Superlatives
Imagine blacking out in 2019, waking up in 2025, and realizing that history is a Möbius strip—Trump leads the White House in the midst of a self-induced trade war, SoftBank is in the headlines investing in real estate (data centers this time), and the dragons of the Game of Thrones universe are once again streaming. Yet, the stakes—and the superlatives—have never been greater. DeepSeek’s launch of R1 triggered the largest single-day market loss in history, while OpenAI announced the largest private compute initiative ever with its Stargate Project. The wildfires that swept Los Angeles are the most expensive ever recorded, while a polar vortex led to record snowfall in Florida. There are more active conflicts around the world than at any time since World War II.

This surreal sense of returning to a world both changed and unchanged underscores the fundamental reality of markets and geopolitics—while the players and headlines shift, the underlying forces of uncertainty, disruption and cyclical repetition remain constant, challenging investors to distinguish between transient noise and durable trends.

In a world of superlatives, our quality of reasoning and the fortitude of our actions must match the extreme risks and opportunities the world presents. Geopolitical and technological fault lines are shifting faster than conventional wisdom can keep up. This reinforces the need for agility in investment and low-probability high-magnitude scenario analysis—an area where Lux’s Riskgaming gives us an edge. As markets move from an era of zero-bound rates and unlimited liquidity to one of more normalized monetary conditions, the premium on discernment rises. Quantitative observations and insights can reveal the what; our task is to divine the why and anticipate the what next.

Markets in 2025: A Year of Contradictions and Conviction
As we enter 2025, the rapid rush of superlatives of recent quarters is moderating as markets are balanced between consensus and contradiction, stability and risk, promise and peril. Apple nearly reached a $4 trillion market cap in December as Bitcoin hit its highest price ever, yet the highest inflation in a generation abated. The S&P 500 rose the fastest for a new administration since Reagan in 1985, even as credit delinquencies reached their highest since the aftermath of the Great Recession. We recently hosted Jason Zweig and the family of Ben Graham at Lux for the release of the newest edition of The Intelligent Investor, in which Graham notes that the market is a pendulum that forever swings between unsustainable optimism and unjustified pessimism.

While the post-pandemic surge and scare has tempered and soft-landing optimists toast 2% GDP growth, the fissures of reality are widening—credit spreads remain tight, dry powder is anything but and the real economy is already showing signs of structural strain. Rising operating costs and consumer belt-tightening have led to a subtle but noticeable uptick in restaurant and store closings. This undercurrent of weakness, combined with ever-shifting political rhetoric, sets the stage for a turbulent year—one where Trump is likely to blame Biden and Powell for any economic missteps, potentially paving the way for Kevin Warsh to replace Powell at the Fed. Even as the Fed Funds Rate is forecast to dip to 3.7%, debate lingers about how long rates will remain elevated, a reminder that markets and monetary policy are rarely harmonious for long.

Inflation is not just a fleeting concern but a structural reality. The trillions in global liquidity that were injected to stave off economic collapse after the pandemic now also set the stage for a prolonged inflationary era. Trump’s push for re-industrialization carries an inflationary sting, as does global aging, greater geopolitical constraints on trade and the intensification of climate disruption on core human staples.

We continue to hold the view that the natural cost of capital is higher and rising, dry powder in private equity and venture capital is “wet” (due to under-reserves and overallocation) and 30-50% of venture firms will exit the market from failed succession planning, partner in-fighting and other Shakespearean dramas. New entrants are also struggling: 2024 was the worst fundraising year for debut funds in a decade.

Broadly elevated valuations cast a long shadow over equities and credit markets alike. The perpetual quest to dampen volatility often achieves the opposite—allocators fleeing public mark-to-market transparency have, in private credit and leveraged private equity, merely displaced risk in time rather than mitigate it. Meanwhile, record retail participation (a classic contra-indicator), the structural distortions of passive indexation (heavily concentrated in U.S. tech), along with soaring valuations herald not just a “Golden Age of America” but the familiar arithmetic of high multiples and low future returns. The United States may remain the best market for funding innovation, but that doesn’t preclude it from being the most expensive, pricing in its own inevitability.

Meanwhile, the “Magnificent Seven” tech giants, which drove over 50% of the S&P 500’s returns in 2024, embody both the opportunity and the concentration risk embedded in today’s markets. That narrow leadership may falter in 2025 as murmurs of sector rotation crescendo into a clarion call. Widespread indexation and passive investors risk being swept along with the tide, while active managers with conviction have the chance to swim against it and thrive. The upshot is that markets have also recalibrated for a second Trump term expecting more DOGE less DOJ, the possibility of exit euphoria and a revival of a 'capitalism without a chaperone' ethos. Trump loves his superlatives, and the markets are waiting for him to act.

Superlative Science from Speculative Fiction
We love being involved in science’s superlative secrets. One such secret is eGenesis. There are 8 billion people in the world, 10 people live in outer space, and only two people are today living with a kidney transplant from a genetically-engineered pig. For centuries, organ transplantation was constrained by the tyranny of biological compatibility. In 2025, that constraint is fading—first for a few humans and then for hopefully many more—proof that the future is indeed already here, and it’s just not evenly distributed.

Superlatives like this abound across Lux’s portfolio. Varda is one of only three companies in the world to have ever flown a vehicle from space to Earth. Anduril was one of only two companies selected for the U.S. Air Force’s Collaborative Combat Aircraft program to build a new autonomous fighter jet, a rare win for a non-legacy defense prime. Physical Intelligence has constructed the largest robotic dataset ever assembled, while Hugging Face has more AI models available for machine-learning researchers than any other platform. Similar superlatives run across Together AI, RunwayML, Applied Intuition, Databricks, Eikon and many more.

These accomplishments seem straight out of science fiction, and that’s because they are. Science fiction has a history of becoming science fact, but rarely at this pace—a pace we see accelerating. We are finding and funding brilliant founders at the intersection of AI and the 3D physical world in robotics, biology, manufacturing, maintenance (or what we’ve dubbed fixware) and defense. In addition to biology and robotics, we are funding ventures (currently in stealth) using AI to transcend the crystalline cartography and silicon determinism of traditional chip design. Like the emergence of biological complexity from simple cellular automata, these machine-evolved circuits represent a critical shift where artificial cognition charts paths through electromagnetic possibility spaces that our carbon-based wetware can’t fully fathom.

Art imitates life, but increasingly, life takes its cues from art. We once looked to science fiction for escape; now, we look to it for instruction. The narrowing divide between sci-fi and sci-fact reflects a dynamic where our scientists and storytellers seem locked in a symbiotic exchange of ideas. The stories we consume don’t just entertain—they mold our perceptions, influence our choices, and convene us in a shared dialogue about the moral and technological dilemmas shaping our future. History suggests that fiction is not just a mirror of reality—it’s also an oracle.

Consider the instructive themes of several upcoming new TV series and movies in 2025. Mickey 17 follows an expendable—a clone that takes on highly dangerous tasks—who unexpectedly survives against its destiny, highlighting the quest for autonomy within the proscriptions of technology and society. The Electric State chronicles a teenager’s journey through a technologically ravaged landscape, reflecting on the human cost of unchecked innovation, while Prime Target dramatizes the high-stakes moral calculus of technological advancement alongside the paranoia of who should control theoretical knowledge.

The interplay between these speculative fictions and our real lives reveals an undeniable truth: just as market prices discount the future, our cultural zeitgeist anticipates it. The stories we consume shape the moral and technological dilemmas we will soon confront. Whether in a galaxy far, far away, a corporate boardroom, or the ruins of a shattered world, these narratives are not mere diversions—they’re clues for what comes next.

Notably: AI is democratizing, and it puts our propensity to tell stories on steroids. Our wildest imaginations no longer take decades of assiduous training to realize on film or paper, but can be conjured today with the right AI tool. It’s not merely automation; it is augmentation of human cognition at scale. If the stories we consume shape the future, then AI itself is the next storyteller—reshaping the narrative, bending reality to its scripts.

Digital Doppelgängers, Simulacra of Self & Breaking Free
One of the superlative fictional works on AI recently is Westworld. In the first episode, a human guest—tentative, probing—asks a robotic host with scrutinizing uncertainty, Are you real? Her response is chilling in its philosophical simplicity: If you have to ask, does it matter?

Spoiler alert: As the series unfolds, we learn that the hosts aren’t merely responding to human guests—they’re learning from them and training on them, iterating and improving their models until they can digitally (and eventually, physically) replicate them in every facet. The guests come for entertainment but leave behind their behavioral data, choices and desires, all feeding a system that understands them better than they understand themselves. If you think twice before accepting all cookies on a website, you might think thrice about what AI absorbs from you—not just what you explicitly give it, but everything it silently infers.

We’ve already made the leap to sci-fact. Many of the companies we’ve backed—from Matterport (3D scanning) to Auris and now Mendaera—are constructing digital twins, simulacra that approximate reality with ever-finer precision. One new thesis we are interested in is the digital twinning of software APIs to agentic AI, which we believe will eventually lead to the death of APIs and require a rethink of the SaaS investor playbook. Where APIs provided standardized exchanges between client and server, AI agents could move more fluidly, navigating interfaces and reducing reliance on traditional integration points. The new battleground might not be who owns the API, but who controls access, permissions and the AI-mediated flow of information.

Yet it’s the digital twinning of ourselves through AI that we are most intrigued by. Every interaction is a data point, every hesitation a signal, every repetition a training loop. AI doesn’t just learn what you type, but how you type—the rhythm of your keystrokes, the insertions and deletions and follow-up questions that reveal your evolving thought process. Imagine this as a kind of digital exoskeleton—an invisible but intimately form-fitting mesh, adhering to you like a Marvel superhero’s suit, but instead of merely replicating your physical contours, it maps your essence.

We’ve all asked an AI model to rewrite a dull article in the punchy financial satire of Matt Levine, the florid Victorian prose of Oscar Wilde or the acerbic, unsentimental wit of Rachel Cusk or Christopher Hitchens. It’s a cultural Turing test in action—AI’s creations may be derivative, but they are increasingly indistinguishable. Yet, with the right training, we can go further. Future AI reasoning models could actually write in the same way as us, stopping on a thorny passage that needs rewriting or pausing to think outside (and perhaps actually observe a digital twin of a garden for inspiration). Some of today’s thinkers even argue that our only goal should be to train AI to imitate our unique personalities and thought patterns—to effectively build that digital exoskeleton as early as possible and liberate us.

Yet, there are limits. How will we evolve our tastes? How will we rebel against our pasts? Today, social media algorithms show us content that we have liked before, never letting us escape our former selves. Can AI handle aging? Or will the data exhaust you leave behind—the weights and biases of a model trained on you—cling like a programmed Python script, defining you, coiling around you, constricting and constraining? Will you be forever locked in the caricature of your past self, unable to break free, unable to disrupt your own patterns, unable to author a new version of you? When will AI models forget?

Offensive openness: Open talent, open source, and decentralized compute
When it comes to superlatives, no country offers more extremes than China. A decade ago, the U.S. was the undisputed leader in nearly every critical technology. More than anecdotal—it’s empirical: today, China commands the lead in 37 out of 44 key fields, from materials and agricultural sciences to photonic sensors and electric batteries. China now produces the majority of the world’s most-cited, high-impact research across these domains, in some cases generating five-to-ten times the volume of breakthrough work as the next country. Research leads don’t always translate immediately to commercial or military dominance, but they are the strongest predictor of who will control supply chains, infrastructure and economic leverage in the coming decades.

Our approach in America must be offensive openness, doubling down on precisely the strategies that made America great the past century. We must continue to attract the world’s brightest minds, continue to open software so that our leading talent has access to the best tools and can become the most ambitious and competitive entrepreneurs in the world, and continue to decentralize digital infrastructure to ensure its resiliency, adaptability and speed in the rapidly changing world ahead.

Today, 50% of undergraduate AI researchers worldwide hail from China, and nearly 40% of AI researchers in the U.S. were originally trained there. Yet, our policies have intentionally made it harder to retain these critical workers. Over the past decade, the U.S. has seen a significant decline in international student enrollment, from 23% in 2000 to 15%. The AI arms race won’t be won with silicon alone—it’s a war for minds. Lose the talent, lose the future. Just as World War II was fought with Europe’s best minds in the service of peace and prosperity, any future war with China will be fought by all Americans alongside dissident Chinese scientists who yearn for freedom and liberty and are willing to fight to return those values to their homeland. Open-source AI and OSINT will also shape the battlefield with the coming rise of algorithmic warfare. Today, GitHub’s global army of committers now outnumber the Pentagon’s contractor workforce 100:1. Democracies debug in daylight—dictatorships train in darkness.

Open talent begets the importance of open-source software. DeepSeek’s R1 model re-opened the minds of investors about the leading role of open-source AI. Having poured tens of billions of dollars into models that could be replicated and commoditized, investors in large foundation models are now left reassessing their strategies.

The pendulum has now swung decisively in favor of open-source AI, a shift we anticipated and positioned for. Hugging Face and Together stand as dominant leaders, as do RunwayML and Applied Intuition with differentiated applications, while Sakana AI—Japan’s foremost AI company where Lux led a founding investment—pushes forward with radically efficient architectures untethered from the brute-force capex scaling laws governing OpenAI and Anthropic. While OpenAI and Anthropic build remarkable products, their business models invite structural risks: Anthropic’s compute reliance on Amazon sets the stage for eventual absorption, while OpenAI faces an insatiable capital appetite, deepening entanglement with Microsoft’s tightening grip and the relentless ire of Elon—who, now politically adjacent to power, can obstruct Sam Altman through Grok/X.ai in ways that may be underestimated. The true differentiator for future AI advantage? Proprietary data. If open models match closed-source performance, then those with deep reservoirs of unique data—Meta (via WhatsApp, Instagram) and Bloomberg—will emerge as kingmakers. Lux’s prior letters previously predicted the pivot from unbridled GPU and large-cluster demand toward smaller, on-device models and the potential resurgence in importance of memory players. We also expressed skepticism toward the overcapitalization of modular nuclear, forecasting that natural gas and large-scale nuclear plant restarts would prove more durable.

There’s a deeper infrastructure change underway though. Open-source AI models allow for immense experimentation, as the combinatorial brilliance of millions of scientists using Hugging Face to edit and update models can attest. Yet the best ideas often hit a compute wall: re-training models or deploying them for inference often requires massive central and proprietary data centers with high fees that are out-of-reach of the tinkerer not employed by one of a handful of companies. That’s why we expect a significant shift in capital allocation toward memory and new chip architectures that optimize for efficiency and speed. Over the next few years, we anticipate that up to 50% of AI inference will happen on-device—reshaping the AI value chain. By increasing decentralization of compute, improvements and specializations in base models can rapidly spread among end users, finding their natural customers much faster. AI winners will not be those who train the biggest models—but those who make them run everywhere, instantly and at near-zero cost.

AI’s encroachment isn’t just personal—it’s structural. The same forces that threaten to ossify individual identity are at play in broader arenas, locking institutions, industries, and even nations into patterns that may be increasingly difficult to escape. This is the defining struggle of our time: the real battle isn’t just human vs. machine but open vs. closed, inertia vs. reinvention, state control vs. market dynamism, incumbents vs. insurgents and West vs. CRINK (China, Russia, Iran, and North Korea). At Lux, we don’t just see this as an investment opportunity—but a moral imperative. The technologies that will define economic strength, military power and societal resilience can’t be ceded to a regime that manipulates markets, suppresses dissent and weaponizes interdependence. Betting on breakthrough science and advantaged technologies is first and foremost about financial returns—it’s also funding the future we want to live in, ensuring that it remains shaped by open societies, free markets and technological leadership rooted in democratic values.

Disrupting Those Who Fund the Disrupters
The conventional wisdom about automation has been upended. While many expected blue-collar jobs to be the first casualties, AI’s initial disruption has largely targeted white-collar knowledge workers—lawyers, accountants, designers and marketers. This raises an uncomfortable question: are venture investors immune?

We can imagine a future where every investment decision—every pitch meeting, partnership discussion and past outcome—is recorded, transcribed and analyzed. AI could quantify what has long been more art than science, applying Bayesian models to assess past decisions and Kelly criteria to optimize portfolio sizing. An AI-enhanced investment committee could emerge, with specialized models acting as advisory partners: one tuned to founder psychology, another tracking market dynamics, a third pattern-matching against historical successes and failures.

But rather than putting us all on the same “greige” page—a dull, homogenized blend of grey and beige—AI might amplify cognitive diversity. By surfacing blind spots, identifying hidden biases and challenging consensus thinking, it could increase variance in decision-making, making firms sharper rather than softer. The implications remain uncertain, but one thing is clear: those who embrace AI’s potential early may gain an edge in a rapidly evolving landscape.

The convergence of AI, geopolitics and capital is happening in real-time. What remains constant is the necessity for strategic foresight, an investor mindset willing to embrace volatility and a relentless pursuit of differentiated innovation. Lux remains steadfast in backing the pioneers who will shape this next era, and especially in the pursuit of funding and commercializing cutting-edge science and technology approaching that elusive asymptote of truth.

With all the talk of artificial and fictions, we must remember that the transitive circular logic of truth is that the pursuit of profits is the pursuit of scientific advantage is the pursuit of truth. It was Roman renaissance man Cicero who is thought to have said, “would that I could discover truth as easily as I can uncover falsehood.” And French painter George Braque who said, “truth exists, only falsehood has to be invented.”

‍

Truth cares not for our comfort or illusions, it just is. It demands persistence, skepticism and the willingness to accept uncomfortable or inconvenient facts. The institution of science in its relentless pursuit of facts—those stubborn, irrefutable fragments of reality—reveals a truth about truth itself: it is not a single, gleaming artifact to be unearthed, but a mosaic painstakingly assembled from countless shards of superlative evidence, tested under the light. Fiat Lux.