A Quarter Century of Thanks 

We approach the end of 2025 with enormous gratitude. We’ve made extraordinary new investments, welcomed new geniuses to the Lux family, exited several companies, and returned significant capital. It’s a surreal moment for us: we viscerally remember the struggle to reach the $100 million headline of our first institutional fund, only to eventually close on $92.1 million. We are forever thankful to the many loyal partners still with us, and we are also thankful for those who passed and added more chips on our shoulders so that we can put even more chips in the pockets of those who believed. Petty, yes, but profitable for those perspicacious partners who perceived precocious pioneers and powered us with their purses. We are only just getting started and we’re still day-one hungry. 

As Lux reflects upon its 25^th birthday and America approaches its 250^th, we wanted to take stock by first returning to our origin story and strategy of multiplication, and then discussing the state of the markets and technology, and concluding by uncovering rich insights from our recent Kinetic Convening at the Lux AI Summit. 

Multiplying Value: Luck x Belief ‍

A cascade of chapters has been written about Lux’s origin story: a tiny team climbing and clawing to break in and break through from hard-scrabble, humble and scrappy beginnings on a steep fundraise slope. It feels like just yesterday we first launched Lux in the midst of the dot-com boom-bust. We were equipped with potent doses of ambition and naivety, a differentiated focus on what we thought would be new and next—the intersection of the physical, material, life and computational sciences—and a lot of luck. 

We went to raise friends and family money and joked that we had friends and we had family—but none of them had money. A series of fortuitous events led us to Bill Conway, co-founder of The Carlyle Group, who put us in business with his capital and that scarce substance that propels people through trials and tribulations: his belief. It helped mark the first of a series of Luxisms we’ve cemented in lore and lingo: the virtue that “we believe before others understand.” 

None of Lux’s past, present or future was inevitable. Had Bill not anchored us, we’d likely have sunk. Had fate forked to an alternative timeline in which he had been in a foul mood or had just received bad  news, or if we had failed to make a strong initial impression, the consequent dozens of years, hundreds of companies, thousands of founders and billions of dollars invested may have amounted to nil. Whatever the circumstances of that morning, the stochastic nature of life demands respect for luck and assiduous attention for its upshot—opportunity. Imagining our own counterfactual keeps us steady. We pay this forward with all whom we fund, forever positively multiplying their trajectories as Bill did ours. 

Multiplying Value: Early Lessons x Learnings 

As we built Lux in the very early years, we invoked the wisdom of Harry Truman, “there’s nothing new, just history you don’t know.” So we studied the history of venture capital, seeing waves of S-curves that went from nascency to growth to maturity before capital transitioned on to the next wave. The PC wave of the 1970s became the biotech buzz of the 1980s and then the media, telecom and internet boom of the 1990s to what we hypothesized would be the physical and material sciences in the 2000s. We were early and wrong—indistinguishable from each other—and learned many lessons the hard way. We overvalued how proprietary a technology could be and undervalued how important the founders were. 

These lessons have multiplied, helping us improve our judgment, quality of coinvestors, and competitiveness to win over highly-contested founders. We’ve also honed our ability to deliver unfair  advantages to every one of our portfolio companies, such as recruiting early teammates and world leading board members, forming useful investor syndicates, gathering competitive intel, teeing up  early design partners and customer introductions, sharpening storytelling and media relations, making connections to corporate development and M&A decision-makers far in advance of an exit, and  advising founders on critical strategic decisions from a quarter-century well of experience. Each gain  for our companies is multiplied across our present and future portfolio in a positive sum equation of excellence, each existing founder a shining beacon for others to join in our pursuit of the possible. 

Multiplying Value: Rising Talent x Unified Culture 

Our team has grown stronger with every fund. We’ve worked hard to attract, recruit, retain and grow new talent with higher standards and greater expectations. We’ve supported several prior partners to pursue their own paths, always grateful for the many contributions they made along the way. We’ve intentionally built a team of young, ambitious and hyper-intelligent investors who have the range, autonomy and cultural alignment to seek out the best founders in the world while acting like  mensches on a mission. We’re now 44 people—10 investors complemented by several extraordinary venture partners as well as a dauntless crew of specialists serving the partnership and portfolio across all segments of value-add (finance, accounting, legal, operations, communications, recruiting, platform, media, content and Riskgaming). 

We operate globally from two kinetically interconnected offices (New York City and Silicon Valley) unified under one culture: Unum Lux. We avoid geographic or sector silos, and each member of Lux is a generalist who learns fast, shares credit, avoids individual attribution and hunts together to the benefit of our partners, who have entrusted us with more than $5 billion of their capital. 

We have the best team Lux has ever had, deeply connected to the best new founders while growing influence and access to add value, with a reputation that compounds not just based on our people inside Lux but multiplicatively through the founders we back. If we serve them well, they serve us well— not just delivering financial returns but helping us win the right to partner with every incremental founder that joins the Lux family. 

Multiplying Value: Decisions x Portfolio Construction 

We are forever improving our judgment and honing our instincts from experience while keeping a beginner’s mind for new sectors that few have invested in before. 

The late economist Carlo Cipolla famously wrote that humanity’s most underestimated force is stupidity—not malice or greed, but the consistent capacity of people to harm themselves and others without gain. His warning wasn’t just about individuals; it applies to institutions, markets and nations. The most dangerous actors aren’t the villains seeking exploitation at others’ expense—they’re the ones blundering confidently into ruin, taking value down with them. 

We’ve inverted Cipolla’s law at Lux by building antifragile decision-making systems, habits and heuristics that make seizing positive opportunities easier while striking against flawed thinking. We  learned from friends like the late Danny Kahneman as well as Annie Duke and Michael Mauboussin on thinking about risk and decision-making to minimize errors of omission (and the permanence of regret, with a single silver bullet per partner per fund, rarely used or exploited) and errors of commission in positioning sizing. We favor process over impulse, structure over spasm. One partner's impatient or impetuous instincts get tempered by another’s pragmatic pacing. Avoiding the unforced error— whether in portfolio construction, partnerships or perception—has compounded quietly like interest  on prudence. We strive to build guardrails against our own overconfidence, and observing the clever cunning and even earnest fools of others, we are reminded that survival and success often belong not  to the boldest, but to the least self-destructive. 

Since Lux’s inception a quarter century ago, our three main sectors of focus have been life sciences, physical sciences and computational sciences, even as the subsectors and boundaries have evolved and deepened. Today, those subsectors span from robotic surgery and cutting-edge biotech to 3D printing/scanning, aerospace, defense, industrial automation and AI and compute infrastructure. 

Likewise, our three styles of investing remain invariant: thesis-driven, people-driven and special situations. We’ve complemented these main avenues for investment with many new initiatives that have become staples of serendipity, sourcing, convening and reputation: 

• SEOs — Seeds, Experiments, and Optionality investments that provide us with key early signals from the frontiers of science and technology 

• Riskgaming — deeply researched and parsimonious scenarios that use low-probability, high magnitude geopolitical and technological developments to convene brilliant, diverse minds around our tables 

• Sci-Tech Diplomacy — bridging science, technology, and policy communities • f(Lux) Fellows — cultivating and connecting the next generation of founders and thinkers 

Together, we have made over 250 active investments, including 39 companies we helped found or for which we “wrote the first line of code.” De novo company creation will always be part of Lux’s portfolio construction, and in fact, we’ve become efficient and effective at founding or co-founding these businesses—often so contrarian or unpopular that no other startups like them existed. Examples include: nuclear waste cleanup; robotic surgery; Nobel Prize–winning microscopes discovering new  drugs; and finding real-life genetic mutants. 

Our strategy of multiplication has worked: we’ve distributed over $1 billion in the last few years and have logged more than 30 profitable IPO and M&A exits. We look forward to sharing more of our  success in the coming years.

‍The State of Markets, Technology and Geopolitics 

Let’s now turn to the shifting terrain of markets and technology, where liquidity has found a new gravitational center. The AI hyperscalers—Microsoft, Meta, Amazon, Alphabet, Tesla and Oracle—are  on pace to spend over $400 billion in capital expenditures by next year in the pursuit of compute  supremacy. 

Alas. Markets pay high prices for cheery consensus while alpha hides at the edges where systems are  breaking, borders are blurring and the future is forcing mark-to-market discipline on the past. Alpha  lives in sectors undergoing rapid, existential change; in assets escaping the gravity of traditional  finance; and in companies that crawl through the valley of venture death to emerge as indispensable infrastructure. 

A first fracture appears to be running through the service economy: as the S&P 500 climbs to record highs, AI is mowing down the middlemen. Accenture, Capgemini and Gartner––consulting empires that  monetized human repetition—are facing obsolescence, down over 30%. AI may prove to be a deflationary scythe through white-collar work. As volatility rises, capital seeks physical refuge: gold is up 45% YTD, trading near $4,000 per ounce, outperforming major indices and proving that in an age of  digital abundance, scarcity still shines (literally). 

Even as Lux is launching new defense companies on the European continent, London’s capital formation pace continues to decline, with IPO proceeds the lowest in 30 years, mirroring the city’s slide from empire to echo. Abu Dhabi and Dubai have become testbeds for a new model of state capitalism:  nimble, cash-rich and future-obsessed. They are absorbing the financial and human capital London is  losing—an inversion as symbolic as it is structural. Like heat, wealth and talent move along gradients. Europe’s old hubs are cooling, their gravitational pull weakening. The new centers of attraction combine capital surplus with strategic ambition—where mass liquidity meets manic velocity. 

The venture market too has—as we’ve long said—become a tale of megas and minnows. In 2025 YTD, ~43% of all venture capital raised by LPs flowed to just 10 managers; ~41% of VC dollars went into 10  anointed companies, eight of them AI. 

The irony of this era is that financial assets have become more imaginary just as real assets have become more real. Markets no longer mirror the world—they project belief. Gold gleams not because  it changed, but because faith did. Bitcoin, born of distrust, now functions as the monetary mirror image  of sovereign debt: finite, transparent, immune to decree. Meanwhile, the things that actually keep  civilization running—energy, water, minerals and manufacturing—are getting harder to produce. The AI boom, often cast as ethereal, is profoundly physical: data centers draw gigawatts, with their copper wiring measured in thousands of tons and their cooling water in teraliters. The future, for all its talk of intelligence, still runs on iron. When everything becomes financialized, scarcity retreats from paper  to the physical world. You can print liquidity, not lithium (yet––stay tuned for more on that soon). The so-called intangible economy turns out to be anchored in very tangible foundations. In an age of infinite code and finite copper, alpha in the future may be returning to the things we can touch,  measure and build upon.

Our task is to back the builders of that future. Globally, the world’s defining rivalry is not fought through threats of tanks and troops, but through transistors, transducers and term sheets. It’s a contest not only of ideology but of infrastructure—where America for now rules the realm of bits while  China consolidates the dominion of atoms. America, unrivaled in its ability to summon novelty and scale, has birthed nearly 7,000 new AI startups since 2013, drawing nearly $500 billion in private  investment. Today, America counts nearly 700 unicorns worth over $2.5 trillion combined, a reflection of both capital abundance, creative compounding and of course, the inevitable failure and  collapse of many of them. That software supremacy rests precariously on hardware dependency. And at the base of it all lies the quiet, coercive leverage of the periodic table: China’s near-monopoly over refined rare earths—metals the U.S. scarcely processes—gives it the ultimate chokepoint in this new  geopolitical alchemy. Beijing deploys 300,000 industrial robots a year, mechanizing the global factory  floor, while Washington debates industrial policy. There are signs of optimism and change in the somewhat refreshing irreverence of the Pentagon and Treasury to move fast and shake things up to be  more competitive. 

We remain carefully attuned to productive shifts in Washington’s attitudes, but we never forget that  the alpha on the frontiers will invariably come from the rebellious and brilliant scientists and engineers who forge their own innovative paths. For every Bessent remodulating policy to improve  America’s future, a Palmer is needed to blaze the trail. America’s 250th is a reminder that we were founded as rebels against orthodoxy, with a readiness to fight for the values of freedom that made this  country the richest and most dynamic in the world. 

Kinetically Convening Genius 

A key thing we’ve amplified in recent years: kinetic convening. We’ve averaged more than an event  per week across over 12 cities in the U.S. and Europe including dinners, receptions, summits, retreats  and Riskgaming gatherings. All told, nearly 3,000 attendees have covered themes like AI, biotech,  pharma, physical sciences, AI, compute infrastructure, geopolitics and more. These gatherings are not  just social touchpoints but strategic catalysts—amplifying Lux’s reputation, deepening deal flow and strengthening our network across science, technology and capital. 

We recently hosted our Annual Meeting, preceded by our third Lux AI Summit. The quality of attendees,  talks and insights raised the bar yet again. We revealed new investments, experienced awe-inspiring demos, and showcased Lux team and Lux founder non-consensus variant perceptions. Next, we share  some actionable insights from both days. 

Kinetic Convening: Healthcare & Biodefense 

In healthcare, we noted biotech today sits at a rare intersection: valuations at decade lows, technology quality at decade highs. That disconnect between capital cycle and capability curve is an arbitrage opportunity. On the one hand, AI lets startups do more with less—using large models to  underwrite biology, focus on single targets and reach early clinical milestones with less capital. Yet the half-life of a good idea has collapsed since Chinese labs can read a paper in Nature and start clinical  trials before the original team finishes hiring. The next U.S. edge won’t be cost or speed but  complexity—ventures so technically hard and infrastructure-intensive that only frontier builders dare attempt them: pig-to-human organ xenotransplantation (eGenesis), single-cell perturbation genomics (Neptune Bio) and AI-driven biodefense (Valthos). 

On the latter, Valthos is our newest Lux Labs de novo company creation and was co-founded with  Kathleen McMahon, who was previously head of life sciences at Palantir. Bioweapons were once limited to state actors (Russia, China, and North Korea), but now AI models, open-source genetic design  and biomanufacturing are lowering the barrier of entry from hundreds of specialists in secure national  labs to potentially a few independent individuals with modest skills and hardware—and malice. Valthos is responding to this new threat through next-generation detection, surveillance and management. 

Societies already gamble by guessing naturally evolving flu strains and producing vaccines, but synthetic pathogens present a much graver threat. Biology has crossed the threshold from  observation to orchestration. What once took battalions of bench scientists now takes a few graduate students armed with GPUs and curiosity. The code of life—once read, now written—has entered a compiler phase, where the boundary between wet lab and cloud is dissolving. A decade ago, protein  folding was the frontier; now the question is how fast we can design, synthesize and test entirely new  and more complex interacting biological machines. The codebase of nature is being forked in real time  (including concerns and risks of “mirror biology” that could yield molecules undetected by our immune  system). The old deterrence model—slow defense versus fast offense—fails when the pace of invention quickens. The most credible antidote is speed itself: a civilization whose countermeasures iterate as fast as threats mutate. In biology, the same algorithms that could conjure danger will be conscripted to generate its remedy. We will need close collaboration between government and life science companies to create a rapidly responsive biodefense ecosystem. 

Kinetic Convening: AI & The Computation Stack 

Our team shared insights and contrarian takes on AI, noting that moats look different this cycle. At the  capex layer, scale compounds; at the application edge, brand and habit may win but switching costs remain low. The stack is cohering into vertical utilities by craft—code (Cognition), legal (Crosby) and support (MavenAGI)—each automating a specialized labor market rather than merely a software niche. Ironically in a world of “artificial” intelligence, the only scarce resource that still decides outcomes: human intelligence. Talent raids, acqui-hires, “Poachapalooza 2025,” wunderkinds deemed barely old enough to drink or rent a car—proof that while capital builds capacity people still build the  winners. 

David Ha of Sakana AI showed how Japan may quietly define what comes after the age of large  language models (LLMs). Where others build fortresses of data and battalions of GPUs, Sakana AI is  cultivating an archipelago of smaller, evolving intelligences—systems that learn by recombination, like  coral reefs of code. The next frontier of AI may not be larger models but living ones: adaptive, self-correcting and communal, proving that the future of intelligence may belong not to empires of scale, but to ecosystems of emergence. 

Alejandro Matamala of RunwayML talked about generative AI tools which began as toys but have  matured into instruments for serious imagination––no longer rivals to creativity but partners in its production. Runway made the wise decision to embed itself in the workflow of creatives. Every tedious ‍edit the machine absorbs is air for the artist’s intuition. What matters next is control. The true artist  doesn’t seek random outputs but refined ones, capable of being shaped and rehearsed like any other craft. 

Early Math Olympiad genius Scott Wu, founder of Cognition, shared what started nearly as a hacker house experiment and evolved into an “applied AI lab” building the first true autonomous engineer.  Devin, their AI software engineer, doesn’t just autocomplete lines—it reasons, debugs and improves itself, collapsing the boundary between user and system. The next leap in software won’t come from faster typing but from code that wakes up tomorrow just a bit wiser than yesterday. 

As communication AI (voice, video, audio, text, chat, code) gives way to physical AI (robotics, biology, defense, manufacturing), we had Physical Intelligence founders Karol Hausman and Lachy Groom discuss the state of play in robotics. For decades, intelligence lived weightless—text in a window, answers in the cloud. Now it’s acquiring mass. Watching a robot fold a t-shirt, clumsily but autonomously, is to witness embodied intelligence learn friction, gravity and grace. The miracle isn’t that machines are imitating humans but that they are beginning to share our constraints. Every  gesture becomes data, every mistake a moment of learning, every repetition a rehearsal. When robots  learn from their own motion, labor itself becomes self-improving—a kind of infinite apprenticeship where the factory learns as fast as it produces. The early demos look humble—linen creased, mugs  toppled—but the trajectory is unmistakable: code becoming corporeal. What language models did for reasoning, these physical models will do for dexterity and navigation. We’re moving from software that predicts the world to systems that participate in it. 

Zongheng Yang of SkyPilot noted how AI teams are running out of GPUs, but new clouds are springing  up—each with different prices, chips and quirks. SkyPilot makes them all feel like one seamless supercomputer, letting companies move their training or inference jobs wherever capacity is cheapest  or fastest. TogetherAI founders Vipul Ved Prakash and Tri Dao talked about how a quieter truth is  emerging from our scramble to build data centers: the bottleneck is no longer computation but  coordination. Models not only have to think, they must cohere as well, and thus, the next frontier of  optimization is orchestration: routing the right model to the right chip with the right context at the right time. The future won’t belong to those who merely own the most GPUs, but to those who can choreograph them—turning stochastic chaos into code-conducted symphony. 

We riffed on the foundation model companies and the rise of the memory players with Dan Grossman, who helps run Amazon worldwide corporate development responsible for their mega investments and  M&A. A decade ago at Zoox (acquired by Amazon), we watched engineers playing what looked like “Grand Theft Auto” with lidar feeds, training cars in silico on unreleased Nvidia chips. Back then, Nvidia  was worth $15 billion and Intel $150 billion—a pair trade of the century as Nvidia sits at a $4.5 trillion  market cap today. As we’ve written about previously, we sense the same inversion coming for memory with SK Hynix, Samsung and Micron set to become the soul of the new machine. If 50 percent of AI inference moves on-device using flash memory and the demand stays high for high-bandwidth  memory attached to GPUs, it may upend consensus assumptions of infinite demand for data centers and power plant build outs. That future is propitious for our edge inference chip investments from  Mythic to Majestic.

Meanwhile, all large frontier lab companies are acting individually rational yet collectively irrational— each chasing 100% market share while together spending 400% or more in capex. If Google offers Gemini for free (as they did with Google Docs nearly two decades ago) or an open-source model offers  compelling good-enough performance, competitors may never grow revenues to compensate for their build-outs. The AI capex super-cycle will be gloriously useful for society and brutally destructive for many investors funding it—history rhymes with fiber, colo and early cloud. Most of the gross margin  will accrue to the toll-keepers—Nvidia and the hyperscalers. The outcome may be familiar: misallocation on a monumental scale, over-investment followed by surplus. Users surely win, and  investors may scatter as infrastructure nonetheless compounds. 

Kinetic Convening: Defense on Offense 

Our five-year psychological bias: everyone wants to be invested today where they ought to have been five years ago is clear in AI and becoming more clear in aerospace and defense. Defense exemplifies the collision of Lux’s themes—hard tech, autonomy and moral calculus. 

The postwar West traded discipline for diffusion. The defense and aerospace supply chains that once won wars splintered into a thousand subcontractors—America’s “arsenal of democracy” devolved into a democratized archipelago of mom-and-pop machine shops. Fertilizer plants replaced munitions factories; a by-product of Chinese cotton shirts now feeds Europe’s explosives industry. What was once  resilient interdependence has become strategic vulnerability. Reshoring isn’t a slogan—it’s survival.  The U.S. builds fewer than six commercial ships a year while China built a majority of new deadweight tonnage last year. Europe, jolted awake by Ukraine, is rearming on a scale unseen in half a century with  new primes rising from the dust, new multi-layered air defenses designed for drone-saturated skies and new supply chains built to last beyond the current crisis. 

Warfare itself is mutating. Tanks now die to cheap quadcopters; anti-tank missiles have given way to anti-drone jammers. Unmanned systems fight below a thousand feet while AI fuses the battlespace  above it, compressing decision time from minutes to milliseconds. From Mosul to Mariupol, the lesson repeats: you learn fast or you learn late. The new race is not for territory or ideology but for adaptation itself—the ability to evolve at the speed of threat. 

Matt Grimm, co-founder and COO of Anduril (proudly wearing a Winnie the Pooh shirt playfully taunting Xi who has not playfully but personally sanctioned Grimm––a clear sign we’re doing something right), talked about aggressively hiring while methodically scaling operations and global manufacturing. Critics said a startup could never build at prime scale. From the start, the ambition was unsubtle: build a company that belongs in the same sentence as Lockheed, Northrop, Raytheon and Boeing—and then earn it. The early programs were controversial; the vision wasn’t. Air, land, sea, space, cyber—every  domain, software-led. Anduril built at risk: Arsenal-1 in Columbus, Ohio—roughly 5 million square feet  at full build, thousands of jobs, modular lines for jets, counter-UAS, undersea systems—was greenlit before the contracts caught up. The company then beat legacy primes into the Air Force’s  Collaborative Combat Aircraft down-select and recently flew America’s first-ever autonomous fighter jet. Anduril stood up sovereign production with allies (U.K., Australia and beyond) and shipped  capability faster and cheaper: Ghost Shark in Australia logged more hours at sea in around 3.5 years than Boeing’s decade-long Orca, at roughly one-seventh the cost. That is the thesis made concrete: software tempo, vertical integration where it counts. 

The moral dimension of defense is key too: the ethic is deterrence through precision. Higher technical resolution yields higher moral resolution: a big, credible arsenal one ideally never has to use. Along the way, Anduril has made defense cool again—“Don’t Work at Anduril” billboards, sanctions-as-a badge, and onboarding classes big enough to need their own auditorium—pulling top-grade talent from big tech and old primes alike. 

The Next Century of Thanks 

Over the past 25 years, to paraphrase Blade Runner: we’ve seen things you people wouldn’t believe— nuclear waste cleanup technology treating hundreds of millions of gallons of radioactive water; robots  performing complex surgeries; transplant surgeons transferring kidneys from genetically engineered pigs to give the gift of life to humans; vessels launched into space returning at hypersonic speeds;  satellites that will reflect sunlight to illuminate dark swaths of Earth at will; revelation of the frenzied  chaos inside of cells tracking microscopic particles to reveal how drugs are metabolized by the  machinery of life; silicon and code conjoined to conjure images and videos from the imaginative  recesses of one mind to the irises and retinas of another’s eyes; cars that drive themselves and unmanned ships and fighter jets; code that instantaneously turns prompts into proteins. 

None of this could have been done without our partners, founders, teammates, peers, friends, contacts and everyone else who dared to join us on these ambitious quests of the possible. We are forever grateful to dedicate our lives to bringing the future to the present, to transition science fiction to science fact, and to deliver the financial rewards of these inventions to many great missions in the service of humanity. Thank you for your support. Fiat Lux.