Why the value of the AI platform shift is accruing before the IPO

Why the value of the AI platform shift is accruing before the IPO

Every platform shift has a moment when the public market "discovers" it. For the internet it was Netscape in 1995. For mobile, arguably Facebook's listing in 2012. For AI, the moment arrived in the spring and summer of 2026, when Cerebras, SpaceX and a queue of confidential filings from the largest model labs turned a trickle of listings into a wave.

It is tempting to read that wave as the start of something. We read it the other way. By the time a company of this generation rings the bell, much of the compounding has already happened, and it happened in a market most investors cannot see.

The company you buy at IPO is older than it used to be

The structural fact behind this essay is simple and well documented. The median company going public in the United States is now roughly twice as old as it was a generation ago. Jay Ritter's long-running dataset at the University of Florida shows the median age at IPO rising from about six years in 1980 to the low teens in recent years, with 2025 listings clustering around twelve to thirteen years from founding (Ritter, University of Florida IPO data, 2025; CNBC, 2025-10-07). Median revenue at IPO has risen accordingly, from tens of millions of dollars in inflation-adjusted terms to more than two hundred million (CNBC, 2025-10-07).

Ritter's explanation is the right one: companies used to go public to raise capital, and today "there are a lot of good alternatives" (CNBC, 2025-10-07). Late-stage private rounds, structured secondaries and employee tender offers let a company fund growth and give staff liquidity without the costs of public reporting. Stripe is the canonical example. In February 2026 it ran another tender offer for current and former employees, at a valuation of $159 billion, up from $106.7 billion a year earlier (CNBC, 2026-02-24; Axios, 2026-02-24). A company of that size once had little choice but to list. Now it can choose.

The consequence is a transfer of where growth is captured. When a company lists at year four, public shareholders own most of its growth curve. When it lists at year thirteen, they own the back half.

The AI generation compresses the curve further

The AI platform shift intensifies this pattern for a reason that is specific to the technology: frontier AI is capital-intensive in a way software was not. Training runs, inference capacity and custom silicon require capital on a scale that only the largest private rounds can supply, and private capital has proved willing.

The trajectory of Anthropic illustrates the point. Reported valuations moved from roughly $170 billion in mid-2025 (TechCrunch, 2025-07-29), to a $380 billion post-money round in February 2026 (Crunchbase News, 2026-02), to a $65 billion raise at a reported valuation near $965 billion in May 2026 (TechCrunch, 2026-05-28). The company confidentially submitted a draft registration statement in June 2026 (GraniteShares research, 2026). Whatever happens after listing, a meaningful share of that company's value creation was priced while it was private.

Cerebras is a smaller but cleaner case study, because we can see both ends. Its final private round in September 2025 was priced at an $8.1 billion post-money valuation (Cerebras press release, 2025-09-30). Eight months later it priced its IPO at $185 a share, above the marketed range, raising $5.55 billion (Cerebras press release, 2026-05-13) in what CNBC described as the largest US tech IPO since Uber (CNBC, 2026-05-13). The listing implied a market capitalization of roughly $67 billion (SiliconANGLE, 2026-05-13).

We are not arguing that every private AI company will follow that path. Most will not. We are arguing that for the ones that do, the steepest part of the curve now sits before the listing date.

What the 2026 wave actually revealed

Renaissance Capital and others have described the second quarter of 2026 as the strongest quarter for US IPO proceeds on record, driven by SpaceX's listing (IG, 2026-05-20; Yahoo Finance, 2026). OpenAI confidentially filed in June as well (IG, 2026-05-20). Commentators have noted that a handful of these listings could raise more than the entire US IPO market did across several prior years combined (TechJournal, 2026).

Read as a signal, the wave says three things.

First, public investors are willing to pay for AI infrastructure and model companies at scale. That validates the categories.

Second, the companies arriving are already enormous. There is no small-cap version of this cycle for public investors to buy early.

Third, the investors who shaped cap tables in 2022 and 2023, when AI infrastructure was a contrarian call and secondary sellers were plentiful, were positioned in a way that simply cannot be replicated in a public order book in 2026.

What this means for allocation

For an accredited investor, the practical question is not whether AI matters. That is settled. It is where in a company's life you are able to participate, and on what terms.

A few principles follow from the data.

Public exposure alone now captures a later slice of the curve. An index or a portfolio of recent listings gives you the mature phase of these businesses. That is not a bad thing to own, but it is a different asset.

Private exposure is about access, not appetite. The best late-stage private rounds are allocated, not sold. Secondary blocks in category leaders move through relationships with employees, early investors and company counsel. Wanting in is not a strategy.

Selection matters more than exposure. Private markets are more dispersed than public ones. For every Cerebras there are infrastructure companies that raised at peak valuations and will not grow into them. A manager's edge is in knowing which is which before the market does.

Liquidity is the price of admission. Capital committed to private companies can be locked up for years, and the IPO itself is often followed by lock-up periods. Investors should size positions for that.

Where CrossWork sits

CrossWork was built on the premise described here. We became a Cerebras shareholder in 2023, before its 2026 listing, and our current holdings include Stripe, Perplexity, Neo4j, SambaNova, DataRobot and Scale AI, with access through an advisory board of former operators, who tend to see which companies are quietly winning inside their customers well before a round is announced. CrossWork Midas II is our current vehicle.

The 2026 IPO wave has been described as a window opening. For many of the companies in it, it was closer to a window closing on the private phase of their growth. The next group of category leaders is still private, and the same pattern is repeating now.

---

Sources

• Jay R. Ritter, "Initial Public Offerings: Median Age of IPOs Through 2025," University of Florida (2025).
• CNBC, "Startups staying private longer with alternative capital" (2025-10-07).
• CNBC, "Stripe valued at $159 billion after tender offer" (2026-02-24); Axios Pro (2026-02-24).
• TechCrunch, "Anthropic reportedly nears $170B valuation" (2025-07-29); "Anthropic raises $65 billion, nears $1T valuation ahead of IPO" (2026-05-28).
• Crunchbase News, "Anthropic raises $30B at $380B valuation" (2026-02).
• GraniteShares, "Anthropic IPO 2026 explained" (2026).
• Cerebras, Series G press release (2025-09-30); IPO pricing press release (2026-05-13).
• CNBC, "Cerebras prices IPO above expected range" (2026-05-13).
• SiliconANGLE, "Cerebras prices shares at $185" (2026-05-13).
• IG, "SpaceX, OpenAI, Anthropic: upcoming IPOs to watch in 2026" (2026-05-20); Yahoo Finance, "Why the 2026 IPO boom is about to broaden" (2026); TechJournal (2026).

Past performance is not indicative of future results. This is not an offer; offering documents govern. Detailed performance is available to verified investors in the CrossWork portal. This is not investment advice.