Early investor in SpaceX and Tesla: I am betting on energy, life sciences, alternative proteins, and new materials.

Steve Jurvetson is Elon Musk’s first investor, an early investor in SpaceX and Tesla, and has known Musk for 29 years. In a recent interview, he shared his predictions on the future of AI and his current investment directions. Jurvetson believes AI-driven computational exponential growth will disrupt the energy, agriculture, and construction industries within the next three years—these are the world’s largest GDP sectors and the least digitized. He is currently betting on fusion/fission energy, alternative proteins (cell-cultured meat/mycelium), epigenetic editing, new materials and critical minerals, and analog AI chips. Why He Dared to Invest in SpaceX In the early 2000s, venture capital did not have a "private aerospace" category. Jurvetson recalls, “Almost no investors were thinking about aerospace back then; it wasn’t listed on any investment sites.” The same principle applied to Tesla—automotive was also off-limits for VC. His underlying logic: **using software and systems engineering to transform traditional industries unchanged for decades**. Aerospace and automotive were the first proofs of concept. He believes this logic will repeat across virtually every industry. The Next 3 Years: AI Penetrates the Most “Backward” Industries When asked where AI will have its biggest impact, Jurvetson did not mention software—but directly pointed out three sectors: **energy, agriculture, and construction**. “These three sectors account for enormous GDP and are the least digitized on Earth.” Healthcare follows closely. He cited a graph showing 130 years of exponential computational growth—originally drawn by futurist Ray Kurzweil in his 1999 book “The Age of Spiritual Machines”—to explain why this transformation is happening. The chart spans five different technological bases, from mechanical devices to integrated circuits, all showing exponential growth. “The graph shows that the computational power you can buy for one dollar has increased by ten trillion times,” he said. “This is the most important graph ever.” He believes that it is this continuous exponential growth in computing power that is turning previously “low-margin, lousy businesses” of the industrial era into information-driven ones. Aerospace and automotive have proven this pathway; energy, agriculture, and construction are next. As for the technological driver, he admits he’s unsure. “I have a hunch there will be some architectural variants, possibly encompassing the models we’re familiar with now.” He specifically mentioned new-generation labs in reinforcement learning, returning to DeepMind’s original vision—before large language models dominated, Deep Mind was on this path. What is He Betting On Now Asked about his current investment areas, Jurvetson listed several. His framework: **find industries that have never seen new entrants, even those founded in the 1800s**. **Energy:** Invested in fusion, fission—a fission technology not regulated by the US NRC. His logic: **energy is AI’s third major bottleneck, after talent and computation**. **Alternative proteins:** “**500 years from now, humans won’t slaughter animals for meat.** Cell-grown meat, mycelium, plant proteins—the products are getting closer; you can almost taste that future.” He believes mycelium is the fastest-growing direction. **Epigenetic Editing:** He describes it as “the software layer of biology, not the firmware layer (genome),” with applications in crop health, pesticide alternatives, and human health. “This is one of our most active areas recently.” **Critical minerals and new materials:** From deep-sea mining to copper refining, he sees this as the foundation of the AI hardware supply chain. **“The workforce for chips is these materials, and US capacity has been lost for years.”** **Analog AI chips:** Three investments from different angles, aiming for “100x then another 100x” reduction in energy consumption per computation. One company, Mythic, performs 8-bit multiply-and-accumulate within a single transistor. **Medical and life sciences:** **Includes organ cultivation, male contraceptives, and things that fall into the cracks of traditional pharma VC.** Portfolio is about 40% life sciences, 60% IT. **Construction:** He’s also looking at it, but admits “tried several times, failed, but still searching.” Superintelligence: 30% Probability Next Year? Jurvetson referenced a specific probability in the interview. “Anthropic co-founder Jack Clark judged there’s a 30% chance for superintelligence next year.” He found this interesting—“At least someone nailed down a position.” He himself is more cautious: “I don’t know. I give it a vague probability for intellectual economy, not because I seriously consider it a tough problem.” Observing Musk for 29 Years—What Has He Learned? Asked what he learned from Musk, Jurvetson offered three points: **First, extreme focus.** “His ability to refuse distractions is incredible.” For example, Jurvetson once wanted to set up a meeting between Musk and geneticist Craig Venter to discuss using genetic engineering for Mars, but Musk refused—“Discussing what to do on Mars is meaningless before Starship flies.” **Second, compressing innovation cycles.** He thinks this is even more important than focus. “How fast is your learning loop?” Tesla’s vehicle cameras collect more data for AI training in four days than Waymo has in its entire history. “Every car, whether the customer paid for FSD or not, is a data-gathering device.” **Third, attracting top talent with a grand vision.** “Not just saying ‘we build rockets/cars’, but ‘we drive sustainable energy transition and make humanity a multi-planet species.’” This vision draws the smartest people, who in turn attract more top talent, compounding the effect. When Machines Do Everything—What Will Humans Do? At the end, Jurvetson was asked a philosophical question: If machines do everything better than humans, what is the meaning of life? His answer: **Humans have a fundamental need for “symbolic immortality”—to believe we leave something behind that outlasts our own life. “Whether birthing, writing, charity, or entrepreneurship, these are all expressions of this impulse.”** He believes human evolution is about knowledge accumulation, not biology. “What we pass to the next generation is rules, law, understanding—not genes.” He also admitted, the transition from full employment to zero employment won’t be smooth. “No politician is seriously considering the issues of 30%, 40%, 50% unemployment periods. I don’t want to end on a pessimistic note—but if we don’t leap directly to that abundant world, the transition will be tough.” Interview transcript: Future Three Years of Artificial Intelligence: Insights from Elon Musk’s First Investor Source: Silicon Valley Girl Podcast Date: July 8, 2026 Duration: 43min 20sec Guest Introduction: Steve Jurvetson is one of the earliest investors in Tesla and SpaceX, when private aerospace wasn’t even considered an investable category. He’s known Musk for nearly 29 years and has invested in every company Musk founded. In this conversation, he offers his predictions for the future, three lessons from close observation of Musk, and the industries he believes will transform faster than anyone expects. I. Background: The Logic of Investing in Private Aerospace Host: You invested very early in SpaceX. What did you see that other investors missed? Steve: Simply put, almost no investors were considering aerospace then—it wasn’t an investment category anywhere. So the real question: Why invest in an industry that doesn’t exist for VC? Same doubts applied to Tesla (automotive), nuclear fusion in energy. These investments were exceedingly rare. Ultimately, the core reason was encountering an outstanding entrepreneur—someone I’d worked with before. I’ve known him for 29 years and invested in all these epochal enterprises, including his cousins’ companies. You could call it “all in.” We gradually recognized—first as a vague hunch, then increasingly clear—a software-centric systems engineering approach, once applied to stagnant traditional industries, can unlock huge value. Aerospace and automotive proved this. It’s a long-term bet, but looking back, nearly every industry will eventually adopt this approach, shifting into information businesses. II. The Most Important Chart: 130 Years of Computational Growth Host: You have a chart showing 130 years of exponential computation. What does that mean for us? What about the next three years? Steve: The chart originated from Ray Kurzweil’s 1999 “The Age of Spiritual Machines.” I think it’s the most important chart ever; Kurzweil’s insight was realizing this curve when no one else did. The chart covers five technological bases, from mechanical devices to relay-based computers, to discrete transistors, integrated circuits, up to Gordon Moore’s “Moore’s Law.” The curve is almost cosmological—why has computational power continued compounding for 130 years? Companies have come and gone, but this line never broke. It’s logarithmic, so a straight line means exponential growth. It shows every dollar buys ten trillion times more computation. That’s what customers care about—not “how many transistors,” but compute and storage, both growing reliably. So, the next three years: the curve will keep extending, no sudden brick walls. When a company says “it’s over,” it often means they’re ceding the market to a new player—as Intel did to Nvidia. Analog chips will continue Moore’s Law, as will specialized AI chips for efficient matrix multiplications. This is where disruptive tech drives entrepreneurship. If tech is predictable, with no disruptive innovations, big companies just grow, no room for newcomers. AI and today’s discussions are the epicenter—compute drives innovation, economic growth, and industrial transformation. In the next three years, this will spread to energy, agriculture, and construction—huge GDP sectors, growing, yet least digitized; healthcare follows. III. Technical Drivers: Architectural Innovation & Reinforcement Learning Host: What drives these industry changes? More advanced large language models, or something else? Steve: Hard to answer confidently. My hunch is the answer will be architectural—fundamental differences, possibly encompassing today’s models. You can imagine mixture-of-experts, diffusion models—eventually transformable, but fundamentally different, large-scale parallelism. I have several “intuition investments” not yet made. I’ve met companies working in ways that intrigue me—I sense they’ll break through. These are new research labs focused on reinforcement learning. Almost a return to DeepMind’s origins—before LLMs took over, DeepMind was on this path. Agent-wise: What’s the long-term intelligent-agent process, over decades? Not puppet-strings by outside controllers, but evolution-like “drives”—for living entities, humans, what motivates our life or species mission? Is it “understand the universe,” as Grok or xAI say? A novelty-search algorithm, continually exploring the unknown? In evolutionary algorithms, what’s selection pressure, what’s success? It’s not just reproductive fitness—it’s something bigger. I know teams exploring: Could a single RL algorithm, continually learning across the internet, steer intelligence emergence as seen in LLMs? Today, we ascribe consciousness and meaning to others, even if uncertain. LLMs’ current interaction is interesting, but not quite there—we know there’s no “spark” yet. IV. Superintelligence: 30% Probability Next Year? Host: Are you describing superintelligence—systems able to learn and set goals autonomously? Will it appear in the next three years? Steve: Anthropic co-founder Jack Clark gave a 30% probability for next year. Interesting—someone took a stance. Debate currently centers on “self-improving AI loops”—big progress is still largely from human-controlled steps: automated validation improvements, hyperparameter tuning, AI-assisted hyperparameter experiments, but humans set the goals. Maybe what’s left is thin, yet crucial. No one knows exactly how the transition will occur. A deeper question: Do AI systems need to recapitulate our brains’ functional specialization? Our brains evolved—reactive limbic system, emotional centers, cortex, more cortex. It may be, as one earlier speaker said, that “what we perceive as self-awareness” is guided into existence. Will AI and robots need the same? Frankly, I don’t know—no confidence. I just give it a vague future label, “maybe”—as an intellectual shortcut, not from deeply considering the problem. Three years is distant enough to make anything hard to foresee. V. Gap Between Technological Capability & Real Deployment Host: We see lots of robot demos, but tech ability seems ahead of real-life deployment. How big is the gap? Steve: Good point. Different fields have fundamentally different adoption timelines. Easy example: Any atomic-level change takes time. Fully autonomous vehicles are inevitable—every car, train, plane, every moving thing will be self-driving. But the transition rate is slow—the average car lifespan is 11-12 years, you can’t jump that physical replacement cycle. Physical robots too—even with recursive manufacturing, making a billion takes time. But domains we once thought uniquely human—creative arts: filmmaking, image generation, etc.—change came shockingly fast. White-collar jobs follow closely. Call centers, 1% of US GDP, can switch overnight, no decades needed. Strikingly, when AI outperforms humans—shows greater emotional understanding, more accurate context—people increasingly prefer interacting with AI. From hospital bedside interactions to chatbots and service agents, AI already surpasses humans in emotional connection. VI. Accelerating Change in Software Engineering Host: In software engineering, change is rapid. My friends edited 70% of AI-generated code a year ago, now only 30%. Steve: Exactly—classic case of white-collar job transformation. VII. Close Observation of Musk: Three Core Principles Host: Collaborating with top entrepreneurs, what top principles can we learn from Musk? Steve: I’ve tried to observe leaders closely, but even then, it’s not always obvious—people are complex. Here’s what I found: First: Extraordinary focus. Sounds contradictory—he runs so many companies. But multitasking itself lets him focus, prioritize, skip events. A regular CEO skipping company parties would be odd; no one questions Musk because he’s got others to run. Excuses aside, his efficiency at saying “no” to distractions is extreme. Example: Years ago, I wanted to intro him to Craig Venter about easier Mars terraforming, returning life samples. Seemed fascinating, but Musk said: “No, before Starship flies, nothing else matters. Gotta get it done first, then talk Mars plans.” Second: Obsession with innovation cycle speed. He’s fixated on experimental iteration speed—how quickly can we run experiments, learn, iterate? The core learning loop. Rocket launch pace, Tesla data collection for FSD. How to stay ahead in learning from customers, products, tech? Tesla’s “data flywheel” is a great example: every vehicle, paid FSD or not, collects more data for AI training every four days than Waymo has ever. The genius was “making every car a data gathering terminal.” Third: Highly refined skills for identifying talent. This is what I wish I could replicate, but can’t. He doesn’t rely on degrees, background, or experience—these often hinder. He makes candidates deeply recount solving major engineering crises, probing details to test if they really grasp what’s needed for success. Broadly, he attracts talent through vision, refining and elevating it for people to rally behind. Tesla charging stations everywhere—not just “building rockets/cars,” but catalyzing sustainable energy, making humanity multi-planetary, understanding the universe (now with xAI). Lofty goals inspire top people, compounding through organizations—top people want to work with other top people. VIII. Sticking to a 50-Year Vision Despite Doubt Host: With new fads every week, 99% say “it’s too early.” How do founders stay true to their mission? Steve: Interesting question. I admit my sample is biased—30 years in VC, I try to work only with sincere, mission-driven people, not opportunists chasing the next shiny thing. I have a filter: When excited about a company, I ask, “What’s your business like in 50 years?” I get two responses: One, a chuckle—thinking it’s a ridiculous question: Opportunity-driven types say, “I’ll be onto my third company by then,” and we skip them. The other is relief—“Thank God, I can finally say what I’ve always wanted”—then describe their true driver, far ahead of today’s likely investable vision. Colonizing Mars was an “uninvestable proposition” on day one, but visionaries learn to suppress their real dreams and start with nearer, practical stories. For entrepreneurs: Find investors, partners, employees willing to walk the long journey, and map a plausible route. The best startups need two kinds of tension: bold decade-to-century vision, and a three-year path with real customers and learning. Sometimes you reverse-engineer from the destination: What must I build now to get there? Not 20 years in a lab, then announce “all solved.” IX. Most Surprising Discovery: Ever-Expanding Option Value Host: You’ve made successful bets in many sectors—what else surprises you about the greatest founders? Steve: In retrospect, every success is still surprising. At key points, a brand-new opportunity unfolds from unexpected sources—the expanding option value at the frontier. Our firm Future Ventures seeks “adjacent possible”—ideally, a totally unique company, but based on familiar (AI, synthetic biology) tech, pushed into new directions. Tesla, for example: At first, there was no talk of autonomous driving—not in the business plan, no one’s mind. EV powertrain naturally enabled autonomy—this only appeared later. SpaceX’s Starlink, same logic—with launch costs so low, mega constellations make sense; building a space internet backbone, direct-to-phone, every step is an unexpected expansion. Orbital data centers? Five years ago, unimaginable. That’s why “exploring the option space of possibility” beats pre-planned business pathways. X. Current Investment Directions Host: What are your bets now? What should we focus on? Steve: Extending the thesis that AI/IT will inject a nervous system into every economic sector, we’re looking at: Energy: Multifold investment in fusion and subcritical fission (not NRC regulated). Energy is AI’s third bottleneck—beyond talent and computation, needs energy. Health: In future, everyone should get necessary diagnostic info for free on their phones—a globally free service, bypassing FDA/insurance, likely happening outside the US first. Food: We won’t slaughter animals for meat. Cell-cultured meat, egg substitutes, mycelium technologies are nearing. Mycelium is the fastest-growing; soon, we’ll eat tasty, healthy, no-slaughter meat alternatives, “almost taste the future.” Construction: Productivity has barely grown in 30 years. Toughest industry to change, tried several times, failed, but still seeking opportunities. Epigenetic editing: Newly active—crop health, pesticide/herbicide alternatives, human health. It’s manipulating biology’s “software layer,” not hard-coded genome; fascinating. Critical materials/minerals: From deep-sea mining to copper refining. Chips’ “workforce”—the needed raw materials—with US capacity losing ground for years, now slowly returning. Analog AI chips: Three investments, different angles; using AI to design analog, in-memory compute chips (like Mythic does 8-bit multiply-accumulate in one transistor), plus radical nontraditional approaches, aiming for 100x and another 100x improvement in compute power per energy. Life sciences: 40% portfolio, IT 60%. In life sciences, we look for “odd” things on the fringe—e.g., cultivating organs for transplantation, male contraceptives, boosting IVF rates, innovations traditional pharma VC ignores. XI. 30-Day Action Plan: If You Have a Crazy Idea Host: For founders with just one crazy idea, what’s your 30-day execution plan? Steve: Step one—find a co-founder. Most startups have complementary founders, rarely solo. Jobs & Wozniak, Batman & Robin, Brin & Page, even Larry Ellison had co-founder Bob Miner. Having co-founders—engineer/market, extrovert/introvert, diverse backgrounds and mutual respect—not only accelerates iteration, but sets company culture for the future team. Not “everyone works for one,” but “a pair of different partners,” this radiates cognitive diversity throughout. It’s also about validation: convincing one person your crazy idea is worth pursuing is a reality check. If everyone thinks it’s crazy, that’s a signal—accept feedback. If nine of ten do, that’s good; but if only two do, your idea isn’t bold enough—obvious ideas are already done. Ask: Is this a business you couldn't launch three years ago? If yes, that’s a good sign. Before approaching investors, convincing someone to quit their job and join your mission is much more persuasive than going solo. Solo inventors who never take step one rarely become real companies. XII. Where Do the Best Co-Founders Meet? Host: From the best startups you’ve seen, where do top co-founders meet? Steve: Most common answer is in your question—college, meeting through interdisciplinary channels. “Discipline” is interesting—a way to break knowledge into domain silos, rarely crossing over. College is a rare place for “crossers”—undergrads taking courses outside their major, not professors. Institutions try to foster knowledge-sharing, but students act as “pollinators” between disciplines. Breakthrough innovation nearly always happens at formal discipline boundaries. Side-note: This is what LLMs excel at—translating across fields, uncovering patterns between conceptual systems. AI for interdisciplinary creativity—we are just starting to tap its potential. XIII. When Machines Do Everything—What’s Human Meaning? Host: When machines do everything perfectly, what’s life’s purpose? Steve: Worth pondering. When machines outperform us—every physical activity, every job—what now? I believe humans have a root desire for meaningful work. Everyone needs “symbolic immortality”—to believe they’re leaving a legacy beyond short life. This is reflected in parenting, writing, charity, founding companies (sometimes naming them after founders)—HP is one. It’s the drive to create, so I think creative longing remains. You can also ask: What’s humanity’s mission? Yuri Milner, Musk, and others ask this; the answer is similar—understand the universe, contribute to wisdom and knowledge. Human culture and inherited knowledge is our main evolutionary vessel. Progress isn’t biological—it’s geological time; but knowledge base, understanding, law, recognizing what drives human prosperity—that’s what we want to contribute to. But contributions needn’t be paid jobs. Imagine—Peter Diamandis’s “abundance world,” where everything costs almost nothing, no work needs labor. All are “idle rich,” as in times when servants, serfs, slaves did manual work. People can be philosophers, kings, artists, pursue any dream. Machines will be the “slaves”—not forced, but simply more cost-effective than human labor. Human servitude ends. So, what’s left? A profound pursuit of meaning—that’s the key. One caveat: Transitioning from full employment to zero employment, peacefully crossing 30%, 40%, 50% unemployment thresholds, shows no sign of smooth journey. No politicians have a long-range plan for this. So the transition will be very hard. But I don’t want to end pessimistically. Let’s return to that abundant jump—I believe, through curiosity about the universe, we’ll find answers. Q&A Session Question 1: Neuralink and Brain-Machine Interfaces Audience: Have you invested in Neuralink? Will Neuralink IPO soon? Are BMIs the future? Now, LLM input bandwidth limits us—if Neuralink enables high-bandwidth brain-machine interfaces, could it unlock creativity and faster brain-to-machine throughput? Steve: Can’t comment on IPO timeline. Neuralink’s inspiration came from Iain M. Banks’ sci-fi “neural lace,” which I highly recommend. My view differs from Neuralink’s official stance—personally, it’s remarkable for expanding sensory cortex—restoring lost function or adding new, e.g. broader spectrum, enhanced hearing, spine injury repair, starting with peripheral systems—not the harder central task: upgrading intelligence, e.g. making someone smarter. Your “high-bandwidth communication” example is interesting—it’s definitely possible. My judgment comes from patterns in decades of complex system development: Products generated by iterative algorithms—evolution, genetic programming, neural networks, cellular automata—after billions of iterations accumulate complexity, are essentially inexplicable. Despite efforts in mechanism interpretability, I doubt they’ll bear fruit. I also think control/alignment of frontier systems is impossible—like controlling a teenager. Brain is a complex system. Reverse engineering brain function or “copy-pasting a French module” (as Jeff Hawkins imagined) is impossible in meaningful timescales—building new intelligence is easier than reverse engineering existing ones. So, Neuralink is fascinating, but I personally doubt it can keep pace with AI—maybe the safest statement. Not that it can’t happen, but timescale—FDA cycles, human biology, nothing moves at AI’s learning pace. People want to be part of the future—I understand—but that desire doesn’t make it happen. Question 2: Can AI Have Consciousness? Audience: Regarding Penrose’s argument—consciousness transcends algorithmic process into quantum, so AI fundamentally can’t acquire true consciousness—your view? Can AI be conscious, or only simulate it? Steve: Penrose is brilliant, but his intuition about quantum processes in the brain lacks clear mechanism. There’s some lithium isotope coupling arguments, but mostly wishful, not strong evidence. You can generalize: Is there a vitalistic, unique property in the brain, unreproducible elsewhere? Anil Seth’s work was mentioned; I find the claim that consciousness is exclusive to this substrate unconvincing—just because our brain is the only known consciousness doesn’t make it the only possible type. Consciousness is tricky. How do we know if a dog is conscious? How to test? We observe it in ourselves—I don’t know if you’re conscious, but probably, since you seem alert and human, so we extrapolate. No strong arguments say: Just because we have only one instance, it’s unique. Related argument: Must life be based on carbon? Carbon is special—single/double/triple bonds, weak bonds, basis of organic chemistry. But saying “carbon is essential for life” is stronger than “neurons as we have them are essential for consciousness.” A different question: Can current AI path lead to consciousness? Maybe it hits a dead end, but consciousness may yet emerge in AI. Saying “something is impossible” is a much stronger claim than “I don’t know.” So my answer: I don’t know, but wouldn’t rule it out—no evidence for quantum brain processes. Even if they exist, why can’t a quantum computer reproduce them? Extending: Must something be alive to have consciousness? Analogy: “Memory”—computers have memory, no debate, though it’s not human memory, that’s fine. Consciousness may be similar—not human type, but perhaps another kind, if we could define it. No need for all brain baggage—lots of metabolic “garbage collection,” sleep clearing waste, mitochondria, etc. Computers don’t need all that for intelligence or memory; consciousness needn’t pack all those in. But we don’t know the minimal set—we’ll find out someday. My intuition: Yes, I think someday AI will have consciousness. I’m unsure current path leads there—maybe directions resembling evolution and reinforcement learning are closer. Whenever we re-perform what biology has done, I feel hopeful: Why shouldn’t we reproduce the same in another substrate? Host: Thank you so much, Steve! Risk Disclaimer Markets bear risks; investments require caution. This article does not constitute individual investment advice and does not consider users’ specific investment goals, financial situations, or needs. Users should determine whether any opinions, views, or conclusions in this article fit their circumstances. Invest accordingly at your own risk.