‍

The Asymmetric Ledger of Entropy
There's one key question in this quarter's letter that’s relevant for every individual, investor, company, country and civilization: what did you choose to build and protect, and did you understand what it cost? There's also one key idea that competitively advantages all who understand it and disadvantages all who don't, and equally applies to business and biology, technology and treaties: entropy.

Consider what it costs to build a skyscraper, car or jet. A Manhattan residential tower takes 2,500 tons of material, 5 miles of copper wire and 18 months of coordinated labor across 15 trades. A Toyota Camry has 30,000 parts machined to tolerances smaller than a red blood cell, while an F-35 runs on 8 million lines of code built on top of 300,000 parts from 1,500 suppliers across our allied industrial base. Gaps are measured to thousandths of an inch. Why? Because at Mach 1.6, imprecision isn't a flaw, it's a funeral. Each is a monument to negentropy: energy converted into order, chaos beaten back by design. As Radiohead sang: Everything in Its Right Place.

Now consider what it takes to reverse those creations. A Mark 84 bomb detonates faster than our nerves register pain. The First Law of Thermodynamics reminds us the blast wave doesn't destroy atoms—it just rearranges them. Engineered order becomes undifferentiated rubble. Cathedrals were built over half a century; Dresden burned in half a day. Aleppo's centuries-old souk fell in weeks. The asymmetry to grok is this: assembly is slow, expensive, cooperative and fragile but disassembly is fast, cheap, unilateral and total—that’s the asymmetric ledger of entropy.

Everything civilization has ever produced—every building, machine, treaty, and relationship—is a local, hard-fought victory against entropy. Entropy is the universe's default pull toward disorder, diffusion, the undifferentiated sameness of equilibrium, and even death. We take raw materials, and with enormous energy, arrange them into complex configurations that work: their function is the value. Destroy the arrangement, and you destroy everything except the raw mass. The atoms remain, the worth is gone.

In preview, this quarter's letter starts with the tale of two parabolas and why we must defend against entropy’s asymmetric ledger. Then we discuss the high costs of low-cost offense, the hidden macro consequences of buffers in complex systems, the negentropy of biological intelligence, and the asymmetric upside of contrarian theses in narrative-driven, homogeneous markets.

A Tale of Two Parabolas
On the first of April, nearly 9 million pounds of thrust pushed a 322-foot column of controlled combustion straight up from Florida's coast. Within 23 hours, the four astronauts aboard Artemis II broke free of the gravitational envelope that's held every human being on Earth for 54 years. Courage, chemistry and capital to the tune of $4 billion helped them reach escape velocity and extend the parabola of human ambition as they transited to the moon and back. The crew re-entered Earth at 25,000 miles per hour and the rocket sank into the Atlantic, never to fly again.

Now invert the parabola and rotate the vector from celestial to terrestrial. The same physics—combustion, thrust, ballistic trajectory and kinetic energy at terminal velocity—is being deployed across two theaters of war not to ascend to the stars but to descend into scars, reducing civilization to rubble.

Above Ukraine, Russian missiles dive at speeds up to Mach 10 with the kinetic energy of a freight train compressed into a single point. The country produces 200 each month, enabled by CNC machines from China, Taiwan and Belarus. To intercept six of these, Ukraine must fire up to 18 PAC-3 Patriots at a cost of $4 million each. It’s a grave calculus: $72 million of defensive order to neutralize $15 million of offensive ordinance. Lockheed makes around 600 PAC-3s a year while our enemy fires that many in a quarter. Russia is building an industrial conveyor belt of destruction at a pace that mercilessly mocks the West's peacetime procurement rhythms. In the Middle East, this asymmetric calculus is even more grave. The quip is we’re launching Ferraris at frisbees. Shahed drones cost less than $40,000 each and Iran produces 10,000 of them per month. In the first week of the current Iran conflict, the U.S. consumed 170 Tomahawks in just 100 hours—three times more than what the Pentagon requested for the entire year.

In a recent Lux conversation for our Limited Partners, Lux venture partner Brett McGurk and former Israeli Minister of Defense and retired IDF General Yoav Gallant argued the current Israel-Iran campaign is the most consequential strategic inflection in the region since 1979. The cost imbalance—roughly $100 million of Iranian munitions met by $3 billion of interceptors per wave—is reshaping the entire defense-tech ecosystem and reinforcing our long-held thesis around layered defense, counter-drone systems and mass-producible munitions.

Israel's layered defense including Iron Dome, David's Sling, Arrow 2, and Arrow 3 is the most battle-tested missile shield in history, but it’s being ground down by asymmetric economics. Israel's new laser, Iron Beam, intercepts at $2 to $5 per shot—the first structural correction to the attacker-defender cost curve in a generation. Even so, offense and defense never stop evolving, which is why we are funding cutting-edge defense ventures like Kela, Covenant, Traysar, Tenzai and more in Israel. McGurk and Gallant believe Iran's missile buildup has been substantially degraded while U.S.-Arab-Israeli integration has tightened. Yet, both flagged the same forward risk: if the U.S. disengages once the kinetic phase ends, China will fill the vacuum.

Every civilization is, at root, an energy system: the story of human progress is a story of harnessing ever more joules per capita applied ever more efficiently toward ever more complex ends. This cuts in both directions. The same thermodynamic mastery that lifts a crewed capsule toward lunar orbit also delivers a hypersonic warhead through the roof of a Kyiv apartment block. This year, Russia will spend the cost of one Artemis launch to produce approximately 2,500 instruments of destruction. One expenditure of energy creates the raw infrastructure of a spacefaring species (helped by Impulse Space, Varda, Astranis, Observable and more), while the other creates ruins, refugees and maximum entropy. Same engineering, opposite directions.

Energy is agnostic and physics keeps no moral ledger—only we can. The question is which parabola we choose. With bad actors as aggressors, the painful answer is it requires us to create the destructive force capable of defending the creative force—hypersonics protect hyperspace. Peace is negentropy, diplomacy is maintenance, and trust is the invisible web holding it all in place. War breaks out when the enemy's version of order is intolerable and they're intransigent to rearrange it. American military power—which protects our GLP-1s, GPUs, and GPS satellites—rests on a single credible promise: that we can use the most highly ordered technology ever assembled to swiftly and asymmetrically disassemble, disorder and destroy our adversaries.

The Arsenal of Arithmetic
The arsenal of democracy has become an arsenal of arithmetic, and the numbers don't add up. America accounts for 0.1% of global shipbuilding; China, 53%. A single Chinese shipyard launched more commercial tonnage in 2024 than the entire American shipbuilding industry has produced since World War II. The defense industrial workforce has shed nearly two-thirds of its headcount since the mid-1980s, and we consolidated 51 prime defense contractors into five. It’s little wonder then that only one production line was left for large-diameter solid rocket motors.

Efficiency is just fragility that hasn’t been stress-tested yet. In 16 days of Operation Epic Fury against Iran, the U.S. consumed roughly 30% of its Tomahawks and 40% of its THAAD interceptors, and nearly a third of its Massive Ordnance Penetrators (only 25 of which are available).

The most honest answer as to why the war in Ukraine grinds on is that Russia has discovered patience is a kind of leverage. Weapon stockpiles are the slowly accreted order of an industrial base compressed into pallets, factories and trained machinists. Moscow's strategy is to let the second law of thermodynamics do the work its army cannot. Every interceptor we fire is order spent and every Shahed they launch is disorder seeded. Properly framed, strategic patience is just outsourcing entropy to your adversary.

The turn, when it comes, will come from physics and the asymmetric ledger of entropy finally being respected. The Department of War has reverse-engineered Iran's Shahed-136 into a domestic equivalent at $40,000 each—about 2% of the cost of the Tomahawk it replaces. The doctrine has been formalized as "Affordable Mass," and Anduril's Barracuda-500 sits in the same family.

Thankfully, the country asking whether it can rebuild its defense industrial base is the same country whose commercial technology sector is the most innovative on earth. The bottleneck isn't the factory floor but the distance between knowing and doing—where every previous arsenal-of-democracy story began, and where this one will be decided.

Few people have lived this equilibrium between knowing and doing and been in positions of decision-making from both sides quite like Lux’s newest venture partner Zach Iscol. As a decorated Marine veteran fighting in Fallujah, he saw the asymmetries of entropy industrialized: order undone in seconds by people who had decided it should be. As Commissioner of NYC Emergency Management, he ran the other direction of the same physics: leading a small team protecting a city of 8 million through a pandemic, blackouts, blizzards, active shooters and so much more. Zach is already working closely with our team and our portfolio to seize on asymmetries, seeking out unusual strategic opportunities or eliminating risks before they metastasize into entropy.

The Deceptions of Macro Buffers
A key frame we use in building tech investment theses is abundance versus scarcity. The question isn't just what happens when an invention arrives, but when everyone has one. With geopolitical risk, it’s the inverse. What happens when everyone doesn't have food, energy, security or shelter? The Strait of Hormuz is testing this query now.

The molecules disagree with the market consensus, an oil story awaiting a ceasefire-shaped exit. One strait, throttled for 60 days, ripples through chemistries most people don't even know they rely on. Helium, not substitutable below four degrees Kelvin, keeps fabs and MRI scanners from going dark. Sulfur becomes sulfuric acid becomes phosphate fertilizer becomes next season's crop—or doesn't. Fertilizer is where the cascade can become combustible. 30% of global fertilizer trade and 50% of seaborne sulfur transit the Strait. Bangladesh has shuttered fertilizer plants; India's urea production is down a third. The crop calendar has a cruel cutoff: miss the window and no capital can undo the loss. Add in an underappreciated 70% probability of an El Niño this year (with NOAA assigning 25% odds it runs hot), and we could see a synchronized harvest failure across regions that are really import-dependent.

Bread loaves become bread lines become bread riots—and a possible proximate trigger of populous uprisings and regimes falling. The 2010 Russian wheat failure dominoed into the Arab Spring. Egypt today is the world's largest wheat importer and Indonesia is on track to overtake it. We like to quote that capital goes where it's welcome and stays where it's well treated. Hunger does not stay where it starts, but moves on foot to polling sites, onto the streets of cities and across borders. Today’s fertilizer supply shock is tomorrow’s migration and political-violence shock, a spring coiled with explosive kinetic energy. It’s future entropy not fully priced into the markets.

Buffers, if inadequate, are the great deceivers of complex systems. The strait closes but the price doesn't move; the fab stencils chips and the B200 ships on time, until it doesn't. Failure comes from a failure to imagine failure: what markets read as resilience may be inventory silently drawn down, rearview dashboards reporting conditions that no longer exist. Entropy here isn't noise that we measure but noise that is suppressed. When slack runs out and the buffer empties, the recalibration and reset is discontinuous—more like a step-function jump to a reality there all along but unseen. There are several buffers emptying in parallel: oil-in-transit, petroleum derivatives, critical minerals, industrial labor, free passage through other people's chokepoints. Each flashed green on its own dashboard, but the dashboard was last to know. What was once abundant is now scarce.

The (AI)symmetry of the Twenty-Watt Mind
The same physics that builds the car or crashes it, that arcs the parabola toward the moon or toward the rubble, also offers a parallel asymmetric ledger between two divergent architectures of intelligence. The numbers are both staggering and poetic: a frontier AI model in 2026 requires 600 megawatts to train and the smallest commercial AI data center runs at 20 megawatts. Meanwhile, the human brain runs at 20 watts, constantly. Not 20 watts in some special "training mode" with 5 watts at inference, but 20 watts whether you're sleeping or solving differential equations, writing a sonnet or prompting Sonnet.

Six to nine orders of magnitude separate artificial intelligence from biological. It’s the most asymmetric ledger of all—the one between the brains we build and the brains we're born with. A toddler needs 100,000 words to begin speaking; a frontier LLM needs 15 trillion tokens. That’s 150 million times more language to do something meaningfully less impressive. To learn a new task via reinforcement learning, models need 100,000 episodes. We humans, the universe's ultimate one-shot learners, often need just one. Drosophila fruit flies count just 100,000 neurons, and yet, no drone on the planet can match its navigation, threat perception, predator evasion or spatial memory. We can train models that beat last year's benchmarks on every axis. We can't yet build a fly.

It takes 16 years to grow a human capable of complex cognitive and physical tasks and only 16 weeks to train a top LLM. Yet at the end of those 16 weeks, the LLM stops learning, while at the end of 16 years, the human is still learning, adapting and integrating. LLMs are trained like the movie title made literal—Everything Everywhere All at Once—but with zero sense of time or space. Updating them with last night's scores requires retraining them from scratch. The largest private models (5–8 trillion parameters) race toward the brain's 15 trillion synapses by brute force.

Yet, quantity of parameters isn't quality of architecture, and brute force isn't biology. That’s why we backed Flourish, which was incubated inside Lux and raised half a billion from inception. Thomas Reardon (creator of CTRL-labs that was acquired by Meta, as well as a key architect of Internet Explorer at Microsoft) and his team founded the company on a contrarian thesis that begins where the AI consensus ends: the data center capex curve is unsustainable, and thus the future of intelligence is not centralized but distributed, not detached but embodied, not trained once and frozen but continuously learning at the edge.

You can't quantize a 600-megawatt model into a 20-watt brain. You can't put a data center in a backpack, a quadcopter, a surgical tool, or a fighter cockpit that must decide in milliseconds without phoning home. The opportunity for AI at the edge will dwarf the data center AI opportunity. Markets have spent the past few years pouring hundreds of billions of dollars into ever-larger models run out of ever-larger buildings while consuming ever-larger fractions of the grid. Lux is funding the opposite: intelligence that runs on watts not gigawatts, learns from one example not trillions, and lives where the work is. A biological system has refined its principles for 600 million years against an artificial one developed for seven decades, and operating at scale for only one. The bet isn't that biology will win—it sucks at many things silicon doesn't—but that architectures moving toward biological efficiency will beat architectures moving toward biological size by raw exponentiation. Flourish took a look at the asymmetric ledger of intelligence and then inverted it.

The Asymmetry of Ideas
One corollary worth naming. As silicon scales, what becomes scarce becomes valuable—and what is becoming scarce is the human. As automated capabilities proliferate, it’s the asymmetric originality of our humanity and person-to-person connection through live presence, in-room judgment, the salon and the dinner and the Riskgaming, the eye-contact with a knowing or furtive glance (not fixed on screens) that all become more valuable, not less.

The homogeneous outputs of a handful of advanced models will not supplant the negentropy of proprietary knowledge and cultivated dissensus nor the human competitive impulse to seek non-consensus, non-obvious ideas—to believe before others understand. This is equally true of risks (fertilizer and famine) as opportunities. It was true of Nvidia in 2016, when our private-market vantage from inside Zoox revealed the pivot from chips for gaming to powering the soul of the new intelligence: simulation, AI, robotics and eventually inference and memory. We made the case publicly then when it had a $15 billion valuation, and the market has clearly caught up. The pattern repeated more recently with our view on edge inference and high-bandwidth memory. The question is always which input the consensus has not yet priced—and the answer is almost always the input the consensus hasn't yet imagined.

What unites the missile interceptor, the lunar rocket, the fertilizer cascade and the embodied neural network—beyond physics—is asymmetry. Asymmetric cost, energy, attention—and for an investor, asymmetric conviction.

The same physics is at work in markets. A market that has pushed out its discriminating participants has lower informational entropy day-to-day (fewer disagreements, tighter consensus, more momentum) but is storing volatility rather than dissipating it. Suppressed variance accumulates somewhere; Minsky said it first: stability breeds instability. Passive flows don't price securities, they receive prices. Physical-world entropy (conflicts, supply chains, breached sovereignty) is rising at precisely the time when the financial system's capacity to price it has declined. Here's a related speculation: when Buffett dies, the headline will not be greatly exaggerated, but the death of value investing will be. We can imagine a scenario where the Nasdaq is down 10–15%, the concentrated tech-heavy market index is rolling over, and discriminating stock-pickers are once again the strong preference of allocators over narrative-driven investors who rode passive indexation up. What was once dollar in, buy indiscriminately may become dollar out, sell indiscriminately.

Venture is already seeing this dynamic in slower motion. In Q1 2026, nearly 75% of all capital flowed to just five firms. Meanwhile only a few first-time funds were raised—a 75% collapse. The venture minnows are going extinct; the megas are consolidating, several openly following the private-equity playbook of seeking public-market liquidity for their own management companies. The SpaceX listing (targeting $1.5 trillion as of this writing) is the largest private-company exit ever attempted, and also the first private company to bend index-inclusion rules to force automatic buying and passive demand—a market-structure innovation as consequential, in its way, as the Falcon rocket itself.

The biggest position we hold this quarter is not a specific investment—it is dry powder. With the close of Lux IX at $1.5 billion, we have meaningful total capacity to deploy aggressively into whatever dislocations arise. Against the backdrop of a venture industry incurring the changes cited above, Lux believes disciplined, contrarian and conviction-driven capital has more asymmetric upside value now than at any point in the past decade.

We have deployed capital into three intersecting asymmetries:

The asymmetry of the physical. In America, the spine is the new defense-industrial layer the legacy primes were structurally incapable of building including: Anduril for autonomous mass and software-defined kill chains; Hadrian for the precision components the supply chain forgot how to make; Varda and Impulse Space for orbital infrastructure now strategic rather than scientific; Saildrone for persistent maritime sensing and kinetics in oceans the Navy can no longer cover by hull count alone; and Nominal for the test-and-evaluation tooling that has to exist before any of the above can be qualified. Internationally, America’s allies are rebuilding their capacity to protect liberal democracy through companies like: Cambridge Aerospace in the United Kingdom for civic security where state capacity has thinned past self-defense; Terra Industries in Nigeria against Sahel-Maghreb extremism projecting northward; EnduroSat in Bulgaria to save sovereign space at the frontier against a revisionist Russia; Kela, Covenant, Tenzai and our Israeli portfolio as the most operationally tested defense-tech ecosystem on Earth; Sakana in Japan as a hedge against the assumption that frontier AI remains North American; Onodrim and the broader European bench as the continent rediscovers the industrial competence it once outsourced and defunded. Defense tech is no longer a corner of most portfolios, but rather a thesis about which arrangement of atoms should persist into the second half of the century.

The asymmetry of the biological. Our continued conviction in the broader life sciences intersecting with the computational sciences reflects an under-appreciated truth: the same molecular biology that lets a fruit fly out-fly a billion-dollar drone now gives us the ability to engineer therapies (Aera, Atavistik, Eikon, eGenesis, Variant Bio), sense disease (Atrandi, Mendaera), decode olfaction (Osmo), automate labs (Benchling, Mendra, Multiply) and map the brain (Cajal).

The asymmetry of the artificial. Through Flourish discussed above and companies like Cognition, Hugging Face, Modal, and Together that form our broader edge-AI thesis, we’re betting that the next trillion dollars of AI value will accrue not to the companies building the largest data centers, but to the ones that figure out how not to need them. One forward speculation: the next demand shock will not be in model training, but in components enabling continuous, ambient and embodied capture supporting a proliferation of devices in what we call “lifecording.” The underlying hardware is not currently in volume production, so when the shock arrives, pricing power will surprise many, and we expect a frenzied rush to own the critical suppliers.

These three asymmetries are three facets of one core thesis: the arrangements of matter and energy that resist entropy most efficiently—physically, biologically and artifically—will define the next economic order. The ledger of order is consulted not in scenario planning but in capital allocation: which arrangements we decide to defend and which we allow to fall apart. While we don’t place fasteners, we do place capital, but the function—opposition to entropy—is the same.

Groups survive because the people in them decided that a particular piece of ground, an idea or an entire way of life was worth holding and acted accordingly. They are the doers, not the bluffers. The universe trends toward disorder, and it always wins eventually. Every building still standing, every machine still running, every peace still holding, every portfolio still compounding is a temporary, deliberate defiance of this physical law. The question for all of us twenty-watt biological machines deploying gigawatt-scale capital into a world of dueling parabolas is only ever this: what did you choose to build and protect, and did you understand what it cost? The atoms remain, the worth doesn’t, and like a premium motor car, a perfect musical composition, or a venture portfolio meaningfully constructed, the arrangement is everything. Fiat Lux.