living with "constant hybrid warfare attacks": undersea cables "being mapped by suspicious ships," a Russian "shadow fleet," and intelligence chiefs describing "daily gray zone conflict."

He was refreshingly blunt about the gap between spending headlines and real capability. Britain will spend "something like 90 billion U.S. dollars" this year, about 2.5% of GDP, but "a lot of that money is going on renewing our nuclear deterrent" (four submarines, new warheads). Strip that out, and spending on conventional forces is "around about 1.7% of GDP, which is way, way too low." By contrast, Baltic states like Estonia, right on Russia's border, are "spending like 5% of GDP, sometimes even more." His read on whether NATO is ready: "it is behind," largely because of "the pace and the scale of the U.S. withdrawal," which "could accelerate." He thinks the U.S. president is "well within his rights" to call the Europeans "free riders."

Two points with a direct market edge:

• The capital taboo is breaking. Chalk described a "cultural squeamishness" in Europe about investing in weapons that is now reversing hard. Britain's Financial Conduct Authority "issued a statement clarifying that there's nothing in our rules, including those related to sustainability, that prevents investment in defense companies," and a minister he knows "told UK pension funds to get off their high horses and invest in British defense," invoking their "patriotic duty." His verdict: "this is stuff that would not have been said five years ago." That's a real shift in the pool of capital allowed to flow into the sector.
• The buying is pivoting to drones. Where's the money going? Chalk pointed to Britain's Defense Investment Plan and its "top level newspaper headline retail takeaway": "the UK is pivoting away from expensive big platforms towards drones." Concretely, the Navy has "decided to bin off" replacing its roughly £1 to £1.5 billion destroyers in favor of "a common vessel platform... used as a mothership for drones," while still funding the nuclear deterrent and a sixth-generation fighter being built with Japan and Italy. The lesson, he said, comes "from the tragic testbed of Ukraine."

He added one practical note for anyone imagining NATO itself as a big customer: don't. NATO's own common budget is only "around about five billion euros" total. "The really big opportunities are in the member states," Germany, Britain, Poland, not at NATO headquarters.

Why it matters: this is the demand side of every operator story above, quantified by someone who used to sign the checks. The 5%-of-GDP commitment and the near-20% real spending jump are the fuel; the pension-fund attitude shift widens the investor base that's allowed to buy in; and the explicit pivot "away from expensive big platforms towards drones" is national policy now, not a forecast. Chalk's 1.7%-on-conventional-forces caveat is the honest asterisk: a lot of the eye-popping top-line spending disappears into nuclear renewal before it ever reaches the conventional kit the drone-makers and mid-tier suppliers actually sell.

6. The incumbents strike back: a surprise missile and a very good week for RTX and Lockheed

Amid all the disruption talk, the traditional primes had a genuinely eventful week, and it came from the missiles, the corner of defense where cheap mass doesn't apply and the incumbents still rule. On Aviation Week's Check 6 Podcast (August 31), the magazine's editors debriefed the Navy's Tailhook symposium, where the Navy sprang a genuine surprise on the last day.

The surprise was the AIM-424 Malice, a long-range air-to-air missile from RTX (Raytheon) that "nobody really knew about." The specs are striking: a range of "more than 250 nautical miles," carried internally inside an F-35C's weapons bay despite being "13 and a half feet long" and weighing "about 1,500 pounds." As one editor put it, "a giant missile and still being able to be internally carried is really remarkable." It's already been in flight testing on the F-18 since April.

The editors placed it in a clear competitive arc. Malice is "the third new air-to-air intercept missile revealed over the past 10 years," each longer-ranged than the last, and each a response to China and Russia:

• The AIM-260 (Lockheed Martin), the answer to China's PL-15;
• The AIM-174B "Gunslinger" (RTX), an air-launched adaptation of the SM-6;
• And now the AIM-424 Malice (RTX).

The logic, as editor Steve Trimble put it: "if you're in an air-to-air fight, you don't want the other guy to have a missile that can go farther than your missile, because they can shoot you before you can get in range to shoot them." The adversary benchmarks he named: China's PL-15 and PL-17 (claimed 265 nautical miles) and Russia's R-37M (220 to 240).

Beyond air-to-air, the Navy confirmed its MACE program (Multi-mission Affordable Capacity Effector) is Castellion's "Blackbeard" air-launched hypersonic missile, with a first test shot targeted "by the end of this calendar year" and a budget contemplating "thousands and thousands" of missiles. There's also a brand-new concept called CAM (Compact Air-to-Air Missile) for very close-in self-defense. Trimble's aside captured the moment: "for 30 years, if you asked what kind of air-to-air missiles do we have, there was two. It was AIM-9 and AIM-120. And now we've got several. So it's a really interesting time." Separately, the long-awaited decision on the Navy's next-generation F/A-XX fighter had still not come; the editors' reporting pointed to "likely end of next month," with the delay piling pressure on the existing F-35 and Super Hornet fleets.

For a longer-term valuation lens on those same primes, The Investing for Beginners Podcast (August 24) had two retail-focused analysts rank them on the numbers. Their scorecard:

• Most capital-efficient, by a mile: Lockheed Martin, with return on invested capital "in the 18, 22% range for the last six years." (Return on invested capital measures how much profit a company squeezes from the money it puts to work.)
• Northrop Grumman and General Dynamics: "low double digits."
• RTX and L3 Harris: faster revenue growers, "but they're also growing through acquisition," which drags down their capital efficiency, so the hosts put them "on the bottom of the pile."
• Boeing: finally back to positive ROIC "after years and years of losses," rebounding, but still "too hard pile."

Their overall ranking put General Dynamics first, as the "sole provider" of the M1 Abrams tank, the Bradley fighting vehicle, and the Paladin howitzer (America's tracked ground forces), plus work on soldier heads-up displays to eliminate friendly-fire deaths, then Northrop (sole maker of the roughly $2-billion-a-unit B-2 bomber), then Lockheed (sole maker of the F-35, F-22, and F-16). Their governing mental model: the U.S. government deliberately keeps multiple primes alive and won't "let one company become the Google of defense contractors," so there are "multiple winners" and no single monopolist.

Why it matters: the missile news is a useful corrective to the disruption narrative. In munitions, especially exotic, long-range, hypersonic weapons, cheap mass doesn't apply, and RTX and Lockheed are cementing exactly the high-end franchises that a war-planning Pentagon most urgently wants to buy in quantity. Malice, the Gunslinger, the AIM-260 and MACE are multi-year production programs landing squarely with the incumbents. The counterpoint, from the valuation hosts: not all primes are equal. Lockheed's capital efficiency stands alone, the acquisition-fueled growers screen worse, and Boeing remains a "prove it" story. The autonomy wave is real, but so is the incumbents' grip on the parts of the arsenal that can't be crowd-sourced.

7. The quiet roll-up in the middle, and a note of caution

Two conversations pointed at where value may be created away from both the flashy startups and the giant primes: the neglected middle of the industry.

On The Aerospace Executive Podcast (September 3), John Albers, a 24-year Marine Corps veteran, described building Albers Aerospace as a "next-gen prime" aimed at "the space where the primes don't want to play." He grew the company by himself to about "$12 million a year in revenue" and roughly 80 people, then started acquiring smaller firms. His thesis is a clever inversion: the decades of "over consolidation at the top" of defense, which he traced to defense-executive Norm Augustine's famous 1990s "Last Supper" when the Pentagon told contractors to merge, left a gap in the middle. "Ironically, when you think of over consolidation, there's a roll-up opportunity down low." Add in demographics, aging baby-boomer owners whose "kids don't want anything to do with the business," and there's a supply of family firms to buy "from founders who care." He confirmed the demand shift the whole newsletter keeps circling: "the big prime Lockheed Martin, Northrop Grumman gravy train seems like it's shifting focus to smaller companies" that can "go fast" at "low overhead," and, in his blunt phrase, "Ukraine changed everything... speed is life."

The consolidation theme showed up again, from the investor side, on The Twenty Minute VC (20VC) (August 27). One VC predicted that in defense, "there'll be two or three companies that get critical mass in public at huge scale, and they'll hoover up the rest of us," precisely because "unlike tech where a single product can kill it... it's a portfolio of products that it takes to survive the interaction with the Pentagon." But this episode is also where the week's most useful skepticism lived: its title flagged "Why Customer Service, Defence and Robotics are Overinflated," and Jason Calacanis, while marveling that "AI created it, defense budgets enhanced it," was clear the sector is now priced for perfection. Worth holding onto as a counterweight to Section 1's euphoria.

And for a ground-level view of why all this is hard, Behind the Win (August 25) featured Josh Johnson, a defense-and-security director at consultancy GuideHouse and a retired Air Force major. He called the U.S. defense industrial base "capable, but under immense pressure," naming workforce availability (Northrop, he said, flagged it directly at a governor's council), supply chain, and acquisition speed. His frustrations were concrete: the Federal Acquisition Regulation is "5,000 pages deep," a cybersecurity certification (CMMC level two) cost his firm "almost 200 grand" and "two years of our lives," and "if it takes us 15 years to field a truck, that's unacceptable." His forward view: over five years, expect routine AI-assisted decision-making, "digital engineering" becoming standard, and an increasingly networked base of primes, mid-tiers, startups, universities, and tech firms working together.

Why it matters: the flashiest money is chasing the disruptors and the safest money sits with the primes, but Albers is describing a third path, an unglamorous roll-up of the mid-tier suppliers the Pentagon now wants for speed, at valuations far below the hyped names. The 20VC prediction (two or three consolidators eventually absorb the rest) and Albers' bottom-up roll-up are two ends of the same consolidation story. And Johnson's litany of red tape and workforce gaps is the reality check under all of it: the demand and the capital are real, but the acquisition system, the cyber-compliance burden, and the shortage of skilled workers are the friction that will decide who actually delivers.

The bottom line

This was the week the operators showed up and made the abstract concrete. The money story is now beyond dispute: nearly $50 billion into physical-AI startups in six months, with aerospace and defense the top two categories, and European allies committing to 5% of GDP with spending already up nearly 20% in real terms. But the more important story is what the money is buying, and two founders spelled it out with hard numbers: an AI that out-flew human fighter pilots in a real jet over the Baltic, attached to a company already making up to 3,000 strike drones a month; and a Ukrainian software-maker whose autonomy modules turn a $100 drone with a one-in-five hit rate into a far more lethal weapon for a couple hundred dollars more. The weapon, increasingly, is the code.

Around that core, the plumbing is being rebuilt to match: the Pentagon's advanced market commitments guarantee orders so factories can scale; European regulators and ministers are actively pushing pension money into defense that was taboo five years ago; and both Britain and Germany are steering their budgets "away from expensive big platforms towards drones." The incumbents, meanwhile, are far from finished. RTX and Lockheed just showed off a new generation of long-range missiles the disruptors can't touch, and Lockheed's capital efficiency still stands alone. And in the overlooked middle, operators like Albers are quietly rolling up the family-owned suppliers the Pentagon now prizes for speed.

The one honest note of caution came, fittingly, from the investors themselves: defense is now "overinflated" in some corners, a signed government contract is where the hard part begins, and the acquisition system's 5,000 pages of rules haven't gone anywhere. The demand is real. The capital is real. The technology, this week, was undeniably real. The open question is whether the West can turn all of it into delivered hardware, at scale and on time, before it's actually needed.