Is Non-Consensus Investing Overrated?
Summary
Martín Casado’s actual call is that ignoring consensus is dangerous, not that consensus investing is good. After 10 years and nearly 200 investments, he sees early venture markets as “pretty darn efficient”: being alone may mean genuine insight, but “you may just be missing something.” A company dependent on follow-on capital eventually must become fundable, regardless of how contrarian its first backer felt.
A difficult seed round is not proof that a winner was truly non-consensus, and anecdotes cannot settle the argument. Casado challenges examples such as Anduril when they involve highly accomplished founders, known market signals, or expensive rounds; even a temporarily unpopular company may later raise above-market capital. The right analysis is a basket: compare outcomes for companies with rapid up-rounds, many term sheets, and above-median pricing.
The debate is not simply about finding bargains. Casado’s productive-asset view is that investors often recognize good companies and price them accordingly, while he acknowledges that human perception can also affect outcomes. Torenberg says investors should not seek returns through price arbitrage, and Leo Polovets cites Peter Thiel’s rule that the faster and higher the up-round, the more investors should invest because the company is working. Polovets also recalls missing a company after rejecting a $20 million valuation that he thought should be $10 million—only to watch it reach $10 billion.
The highest-alpha early bets often begin non-consensus but must cross into consensus before their capital needs overwhelm them. Six to eight of Polovets’s roughly ten best investments took months to raise seed rounds, then sometimes jumped 20x or 50x between seed and Series A or B. In deep tech, the wager is whether a $3 million round can hit milestones sufficient to raise $10 million; requiring a $50 million–$100 million next round instead assumes the company will become a top-5% Series A.
Scarce capital can force frugality, while easy capital can create fragility. Casado argues that hard fundraising makes companies more cash-efficient. Torenberg counters that rapid markups premised on perfect execution can create a “house of cards” and says most companies fail from indigestion, not starvation. He identifies the 2021 cohort of billion-dollar Series Bs as a possible major capital wipeout because companies could spend without listening to customers.
Sector enthusiasm is investable only when growth, defensibility, and unit economics survive the narrative. AI has produced genuine growth at OpenAI, Anthropic, and Cursor, with the best companies compressing the old five-year “triple, triple, double, double, double” path into one or two years—but a business reaching $100 million ARR might now fall back to $50 million when a better product appears. Humanoids, autonomous vehicles, and defense illustrate the counter-risk: enormous TAMs and strategic interest can bid prices up before standalone economics are known.
Larger outcomes may justify larger funds and higher prices, but the thesis remains unproven without return-level data. Casado cites Stripe, Databricks, Coinbase, and OpenAI around the $100 billion mark in a16z’s portfolio, while Polovets counters that only perhaps 10–20 companies have crossed that scale over two decades. Their proposed test is decisive: determine whether winners were priced above stage medians and whether most venture profits actually came from those high-priced companies—then distinguish company alpha from mere price arbitrage.
Deep dive
1. Consensus awareness is not consensus investing
Casado’s original distinction: “It’s dangerous to do non-consensus investing” meant that ignoring other investors is dangerous, not that following them is wise. Having completed nearly 200 investments over 10 years, he believes early markets are “a lot more efficient than people realize.”
His academic analogy carries the mechanism: excellent research still failed if its author ignored how the program committee would evaluate the paper. Likewise, a startup dependent on follow-on capital must eventually become legible to the investors funding its next round.
Polovets agrees that “eventually, you have to get to consensus,” but his strongest pre-seed and seed investments often began outside it. Before proof points, the businesses looked doubtful; once evidence arrived, valuations rose so quickly that later investors retained upside, but at far lower multiples.
2. Famous winners do not prove that contrarian rounds outperform
Casado’s objection to calling Anduril non-consensus is definitional. Polovets notes that Palmer Luckey was a second-time founder with a billion-dollar exit, Trae was phenomenal, and the company operated in the shadow of Elon’s defense-tech example. Casado says that calling such a deal non-consensus indicts venture’s insularity; Torenberg recalled its seed as around $100 million, and Polovets said every round was expensive.
Torenberg offered Scale as non-consensus because Alexander Wang was 18 at seed. Casado pushed back that it occupied a known market and involved exceptional investors; Polovets agreed that nearly all of its rounds were competitive. The exchange exposed why “hot versus not hot” or “competitive versus noncompetitive” may be more measurable language than consensus.
3. Hot rounds may contain both information and reflexivity
The discussion rejects treating price as a simple bargain signal. Casado’s productive-asset view is that investors are smart and pay for companies they think are good, while he acknowledges a separate view in which human perception can affect outcomes. Torenberg says investors should not seek returns through price arbitrage; Polovets cites Peter Thiel’s heuristic to invest more when the up-round comes faster and higher because “it’s working.”
Casado proposes testing whether the strongest external correlate of a high up-round is that the previous round was already hot. Polovets says that, if true, it would suggest market efficiency. Casado also asks whether the much larger pool of non-hot companies produces more eventual hot companies than the small set that is already hot.
Both reject single-company storytelling. The useful basket would track companies with rapid follow-ons, ten Series A term sheets, or above-median round prices; even where the operating business disappointed, Polovets has seen investor enthusiasm help preserve a strong outcome, showing that perception can matter independently of productive value.
4. Founders must sell consensus without surrendering product alpha
Torenberg says founder reactions were strikingly consistent: they know they must often be non-consensus in the product market to generate alpha, yet “look consensus” while fundraising. Because another round usually arrives within 18–24 months, celebrating universal rejection can actively damage the company’s next financing.
Casado sees a countervailing benefit in scarcity. Teams that struggle to raise are often more frugal, while hot companies may spend against an assumption of flawless execution; once growth slows, financing disappears abruptly and an operating model built for abundance becomes difficult to unwind.
Torenberg argues that too much capital can prevent founders from hearing the actual market—the customer—and says, “Most companies fail from indigestion, not starvation.” He suspects the 2021 cohort of billion-dollar Series Bs produced one of venture’s largest capital wipeouts.
The market can therefore be efficient on average while failing at both tails. Casado says traditional infrastructure companies that would have been attractive two years earlier can barely raise because they sit outside today’s AI sweet spot; Torenberg adds that some AI companies are receiving speculative funding even when nobody understands the business model.
5. Stage determines how much contrarianism a portfolio can tolerate
Across Polovets’s roughly ten best investments, six, seven, or eight took months to complete a seed and faced extensive rejection. Some failed because skeptics were right, but successful ones later recorded 20x or 50x valuation gaps between seed and Series A or B; the transition from non-consensus to consensus was important, because never making that transition is difficult.
In deep tech, Polovets does not expect a working asset by Series A. He instead asks whether the company can hit technically credible milestones and whether those achievements will be compelling enough for the next investor, explicitly underwriting what that follow-on fund will need to see.
Capital requirements change the wager: raising $3 million to reach milestones that support a $10 million round can be feasible. Planning to use that same $3 million before demanding a $50 million–$100 million Series A requires the startup to become consensus quickly and qualify for a top-5% financing.
Casado’s own company traversed every state: a hot $10 million post-money seed in 2007, no fundability after the 2008 crash, another hot round on “signs of life,” and eventually a $1.2 billion acquisition despite less than $10 million ARR. At its low point, it was perhaps one month from bankruptcy and its original switch-hardware pitch “didn’t make any sense.”
6. AI speed and deep-tech hype pull valuation in opposite directions
Polovets says AI has made the old “triple, triple, double, double, double” journey from $1 million to $100 million look antiquated; leading companies can now do it in one or two years. Yet moats feel weaker: a company can hit $100 million ARR and fall to $50 million after a superior product launches.
Torenberg points to the tremendous growth of OpenAI, Anthropic, and Cursor as underlying market signals beneath the chaos. Polovets, whose portfolio is only about 10%–15% pure AI, remains unsure how to balance unprecedented growth against uncertain endurance.
Deep tech supplies clearer price-cycle examples. Polovets invested heavily in defense three or four years earlier, then paused for a year and a half or two years after the Ukraine and Israel developments drove valuations up 2x–4x without changed fundamentals; a defense company at $40 million then competed against an excellent energy company at $15 million.
Biotech has also cycled repeatedly, while humanoids are one of the most hyped areas, with valuations becoming extreme before revenue provides much grounding. Polovets generally avoids consensus areas where companies have already raised hundreds of millions and a new entrant would start with near-zero resources.
7. Unit economics must survive the giant-TAM story
Casado says a humanoid strategy based on backing several excellent teams and expecting acquisitions is legitimate, but he cannot underwrite it himself. He requires a standalone business at scale, and “competing with a human body is a very, very hard thing to do”; verticalizing into factories also turns the startup into a constrained manufacturing company.
Casado describes the distortion from a roughly $5 trillion human-labor market: at that TAM, almost any seed price can be rationalized. His reductio is cold fusion—calling it the largest market cannot turn laws of physics into an engineering problem for a talented software founder.
Autonomous vehicles reinforce the economics test. After roughly $100 billion of industry investment, Casado characterizes unit economics as approximately on par with Uber: viable for Google or Tesla, difficult for an independent startup, except through acquisition or picks-and-shovels businesses such as Applied Intuition.
By contrast, Casado can understand AI model companies such as ElevenLabs and Midjourney because their unit economics and rapid growth are visible. His objection is transferring those proof points to unrelated sectors where neither the economics nor the technical case has been demonstrated.
8. Outcome expansion changes fund mechanics, not the need for evidence
Torenberg argues that outcomes are now one or two orders of magnitude larger, potentially allowing seed-like returns from Series A or B prices. Casado agrees that getting into the defining company may be the “high-order bit,” while arguing that larger bets require larger funds and access to more LP capital. Polovets adds that a diversified portfolio still needs enough companies.
Casado points to Stripe, Databricks, Coinbase, and OpenAI around the $100 billion mark in a16z’s portfolio, while Polovets estimates perhaps only 10–20 such companies emerged over 20 years. Polovets adds that decacorns are probably an order of magnitude more common than they were 10 years ago, even if $100 billion outcomes remain rare.
SoftBank, Tiger, Coatue, and Insight tested the giant-fund thesis with mixed results; Casado says high prices may not be the sole explanation, citing macro cycles and those firms’ distance from traditional Silicon Valley early-stage networks. Thrive, Founders Fund, and a16z also raised larger vehicles as the opportunity set expanded.
Polovets outlines two viable adaptations: grow the fund 10x, preserve ownership, and let a larger exit return the same share of the vehicle; or make more investments at fractional ownership to increase the odds of catching “the Stripe of the year.” At $100 million per check, however, true non-consensus investing becomes structurally difficult.
A purely consensus market would eventually reduce venture to cost of capital: LPs accepting 2x could outbid those requiring 5x without seeing anything different. Casado values venture’s upside orientation and its role in creative destruction, while the best products remain non-consensus to customers even when sophisticated investors recognize their disruptive potential.
The proposed empirical resolution has two parts: compare winners’ round prices with stage medians, then calculate whether most realized returns came from companies that were consistently high-priced. Polovets supports the test and agrees that investors should not seek price arbitrage; some of his largest misses came from passing at $20 million instead of $10 million before the company reached $10 billion.
Seed remains segmented rather than conquered by multistage firms. Of roughly 10–12 unicorns in Polovets’s portfolio, perhaps one-quarter to one-third had a meaningful Series A investor at seed; repeat founders in familiar markets may command $40 million or $80 million instead of $20 million, but less-obvious companies remain predominantly seed-fund territory.