With another whip around the sun and in preview of our new letter, we discuss marked progress in our portfolios; turn to markets and the macro in the context of Moravec’s Paradox; magnify our investment theme of ‘maintenance’; recount our Riskgaming initiative and how it finds possible adjacencies; and lastly, advocate for investing in the adjacent possible.

The Lux partnership has been very active, thoughtfully and intentionally adding new rising stars with fresh ideas and networks to our investment team, exploring new interdisciplinary cutting-edge science and tech domains, creating new companies from scratch, spinning-out high-performance teams from large tech companies, and spotlighting our leading investments in global media and policy circles. We have embedded our portfolio leaders into spheres of intel and influence to give them competitive advantage––from the executive suites of major customers and partners to the power centers of Washington DC.

Stewarding American prosperity for generations to come while thwarting America’s enemies, the Lux partnership has been requested to offer expert testimony on the future of artificial intelligence as well as U.S. technology competition with China, has been asked to host and advise senior military commanders on new technologies that can help their missions, and has been solicited for strategic counsel by senior White House national security leaders on the complex interconnections between AI, energy and geopolitics.

Many Lux-family companies have heeded our advice and capitalized well, husbanded their cash, completed opportunistic acquisitions when financially advantageous, exited when prudent and yielded great positive momentum amidst our consistent concerns for tightening market conditions.

To wit: Hugging Face, the leading open-source AI platform hosting over 500,000 AI models, 250,000 datasets, and 1 million repositories, last year raised a $235 million financing at a $4.5 billion valuation from Google, Amazon, Nvidia, Intel, and Salesforce solidifying its "position as a neutral platform, or the Switzerland for AI." While there are tens of thousands of large language models for text, there are only a few for biology — stay tuned for news on Evolutionary Scale (another investment using our strategy of spinning talented teams out of large tech companies and turning them into turbo-charged new ventures). There are even fewer AI models for robotics, which is why we funded a gaggle of geniuses from Google and Stanford in a new venture named Physical Intelligence at the intersection of unstructured robots and AI (think: not warehouses). Philosophically, we’re skeptical about humanoid robots, because we believe brilliant bursts of engineering beat billions of years of evolution. If you were designing a car today, it wouldn’t run anthropomorphically on Fred Flintstone’s furious flying feet and cylindrical stone wheels careening through crowds with those yabba-dabba-doo eyes, but rather on high-torque motors, engines, powertrains, computers and batteries, coupled with a fusion of sensors operating on the best telemetry of the spatial environment available.

Speaking of environments and their 3D spatial digitization, Matterport, which we funded from its founding, is being acquired by CoStar Group for $2.1 billion at a significant premium to its public valuation. Databricks, which acquired our generative AI tools and models company MosaicML for $1.3 billion, revealed that it had generated $1.6 billion in revenue for the year ending January 31, 2024, representing growth of more than 50% over the prior year. Applied Intuition — uniquely cash-flow positive for several years — raised $250 million and signed a key partnership with Porsche, boosting its valuation from $3.6 billion to over $6 billion with a cash war-chest of over $1 billion to invest and allocate. Lux is the largest shareholder in Together AI, which is the fastest cloud platform for building and running generative AI workloads and raised $106 million from new investors. Closing the gap between sci-fi and sci-fact, Varda raised a $90 million Series B to manufacture next-generation pharmaceutical drugs in zero-gravity space, and also landed its first space factory in Utah, marking the first time a commercial company has landed a spacecraft on U.S. soil, ever. Lux is very bullish on Japan, and we led a $30 million founding round in Sakana AI, backing an AI thesis we believe will win: cutting the costs of training and inference through "co-evolving" smaller AI models together. Finally and notably, our next-generation aerospace and defense contractor Anduril announced it was down-selected alongside General Atomics to develop the Air Force's multi-billion dollar Collaborative Combat Aircraft or "loyal wingman" program, emerging ahead of defense primes Boeing, Lockheed Martin and Northrop Grumman. In addition, Anduril delivered Ghost Shark to our defense allies in Australia ahead of time and under budget (a rarity among incumbent defense contractors).

The Moravec Paradox, Markets, and Minnows & Megas

Observing the state-of-the-art robots of the 1980s, artificial intelligence researcher Hans Moravec formed a theory that came to be known as the Moravec Paradox: tasks that humans conduct effortlessly and unconsciously present an enigmatic equation that AI is unable to resolve, while tasks that humans find dastardly difficult (reading an extensive legal contract or playing Go) would ultimately be easy for AI to learn. The launch of ChatGPT turned his theory into prophecy. AI can now read thousands of pages of dense legalese in seconds, but no robot can walk through a garden and pluck a dandelion, a skill available to an inquisitive two-year-old. His thesis was predicated on natural evolution: the human mind is extraordinarily high fitness and high efficiency, and the perception skills that were critical for survival (identifying threats, understanding motivations) would be deeply embedded in the structures of the brain in contrast to skills like logical reasoning, which evolved eons later.

Moravec’s Paradox was an observation of a field during a crucial era of research, but the insight — that the stuff most people think is hard is easy and the stuff we think ought to be easy is hard — applies to far more than robotics in the 1980s.

Let’s turn to markets where Moravec’s Paradox abounds. Curtailing inflation is considered hard, and a so-called soft landing even harder, but aggressive rate increases by the Fed coupled with the transient supply shocks precipitated by Covid-19 arrested consumer inflation from an annualized peak of 9.1% in June 2022 to 3.5% today. Yet, the easy stuff proved hard: the U.S. government deficit hit $1.7 trillion in 2023 while the U.S. economy maintained its resilient growth. Now, plans for difficult rate rises followed by a tactical tapering have been replaced with a far more perilous and unpredictable future of the global economy. One factor we’ve focused on that’s missing from the front pages with implications for both inflation and the U.S. election is labor, unions and wage wins. Median wages are up an annualized 4.7% compared to 2.8% over the last decade, thanks in no small part to an aggressive and successful series of strikes last year, most notably with the UAW and the Big 3 automakers (their final contract will increase starting wages 65%, while all wages will increase an average of 25% over four years). We expect unions will play an outsized and perhaps even kingmaker role in this year’s election given their newly energized memberships and density in key Midwest battlefield states. Our view is that the true cost of capital will continue to rise.

We hold that much of the purported $300 billion of VC dry powder is, in fact, wet, pre-allocated by funds to prop up old portfolio companies at prior prices deemed too high so as to avoid future write-downs deemed too low. In the last quarter, down-rounds in U.S. VC made up nearly 25% of financings (its highest level in years), while bridge rounds (which may prove to be dead-end piers to nowhere) reached a high of 40% of Series A and B rounds.

We continue to expect a cyclical contraction in venture, with as many as half of venture firms fading or failing and exiting the industry. Contraction will be driven by fractured partnerships and feckless fund strategies too weak to fix as well as LPs overcommitted to the illiquid venture asset class. From about 1,000 VC firms in 2008, the industry reached a peak of over 3,400 VC firms today, and we expect that number to decline to 1,700 in the next few years. The industry will likely be a minimizing mixture of many “minnows” and a few “megas”. Minnows are the subscale funds below $100 million or niche-specialist firms unable to raise a second fund or become more institutional as they, their LPs and their founders view them as undercapitalized to support a tougher financing environment where endless easy-money up-rounds are no longer the norm. Already, first-time funds are falling out of favor, with the fewest new funds raised since 2014. At the other end, many “megas” (funds greater than $5 billion) must contract their prior outsized fund targets as they face LP disinterest, disapproval or indigestion from overallocation. The very best founders once targeted these megas while seeking princely unicorn-status valuations, and now those same founders see the very opposite: a less-selective, last-resort, low-signal stamp of approval. Moravec’s Paradox strikes again: the hardest part of building a mega was considered raising capital, with deployment considered relatively easy in the frenzied bubble era of 2018-2022 where returns were a cinch. Now, these megas have learned that the capital fundraising was in fact easy, and profitable deployment mathematically improbable if not impossible.

We expect total dollars from LPs to VC to fall, and for those scarcer dollars to become concentrated in fewer managers. Indeed, just two established firms with large new fundraises accounted for nearly half of LP capital committed this year so far. The most advantaged managers will likely prove to have funds in the $1-2.5 billion range, with the ability to back needle-moving, early-stage investments as well as larger, high-conviction growth checks for cash-on-cash multiples that can return the entirety of a fund.

Magnifying Maintenance as a Lux Investment Theme

In recent LP letters, we’ve highlighted the five-year psychological bias where everyone wants to already be invested today where they ought to have been investing five years ago. Our vanguard investments half a decade ago in AI platforms and infrastructure — as well as aerospace and defense — are now increasingly the mainstream thesis for the industry. What might others seek to fund five years hence? Maintenance.

Let’s bring Moravec’s Paradox to the built world. Most consider building something new to be hard, while maintaining that physical capital to be easy. After all, combining design, planning, construction and finishing would seem to be a far more challenging orchestration of skills than cleaning the facades, oiling the gears and running through checklists of safety protocols. Consider again that CapEx — the lifeblood of industry — flows from two fonts: growth and maintenance. Yet, while the recent fever-pitched pursuit of growth gorges on attention and investment, the vital virtue of maintenance has been minimized and marginalized. As rates rise and the cost of capital climbs though, the myopic fixation on the new has left the old at risk of rust and ruin. It is in fact the simple act of maintenance that we find so hard.

In the realm of defense, this neglect isn’t merely imprudent — it’s imperiling. At Lux, we pride ourselves on funding capabilities across air, land, sea, space, and cyber. There is no doubt that Lux companies like Saildrone with a fleet of attritable unmanned drones and Anduril with unmanned subsea Ghost Sharks will play a key role in deterrence and defense of America and our allies. However, we note that the most critical defense systems aren’t just new platforms, but also new technologies that help maintain the materiel we already have. Consider our Navy, the bulwark of American might. In the halcyon days of the Cold War, 600 ships and 600,000 sailors stood sentinel, but as the Berlin Wall fell and global conflict cooled, the military-industrial complex contracted during the so-called Peace Dividend. The defense base, once a robust 6.5% of GDP, withered to a mere 2%. The Pacific fleet, with its 688-class subs, Nimitz carriers, Arleigh Burke destroyers, F/A-18s, and UH-60s, all share a common ancestor: the 1980s. More than 35 years into their lifecycles, America’s defense increasingly relies on our ability to maintain these systems cheaply and even autonomously. With computer vision and better sensors, maintenance can be proactive and preventative, and not reactive and ruinous. Labor-saving innovations can help maintenance teams prioritize the right repairs at the right time, cutting costs while crucially increasing readiness and availability.

Redoubling “Riskgaming” + Possible Adjacencies

We say at Lux that failure comes from a failure to imagine failure (a reasonable rewrite of Moravec’s Paradox)––and this was part of our inspiration for our Riskgaming initiative. We convene scientists, engineers, startup founders, flag officers, politicians, policy entrepreneurs, journalists and more to assume roles in complex scenarios with enthralling game mechanics all conjured by our brilliant Danny Crichton. The purpose is to convene diverse minds with fresh ideas to tackle wicked problems, ranging from Naval shipyards vulnerable to climate catastrophe and AI election integrity in an era of deepfakes to a President elected on the premise and promise to replace government bureaucracy with AI. Player experts must balance short-term transactional mindsets with the patient pursuit of their long-term strategic advantages, all while considering the incentives of labor unions, self-interested entrepreneurs as well as patriotic and corrupt government officials alongside foreign espionage, IP theft, commercial interference, open-source and proprietary technology development and more.

A key benefit for Lux from riskgaming is how it opens our minds to possible adjacencies by engaging us in scenarios under-considered by others. Riskgaming also helps us make new connections between our portfolio companies and influential minds, creating unique venues for cross-pollination and relationship-building. Already, several hundred players have joined, and we’ve even had startup CEOs bring their own colleagues and peers together to play independently. In a flagship event earlier this year in partnership with former New York City mayor Mike Bloomberg, we hosted five three-star generals, two congressmen, a handful of DC policy executives, and a group of tech CEOs to consider the future of AI and national security.

In our first Riskgaming scenario that we’ve now opened to the public, Hurricane Helen, a once-in-a-generation Category 5 superstorm, has left the Hampton Roads region in Virginia — the heart of America's naval might — in ruins. With ships shattered and shipyards shuttered, the Pentagon is faced with a portentous decision: double down on the beleaguered Gerald R. Ford carrier program, a potential boon to a region in ruin but a budgetary behemoth, or cut its losses and cut its contracts, circumscribing both regional recovery and the Navy's future force projection. All the while, the winds of labor discontent blow, as a pro-union President spies political opportunity in a critical swing state. At this crossroads of calamity and calculation, the future of the region — and perhaps the nation — hangs in the balance. The easy stuff may turn out to be hard, but at least with riskgaming, we will be prepared for whatever future is in store.

Non-Interference and the Quest for the Adjacent Possible

Let’s now turn to Moravec’s Paradox in the context of the innovation economy. What is hard — innovation — is in truth astonishingly easy: give talented people open freedom and resources and ingenuity abounds. Yet, what is or at least should be easy — just don’t interfere with them — is, in reality, astonishingly difficult to do. Over the past year, we’ve witnessed governments around the world, including the United States, the United Kingdom and the European Union, begin to regulate and place an imagined ceiling on innovation in artificial intelligence. Too much is happening too soon, politicians argue across partisan lines. Yet, that last great fertile open field of innovation — the internet — has in fact transformed billions of lives for the better while ushering in an extraordinary treasure of global wealth. What is easy is to learn from this past experience; what is hard is to sit still and do nothing to stop a promising future from arriving. As French mathematician Blaise Pascal once wrote, “All of humanity's problems stem from man's inability to sit quietly in a room alone.”

These limits don’t just portend more paperwork for early-stage startups that can be handled effortlessly by well-budgeted incumbents, but rather that such limits can circumscribe the combinatorial ferment of the next generation of human ingenuity. From the possible adjacencies we witnessed with riskgaming, let’s turn now to the adjacent possible. First coined by Santa Fe Institute complexity scientist Stuart Kauffman, the adjacent possible describes how all new things come from combinations of old things. The more new things we have in the present, the more new things we get in the future — an exponential growth function. Take the cornucopia of combinatorial chemistry: just five elements — carbon, nitrogen, oxygen, phosphorus and hydrogen — encodes the entire edifice of earthly existence, a testament to the triumph of the recipe over the raw. Or take compound chemistry: we get bronze from mixing copper and tin; steel from mixing carbon with iron; and superconductors from mixing barium, copper, oxygen and yttrium. If you took the 94 naturally-occurring elements on the periodic table and combined just two atoms, you’d have 8,742 possibilities. Picking four elements yields over 73 million possible combinations, and five elements yields over 6 billion. And even this score underestimates the true diversity of possibilities by omitting the different proportions possible between elements.

Turn now to technology. 2.6 million years ago, Australopithecus had maybe 10 simple stone tools. 2.5 million years later (or about 40,000 years ago), Cro-Magnons had a few hundred. 3,000 years ago during the Bronze Age, we had several thousand tools and today, there are billions of tools in existence. The adjacent possible explains why that exponential expansion in human capabilities took place. Gutenberg’s printing press was a remix of two inventions: movable type and the wine press. The Wright Brothers’ plane was a remix of four inventions: the airfoil, a lightweight engine, a propeller, and a bicycle wheel. Each new invention is nothing more than the matrimony of a multitude of once-isolated inventions that came before it. Combination is the catalyst of creation, and in that birth, lays the seed for its sustenance and several successors.

The adjacent possible is a kind of shadow future, lurking on the edges of the present state of things, a map of all the ways in which the present can reinvent itself, a realm of potential that is just a recombination away. At any given moment, the system can only change in certain ways; that is, the system can only move into the "adjacent" possible states from its current state. To explore it is to dance at the intersection of the actual and the achievable, the real and the realizable. The adjacent possible captures both the limits and the creative potential for change and innovation within biological, chemical, or social systems. The only force that can stop this forward progress is the logic of limits and the regression of regulation, particularly overregulation as a function of fear. Fear isn’t the right emotion to feel though, but rather awe. It’s inspiring and unsettling to imagine the combinatorial pairings that AI will usher in, perhaps accelerating a number of technological, social and economic revolutions. Yet, fears of a singularity have a singular weakness: one of the quirks and possible limitations of AI is that extrapolations depend on weights and biases derived from the past, and the real world is not a linear machine. There isn’t a mathematical equation to deduce what the biosphere, technosphere or economy will evolve into. At each step, some new things are possible that were not before, and new things that will be possible are adjacent. We recombine what we have and keep what works, which becomes combinatorial fodder for the future in ways we — and AI models — can’t anticipate.

America and Moravec’s Paradox

What is easy for America is what is hard for most other countries: two placid oceans that offer the comfort of incredible distance, an unfathomable endowment of physical resources from petroleum and hydrogen to abundant farmland and pastures, a rich and varied economy based on freedom and the rule of law, and most importantly, a permissive and open culture that welcomes the globe’s best talent and most ambitious immigrants alongside our country’s industrious and intelligent sons and daughters. What should be easy for America — don’t break it — is precisely what we’ve witnessed is so hard for our nation’s leaders and its citizens to restrain themselves from doing.

Our adversaries, cognizant of their relative disadvantages, seek to fray America’s fabric from within. From TikTok to universities, they sow dissent, twist values, and seed and incubate self-loathing. And no society or civilization has ever succeeded on the strength of self-loathing or a hatred of the other. The recent eruptions of antisemitism and anti-American sentiment on our college campuses are not mere youthful zeal — they are the bitter fruit of a long and insidious campaign by America’s adversaries requiring constant vigilance.

Against this persistent threat, we need not just wokeness, but an awakening — a renewal of the righteousness and moral clarity that has been the bedrock of American greatness for more than two centuries. America is defined by not what it is against, but what it is for. We don’t find differences to define divisions, but seek contrasts to construct combinatorial creations. We don’t look inward at the already invented, but outward to the adjacent possible. We must remember the roots of Silicon Valley itself, born from the union of technology and national defense. As investors, we will continue to find and fund those entrepreneurs inventing a brighter future for all of us. In the face of mounting threats at what we have previously dubbed the entropic apex, we must maintain not just our machines, but our resolve. Fiat Lux.