A Billion in Headlines, Fifty Million in Reality
There is an image worth more than any subsequent analysis: David Silver, one of the most respected researchers in reinforcement learning, connected to a video call with a venture capital firm, no presentation, no supporting document, describing an artificial intelligence system that would eventually learn to interact with toasters. An investor present at that meeting recalled it as "absurd." Another said they left with more questions than answers.
Weeks later, headlines announced that Ineffable Intelligence had raised $1.1 billion in the largest seed round in European history, with a valuation of $5.1 billion. A company with no product, no revenue, and a business thesis that its own blog describes as carrying a significant risk of failure in exchange for an opportunity for spectacular success.
Both things are true at the same time. The round existed. The headlines appeared. And the structure behind both was designed to make it so.
The Mechanics That Turn Fifty Million Into Five Billion
Ineffable's round was not a block of $1.1 billion delivered under a single valuation. It was a two-part sequence. In the first, Sequoia Capital and other investors put in $11 million at a pre-money valuation of approximately $55 million. A few weeks later, a second group of investors arrived, including Lightspeed, Index Ventures, and DST Global, who contributed an additional $1.1 billion at a pre-money valuation of $4 billion.
The difference between the two tranches is more than 70 times the valuation of the same company, at the same stage of development, just weeks apart. There was no new product. There were no revenues to justify the leap. Instead, there was a financial logic entirely consistent with the incentives of each party involved.
Sequoia entered first, at a discounted price, with a larger ownership stake. When investors in the second tranche paid 70 times more for the same company, the firm that entered first obtained a paper valuation increase before a single dollar of operating expenditure had left the account. That markup is not accidental: it is the designed reward for whoever assumes the risk of entering first with conviction in a company with no traction.
The problem is not that Sequoia paid less. The problem lies in what happens with the number that circulates externally.
The real combined valuation of Ineffable Intelligence — the weighted average that considers the proportion of equity ceded in each tranche — sits somewhere between $55 million and $4 billion, much closer to the lower end if the majority of the external capital came from the second tranche but the greater dilution corresponded to the first. That number never appeared in any headline. Sarah Catanzaro, partner at Amplify, described it bluntly: the combined valuation "is usually not reported." What is reported is the highest figure, the one from the last tranche, because that is the one that generates coverage.
Jaya Gupta, partner at Foundation Capital, was even more direct with Forbes: "In a market where financing flows on narrative momentum, a billion-dollar headline is worth far more than a precise one." That sentence is not a criticism of the system. It is an operational description of how it works.
What the Headline Does for the Company That the Data Still Cannot Do
Understanding why this works requires thinking about who reads those headlines and what decision they make based on them.
The first group affected are senior-level candidates. An engineer with experience in foundational models has multiple offers on the table. Cash compensation is comparable across companies. What differentiates one offer from another is the stock option package and the implied valuation of those options. If the company courting them can say it was valued at $5.1 billion by Sequoia, Lightspeed, and Index in its seed round, that figure enters directly into the calculation of how much the options they are receiving are worth.
What that candidate does not know — unless they conduct active due diligence — is that the exercise price of their options will likely be set close to the valuation of the last tranche, not the combined valuation. Gupta framed it precisely: "They are taking more risk and capturing less potential gain. The social contract that made equity participation in startups attractive is eroding, and most candidates don't discover it until a liquidity event occurs, or doesn't."
The second group are follow-on investors. A fund that did not manage to get into OpenAI or Anthropic at early stages is under institutional pressure not to miss the next wave. When it sees that Sequoia, Lightspeed, and Index already have a position in a company, the signal it reads is one of validation. What it does not see, unless it has access to registration documents, is that Sequoia bought in at $55 million and that the company it is considering financing in a Series A round will be starting from a base valuation built on the highest tranche, not the average.
The third group are sovereign wealth funds and public development entities. In the case of Ineffable Intelligence, the UK Artificial Intelligence Sovereign Fund and the British Business Bank, both backed by British taxpayer money, participated in the round. It is not known at what valuation they invested. Neither institution responded to Forbes' inquiries. The fact that public capital enters a company with no revenue, no product, and an opaque valuation structure — driven by industrial policy motivations rather than financial return — introduces an element of systemic risk that goes beyond private investors.
Why the Market Tolerates Structures It Would Not Tolerate in Another Context
Brendan Foody, CEO of Mercor, published a description on X of the pattern he observed across multiple recent rounds, using the term "Sequoia Scam" to describe the practice of investing in two tranches at radically different valuations and then presenting only the higher figure. Hours later he added that "in fairness, this is common practice across the industry at all major firms."
Shaun Maguire, partner at Sequoia, responded on the same platform that the description was unfair, that this type of structure had occurred about five times in his seven years at the firm, and that the dynamic is explained by the fact that "other investors are willing to pay a high price for a hot company, multiples above what we are willing to pay."
That public exchange captures the tension precisely. The practice is not illegal. There is no evidence that it is coordinated to deliberately deceive. It is the result of a market in which demand for exposure to frontier artificial intelligence far exceeds the supply of companies that can offer it, creating conditions where investors who arrive late are willing to pay any price to get in.
Zach DeWitt, partner at Wing VC, described the pressure from the inside: "At some point there will be pushback from funds, but the market is so hot right now that you have no other choice if you want exposure to the best companies."
That sentence deserves a pause. "You have no other choice" is precisely the psychology that sustains valuations disconnected from operational reality. When the alternative to paying $4 billion for a company with no revenue is having no position in frontier artificial intelligence at all, many funds choose to pay. That dynamic feeds itself: every round that closes at record valuations raises the floor for the next one.
Deedy Das, partner at Menlo Ventures, quantified the scale of the phenomenon in May: 63 neolabs with aggregate valuations exceeding $300 billion and approximately $48 billion raised. That figure represents around 16% of the $283 billion that flowed toward startups other than OpenAI and Anthropic over the past year. Frontier artificial intelligence capital is not a corner of the market. It is a structural part of the current investment cycle.
The Number Nobody Publishes Is the Only One That Matters in the End
The weighted combined valuation is the data point that precisely describes how much a company is worth when several investors paid different prices for the same asset. It is the number that a buyer in an acquisition or a bank in a public offering process will use as a reference to build its own price thesis. It is not the number that appears in headlines, nor the one used in presentations to senior-level candidates, nor the one that circulates in conversations between funds.
This is not merely a problem of informational transparency. It is a problem of market signaling. When Baseten, an artificial intelligence infrastructure company, raises $1.5 billion in two tranches valued at $11 billion and $13 billion respectively, the number that circulates is the one from the second tranche. Founders competing with Baseten for talent negotiate against that figure. Investors evaluating the sector use that number as a reference for their own valuation models.
The problem is not that some artificial intelligence companies are overvalued at some stage of their financing. The problem is that the price formation mechanism has become decoupled from any operational anchor. In sectors where there is proven traction, recurring revenue, or at least a sustained usage metric, valuations have a floor. In the case of the neolabs, the floor is narrative: the quality of the founding team, the boldness of the vision, and the ability to generate headlines that attract the next tranche.
Ineffable Intelligence has the advantage of having a founder whose technical credibility in reinforcement learning is genuine. David Silver built part of the research that today underpins many of the most advanced model architectures. That is not in dispute. What is in dispute is whether that credibility — with no product, no revenue, and a round structure designed to maximize the headline rather than the accuracy — is sufficient to sustain a $5.1 billion valuation under any test other than another financing round.
The company's blog answers that question with unusual honesty: there will be a significant risk of failure in exchange for a possibility of spectacular success. The investors who entered in the second tranche bought exactly that. The employees who join after reading the $5.1 billion headline bought something different, without necessarily knowing it.
The Pattern Reveals a Market That Rewards Structured Opacity
The practice of multiple tranches at different valuations will not disappear as long as the appetite for exposure to frontier artificial intelligence exceeds the supply of companies that can offer it credibly. What can change is the ability of secondary participants — candidates, follow-on investors, and public entities — to read the real structure behind the headline.
The number that does not appear in coverage is always more informative than the one that does. The valuation of the first tranche of Ineffable, $55 million, describes the price paid by the investor with the most information, the greatest diligence, and the strongest incentive to be precise. The valuation of the second tranche, $4 billion, describes the price paid by investors with less information, greater competitive pressure, and greater willingness to pay for access.
That gap of 70 times is not an indicator of how much the company grew in a matter of weeks. It is an indicator of how much the scarcity signal is worth in a market where missing the deal matters more than understanding the deal. As long as that asymmetry exists, headlines will continue to be the primary product of many artificial intelligence financing rounds, and the combined valuation will continue to be the number that nobody publishes because nobody has an incentive to do so.
