The herd mentality financing the future and its hidden costs
Three quarters of the venture capital raised in the last year went to five companies. Not five sectors. Not five categories. Five companies. That figure, stated without euphemisms by Niko Bonatsos of Verdict Capital at a recent TechCrunch panel in Athens, summarizes with more precision than any market report what is happening in global venture capital: an unprecedented concentration that coexists, paradoxically, with a discourse of distributed innovation and open opportunity.
The event brought together Bonatsos alongside Andreas Stavropoulos of Threshold Ventures and Ben Blume of Atomico to discuss the state of venture capital, the imminent wave of major IPOs led by SpaceX, and where they see real space for opportunity. What they left behind was less a map of the future and more an honest X-ray of a market that mixes genuine signals with distortions that no one quite wants to name.
The question that organizes all of this is not whether artificial intelligence is going to change the economy. That debate is already closed. The operative question is how much of the capital flowing into that sector today is buying real value and how much is buying a position in a narrative that still cannot sustain itself on revenues alone.
When liquidity creates the illusion of a market
SpaceX is approaching an IPO with a reported valuation of $1.75 trillion. Stavropoulos compares it to Google's IPO in 2004, which reactivated markets that had lost confidence in technology following the dot-com cycle. The argument is solid in its structure: large exits generate returns that flow back into the ecosystem as new capital, and that capital opens doors for the next generation of founders.
Blume adds that SpaceX is such a singular company that its public access could capture the imagination and investment of segments that historically did not participate in private technology. Space as an investment domain open to the general market is, effectively, a category shift.
But there is a tension that none of the three fully resolves. Blume himself names it: a portion of the capital that goes to SpaceX consists of funds that would have gone to the next twenty or thirty software businesses. That is not neutral. In a market where early access to the right capital can separate a company that survives from one that does not, the reorientation of those allocations has real consequences for what gets built and what does not.
The dominant narrative says that liquidity generates more liquidity, that the cycle is virtuous, and that the returns from a major exit fertilize the next cycle. That is historically true as a trend, but it conceals a lag. Between the IPO of a company valued at $1.75 trillion and the moment that capital returns to a seed fund that finances a 23-year-old founder in Buenos Aires or Mexico City, there are years. And in those years, the distribution of capital is not homogeneous: it goes to the same managers, in the same markets, with the same selection biases.
The most uncomfortable question about SpaceX is not whether it will affect market liquidity in the short term. It is whether an exit of that size will concentrate institutional attention even further on benchmark assets and reduce the relative appetite for what is difficult to categorize — which is exactly where Bonatsos says the opportunities with low valuations are to be found.
The price of capital when everyone wants the same thing
Bonatsos describes Verdict Capital's strategy with a word that rarely appears in the language of venture capital: "freaks." Founders who advance in a single day what the average person would accomplish in a week, who build in markets that do not yet have a name, and whose valuations are low precisely because large asset managers cannot give their teams the mandate to look for companies in categories that do not yet exist.
It is a strategy of first-money-in on unmapped territory. What makes it viable is not only the investment thesis, but the structure of competition: funds of ten or fifteen billion dollars cannot operate efficiently in that space. Blume, who manages a five-hundred-million-dollar fund, already feels the pressure of competing with those vehicles in artificial intelligence rounds, where the incremental value of a dollar to a small fund versus a large one is radically different. That distorts the size of rounds and makes it almost impossible to compare offers on equivalent terms.
What is happening at the application layer of artificial intelligence is an accelerated version of what occurred in mobile between 2009 and 2013: too much capital chasing too many similar theses, with a very small subset of companies capturing the majority of returns. Stavropoulos anticipates this with more honesty than evasion: there is going to be a correction. The promise and the optimism are significantly ahead of the capacity to show results in the short and medium term. That does not cancel the long-term argument, but it does imply that many of the current valuations are not anchored in sustainable revenues but in expectations that still have no delivery date.
What creates noise from a commercial analysis perspective is the combination of three factors that the panel describes with varying degrees of discomfort: unprecedented capital concentration, selection biases based on age and profile as substitutes for business signals, and revenue metrics defined in increasingly creative ways.
Bonatsos says it plainly: when there is a lot of money chasing specific themes, some people develop a short-term mentality that prioritizes appearance over substance. He receives emails from companies in his portfolio with annualized revenue figures that turn out to be 365 times what they billed on a good day following a campaign. The solution he proposes is to use minimum quarterly baselines. But the underlying problem is not the calculation method: it is that there is a market willing to finance those figures without asking the questions that should be asked.
Where capital does not reach and why that matters
The most interesting space in the analysis is not in what everyone is financing, but in what no one is looking at. Bonatsos points out that venture capital has practically abandoned the digital consumer: where previously half the partners at a fund worked in consumer internet, today barely half a partner is dedicated to that area. The argument is that ChatGPT, the most widely adopted consumer product of recent years, came from an artificial intelligence company. The consumer did not disappear: what disappeared was capital's interest.
That creates an asymmetry. If there are five investors available to finance a founder in consumer versus fifty to finance one in language model infrastructure, the price competition in the first case is lower. Entry valuations are more reasonable. The market is less efficient. For a fund with discipline in entry pricing, that is a structural advantage, not a concession.
Blume sees the greatest opportunity at the intersection of artificial intelligence and the physical world. Not the humanoid robot performing pirouettes in a demonstration video, but the penetration of automation into the sectors that still move the largest share of global gross output: manufacturing, logistics, construction, agriculture. The proportion of economic value that still depends on non-digitized physical processes is enormous. The software infrastructure for those sectors is at an early stage compared to what was built over the last twenty years for purely digital processes.
That thesis has an advantage over those competing at the core of the language model market: it does not require winning against OpenAI or Anthropic. It requires understanding, in sufficient detail, the physical processes of a specific industry in order to make automation work under real conditions, with real variability, with real workers. That friction is also the barrier to entry. What makes that category difficult to attack is the same thing that makes it difficult to replicate once it works.
The market that finances itself needs an external buyer
The commercial architecture of the artificial intelligence boom has a structural problem that the panel orbits without fully landing on. A disproportionate fraction of the capital entering the sector comes from funds that also hold positions in the infrastructure on which those startups run. Spending on computing goes to the same providers whose venture capital funds or corporate investment arms are financing the rounds. That is not necessarily fraudulent, but it creates a circularity that inflates activity metrics without there being a net external buyer validating the value.
A business sustains itself when someone who has no financial incentives in the chain decides to pay for the product with money they could have spent on something else. That is what Stavropoulos calls "the capacity to show results." And that is precisely what is lagging behind the optimism of the valuations.
The cycle of major IPOs can inject liquidity back into the market. But the question of whether the companies being built with that capital have external buyers with a genuine willingness to pay prices that justify entry valuations still lacks a clear answer. Until that answer arrives in the form of verifiable revenues with real margins, the correction that Stavropoulos anticipates is not a possible scenario. It is an adjustment pending only in timing.
