In venture capital, conventional wisdom counsels incremental thinking: back a proven market, look for a team with domain experience, find product-market fit before writing a large check. That framework has produced competent returns for generalist funds for decades. It has also systematically missed every defining technology company of the last twenty years. Google, Tesla, SpaceX, OpenAI — each of these was, at seed stage, a bet that looked reckless to the consensus. Each required investors willing to believe that the world would change in ways that the market had not yet priced.
At Estes Capital, our investment thesis begins with a different question than most seed funds ask. Rather than "is there an existing market here?", we ask: "if this technology works at scale, what size market does it create?" That distinction is not merely semantic. It determines whether you can see the opportunity that is invisible to everyone else — and whether you have the conviction to act on it when the evidence is still thin and the ridicule is still loud.
This essay is about why audacious bets outperform cautious ones in deep technology investing, what the historical data actually shows about radical vs. incremental bets, and why the specific moment we are in — 2024 and beyond — demands even more willingness to back the seemingly impossible.
The Mathematics of Venture Returns Rewards Extremism
Start with the math. Venture capital returns follow a power law distribution. A small number of investments account for the overwhelming majority of returns in any fund; the rest either return capital or are written off. Cambridge Associates data shows that the top 10% of venture outcomes account for approximately 90% of aggregate industry returns. This is not an accident — it is a structural feature of markets where winner-take-most dynamics apply, where technology advantage compounds, and where network effects create durable moats.
The consequence for investment strategy is counterintuitive but mathematically clear: if you are managing a seed portfolio, the expected value of a 100— return on a 1% position is greater than the expected value of a 3— return on a 15% position. The fund-level math rewards backing outliers, not avoiding them. A cautious portfolio of "reasonable" bets — companies in established markets with clear paths to modest exits — will consistently underperform a concentrated portfolio with the occasional home run, even accounting for the higher miss rate that comes with more ambitious bets.
This is why Khosla Ventures, arguably the most disciplined practitioner of what we might call "audacious seed investing," has consistently produced top-quartile returns despite — or rather, because of — backing technologies that were widely dismissed at the time. Vinod Khosla's public writing on this is unusually direct: he views a 90% failure rate on his portfolio as a sign that his team is being sufficiently bold. The 10% that work, work spectacularly, and the mathematics takes care of the rest.
OpenAI: The Investment the Market Didn't Believe In
Nothing illustrates the power of audacious conviction more clearly than OpenAI's early funding history. When Sam Altman and Greg Brockman approached seed investors in 2015 for what was then a non-profit AI safety research organisation, the reception was polite but skeptical. The goal — artificial general intelligence — was described by mainstream machine learning researchers as "science fiction." The organisation had no product, no revenue model, and a stated objective that sounded like something from a 1950s science fiction novel.
Microsoft's initial $1 billion investment in 2019 was widely mocked in enterprise tech circles as "paying for science." The thesis that large language models could become general-purpose commercial tools was not yet demonstrated. GPT-2 had shown interesting properties, but the leap from "interesting research result" to "business with tens of millions of paying customers" required believing that scaling would produce qualitative improvements in capability — a hypothesis that the consensus ML community was deeply skeptical about.
By 2023, OpenAI's valuation had reached $29 billion on secondary markets. By late 2024, following the company's latest primary round, the valuation stood at approximately $157 billion — one of the most rapid value creation events in venture capital history. The investors who acted on early conviction — who saw the potential for general-purpose AI when the product did not yet exist — generated returns that no amount of careful market analysis could have predicted. The returns were created by audacity, not caution.
The broader lesson from OpenAI is not that you should back research organisations indiscriminately. It is that technologies which appear to have no plausible path to commercial value often turn out to be the ones that reshape entire industries. The analysis that correctly identifies these moments looks less like financial modeling and more like scientific judgment: does the underlying technology have the potential to improve at a rate that will eventually make it the dominant solution across a broad class of problems?
Anthropic and the Return on Scientific Conviction
Anthropic's story is instructive for a different reason. When Dario Amodei, Daniela Amodei, and a team of senior OpenAI researchers spun out to found Anthropic in 2021, the conventional fundraising logic would have suggested backing them on the strength of their team credentials alone. Instead, the early investors — Spark Capital, Google, and a handful of others — were backing a specific technical thesis: that constitutional AI, a set of training techniques designed to make large language models more aligned with human values and more reliably safe, would produce commercially differentiated models.
That technical bet has proven remarkably well-calibrated. As of 2024, Anthropic had raised approximately $7 billion in funding at a valuation of around $18 billion, with Google and Amazon each committing multi-billion dollar investments that reflected their view of Claude as a strategically critical AI platform. The differentiation that Anthropic's founders predicted — that safety and reliability would become competitive advantages as AI moved into enterprise — has proven correct. Enterprise customers running critical workflows on AI infrastructure have shown a willingness to pay a premium for models they can trust, and Anthropic's technical investments in alignment research have translated directly into commercial positioning.
What makes Anthropic particularly relevant as a case study for audacious seed investing is the nature of the bet. At founding, the company was backing a scientific hypothesis about training dynamics and alignment that had not been empirically confirmed at scale. The investors who wrote the earliest checks were not evaluating a product that existed; they were evaluating a technical thesis and a team's ability to execute on it. That is precisely the kind of judgment that distinguishes deep technology seed investing from growth-stage investing, and it is precisely the kind of judgment that requires both scientific literacy and investment discipline simultaneously.
The Deep Tech Inflection We Are Living Through
The argument for audacious deep technology investing is not merely historical. It is structural, and it is urgent. We are living through a period of technological inflection that is unusual even by the standards of the last fifty years. Several distinct technological capabilities — large language models, protein structure prediction, gene editing, advanced robotics, nuclear energy, electrochemical manufacturing — are crossing commercial viability thresholds simultaneously. The interactions between these capabilities are unpredictable and potentially enormous.
Consider the trajectory of the robotics sector alone. Figure AI, founded in 2022, raised $675 million at a $2.6 billion valuation in 2024, with backing from Microsoft, OpenAI, and Nvidia — a fundraise that would have been inconceivable for a hardware company at that stage five years earlier. Physical Intelligence, founded by researchers from Google, Carnegie Mellon, and Stanford, raised $400 million at a $2.4 billion valuation in 2024, despite having no commercial product at the time of the raise. 1X Technologies, a Norwegian robotics company building humanoid robots for physical labor, raised $100 million in Series B funding. These valuations are not irrational — they reflect a judgment that robotics is approaching a capability threshold where general-purpose physical automation becomes commercially viable, and that the companies that establish technical leadership now will have durable advantages as the market scales.
In biology, Ginkgo Bioworks' path from a small team of MIT researchers to a publicly traded company valued at over $15 billion represents one of the clearest examples of technical vision translated into market leadership. Recursion Pharmaceuticals is applying computational approaches to drug discovery at scale. Asimov, which raised $35 million to build a platform for designing biological systems the way engineers design electronic circuits, is pursuing a thesis that genetic programming will become a mainstream engineering discipline within a decade.
None of these companies, at seed stage, would have passed a conventional market-sizing exercise. The markets they are creating did not exist at the time of founding. The investors who backed them were not doing market analysis; they were doing technology forecasting — asking whether the scientific results that existed in laboratories would translate into commercial capabilities, and if so, over what timeframe and at what cost.
Why Caution Is the Riskier Strategy in Deep Tech
There is a widespread belief in venture capital that deep technology investing is inherently riskier than software investing because the technical risk is higher. This view is partially correct but fundamentally incomplete. Technical risk — the risk that the technology does not work — is higher in deep tech. But market risk — the risk that the technology works but no one buys it — is much lower. When a team successfully solves a hard technical problem (cheap clean hydrogen production, general-purpose robotic manipulation, reliable AI reasoning), the commercial path is typically straightforward. The demand is established; the difficulty was always on the supply side.
Software-only businesses, by contrast, often solve technically easy problems in highly competitive markets. The marginal cost of creating a new SaaS application has fallen to near-zero, and competition is correspondingly intense. The companies that have built durable software businesses — Salesforce, Workday, Snowflake — have done so through distribution and network effects, not through technical differentiation. Those moats are real but they are built over long periods of time and are increasingly difficult to establish in a market where well-funded competitors can match features within months.
For a fund investing at seed stage, this asymmetry is significant. A deep technology company that successfully solves a hard technical problem often has years of lead time before competitors can replicate its capability stack. A software company that identifies a new market niche typically has months. The apparent safety of backing a proven software market is illusory — the competition that will eventually compress margins is already building, even if it is not yet visible.
Our Framework for Evaluating Breakthrough Bets
At Estes Capital, we have developed an internal evaluation framework for breakthrough technology investments that explicitly attempts to correct for the tendency to underprice audacious bets. We call it the Technology Inflection Assessment, and it has three components.
The first is capability trajectory: does the technology have a credible improvement curve, and are we currently sitting at a point on that curve where commercial applications are approaching viability? This is not a binary question — most important technologies have been "five years away from commercial viability" multiple times before they finally arrive. The question is whether there are recent results that indicate the curve is steepening, and whether the team has the insight to identify where on the curve they are.
The second is market magnitude: if this technology works at scale, what is the total addressable market? For truly transformative technologies, this question often has an astronomically large answer — and that is a signal, not a warning. A technology that could reduce the cost of industrial hydrogen production by 80% would address a $130 billion annual market and unlock applications in steel, ammonia, and fuel that are currently economically infeasible. A technology that enables reliable general-purpose robotic manipulation would address essentially all of physical labor, which is a market measured in tens of trillions of dollars globally.
The third is founder-technology fit: does this specific team have the combination of scientific depth and commercial judgment needed to navigate from early technical results to commercial scale? This is the hardest component to evaluate, and it is where the most important work of seed investing happens. A brilliant scientist who cannot build a company and a skilled operator who cannot understand the science are both bad bets. The rare combination — technically rigorous founders who are also capable of thinking about markets, capital efficiency, and organisational design — is what we are looking for.
The Cost of Missing the Audacious Bet
The most underappreciated risk in deep technology seed investing is not the risk of backing a company that fails. It is the risk of missing a company that succeeds. In a power law world, the cost of the missed investment can be orders of magnitude larger than the cost of the failed one.
An investor who passed on OpenAI's early rounds did not just miss a good return — they missed the defining investment of the decade. An investor who passed on Anthropic because the safety-first AI thesis seemed too speculative did not just miss a modest multiple — they missed a company that went on to raise at an $18 billion valuation in three years. The asymmetry is brutal: failed investments cost you 1—; missed investments can cost you 100— or more in opportunity cost terms.
This is why, when we are evaluating a technology that seems genuinely audacious, we deliberately ask ourselves: "what is the story we would tell in ten years if this technology works and we did not invest?" If that story is uncomfortable enough — if the consequences of being wrong in the direction of excessive caution are serious enough — that is often a signal that the investment deserves closer scrutiny rather than an easy pass.
Venture capital exists to fund the things that cannot be funded by any other means — the bets that are too risky, too long-dated, or too technically uncertain for capital markets that require near-term predictability. The funds that take that mandate seriously, that use their structural position to back the companies that most need patient, conviction-driven capital, are the ones that generate the returns that justify the asset class. The ones that use their structural position to replicate the safety-seeking behavior of public market investors are simply a more expensive, less liquid version of something that already exists.
At Estes Capital, we are here to back the companies that most need what we can provide: domain expertise, long-term conviction, and the willingness to write a check when no one else will. If you are building something that sounds impossible, we want to hear about it. Get in touch.
