Data Is the Next $1 Trillion Market

The venture secondary market has matured significantly, with Stripe and SpaceX running regular tender offers that effectively create a public market for private companies. Secondary funds are unlikely to produce five to ten times returns but have consistently delivered two to two and a half to three times returns over the past ten to fifteen years. SPV stacking creates provenance risk because investors often skip the diligence needed to verify that each layer of a structure is legitimate, and fund managers sometimes use SPVs tied to desirable assets as an enticement to get LPs to commit to their main fund. The LP universe is increasingly bifurcated, with a small basket of roughly 6 to 20 companies including Anthropic, OpenAI, Anduril, and Databricks dominating hit lists, and the coming wave of distributions from those names will create a pronounced have and have-nots dynamic among LPs.

Nikhil Basu Trivedi of Footwork argues that energy, compute, and data are the three fundamental substrates for AI models and that market attention cycles through each in turn. Unlike energy and compute, there are no large pure-play public companies focused solely on data, which he sees as potential white space. Protege has emerged as the leader in facilitating licensing deals between data providers and AI companies, is in its second year of business, and is already generating hundreds of millions in revenue. Public deals like Reddit and the New York Times licensing to OpenAI represent a small visible fraction of a much larger volume of such agreements, and proprietary data is where the speakers believe economic value will ultimately accrue in the AI stack.

Public markets are described as systematically undervaluing companies that hold proprietary data. Salesforce is growing at 10 to 12 percent but is down 40 percent for the year despite holding extensive proprietary CRM data. Box is trading at 3.4 times trailing price-to-sales. Intercom rebranded early around AI, recovered from stagnant and shrinking ARR, and ultimately sold for approximately 3.3 to 3.4 billion dollars, offered as evidence that incumbent SaaS companies can leverage their data and customer relationships rather than be destroyed by AI. The MeritEx index for enterprise software trades at approximately 3.6 to 3.7 times revenue, and HubSpot has over three billion dollars in ARR yet is valued at under ten billion dollars in public markets, while some private companies are valued at ten billion dollars or more with minimal revenue.

M&A activity is accelerating. Crunchbase data shows a steep ascent in total dollar value of global venture-backed M&A from a nadir in late 2023 and early 2024, though the total number of exits has not changed significantly. Cursor was acquired by SpaceX for approximately 60 billion dollars. Databricks, valued at approximately 150 billion dollars, has sufficient currency to acquire companies with limited dilution. The re-rating of enterprise software and fear of AI displacement are contributing to lower valuations and increased willingness among legacy SaaS companies to sell.

The AI infrastructure economy is described as circular, with the warning that if any part of the financing or capital-raising chain stops, growth estimates collapse and a serious correction follows. Nikhil says the key signal he is watching is any decline in the compute crunch at the hyperscaler level, and that one major hyperscaler pulling back CapEx by at least 10 percent would put significant pressure on the upcoming earnings cycle. Financing risk on large projects such as the Oracle and OpenAI Star Cluster is cited as a potential catalyst, with bond premiums on those projects already elevated. A Nasdaq decline of 20 percent is described as easily possible if growth assumptions fail to materialize, with the timeline estimated at 12 to 24 months. Nikhil says he worries more about bottom-line performance than top-line growth for companies consuming tremendous capital, with OpenAI high on that list.

Two months before recording, many companies shifted from subscription AI pricing to usage-based pricing. Cursor used Kimi K2 and Airbnb used Qwen as examples of companies adopting open-weight Chinese models. One speaker notes that Nvidia Nebatron and Google Gemma may be insufficient replacements for DeepSeek, Kimi, Moonshot, and Qwen for startups seeking open alternatives, and that startups without AI talent have no option but to use closed-source models or work directly with frontier labs.

The speakers observe that the number of companies, founders, and researchers that matter in AI appears to be concentrating even as it has never been easier to build. They warn that getting into YC and becoming a Thiel Fellow are increasingly treated as credentials rather than genuine signals of founder intent, and that tourist founders chasing easy conditions will not last because building companies is genuinely hard regardless of market conditions. Alpha in venture is expected to come from backing non-consensus founders on non-consensus companies rather than late-stage consensus deals in names like SpaceX, OpenAI, and Anthropic, which the speakers describe as resembling IPO investing under a venture capital label.