Key Takeaways

  • Institutional consensus places the AI cycle between pre-game warmups and the second inning, with zero managers tracked by Adams Street Partners viewing it past the fifth inning.
  • Historical precedent from the 1993 Netscape browser launch suggests the enduring corporate giants of a tech wave appear four to seven years after the opening spark.
  • Hardware and infrastructure supply constraints continue to dictate early-cycle economics, as raw enterprise demand for compute outstrips actual capacity.
  • Adams Street's historical data across venture portfolios reveals that just 7% of backed startups generate 100% of the industry's net capital returns.
  • Even frontier model architects like Sam Altman at OpenAI and Dario Amodei at Anthropic cannot predict the downstream commercial forms their platforms will take.

Tech Waves Take Five Years to Reveal Their Real Winners

Every major computing shift starts with an undeniable proof of life. For the web, that spark was Netscape in 1993. For modern artificial intelligence, it was the fourth quarter of 2022, when OpenAI released ChatGPT to the public.

Yet early validation rarely creates the market's biggest financial winners. Brijesh Jeevarathnam, Partner and Global Head of Fund Investments at Adams Street Partners, points out that the real commercial anchors of the internet did not arrive alongside the first browser: “When there's a seminal moment in a particular technology, and maybe the best example is Netscape, the first kind of real public browser, I think it was 93, early '90s, there were many companies that were started in the internet 1.0 era, but the best ones came arguably kind of 4 to sevenish years out.” (play from 1:00:30)

Amazon took years to find its stride, and Google did not incorporate until late 1998, five full years after Netscape demonstrated what consumer web browsing looked like. Applying that same clock to ChatGPT puts the current cycle squarely in the second inning. The market today is busy financing base infrastructure layers and experimental application wrappers. The enduring businesses, the ones built on top of mature, reliable rails, likely have not raised their Series A yet.

Compute Shortages Trap Capital in Hardware Instead of Software

In the opening innings of a platform cycle, capital naturally aggregates around physical limits. Today, that limit is raw silicon and data center power.

“At this point in time, the demand for compute seems to be far outstripping the supply into the foreseeable future,” (play from 1:03:30) Jeevarathnam notes. “And from a broad use case standpoint, it appears to be still very early in the AI wave.” (play from 1:03:40)

Because computing demand outpaces supply, capital flows upward into silicon, cloud capacity, and foundation models, rather than dispersing across a defensible software application layer. The operators creating the technology cannot even pinpoint where the end-state margins will settle. As Jeevarathnam observes: “I like to sometimes go back and watch a presentation or a new product release from OpenAI or Anthropic and see what those very smart people, Sam Altman, Dario say. And what I find interesting is with the passage of time even they cannot quite foresee where this is all going.” (play from 59:38)

When model creators like Altman and Amodei cannot chart the downstream roadmap, underwriting early-stage application software requires extreme caution. The venture universe has grown from 500 firms to more than 4,000, all chasing exposure. Yet Adams Street's proprietary dataset shows that only 7% of venture-backed companies produce all net returns. Chasing first-generation application tools built on shifting model capabilities exposes allocators to outsized obsolescence risk before the foundation stabilizes.

Why It Matters

When platform cycles take four to seven years to produce their defining winners, aggressive early deployment into thin software layers creates stranded portfolio companies. The current phase favors infrastructure providers that monetize raw compute demand immediately, while late-stage venture and private equity will capture the true application winners only after baseline model capabilities and pricing settle into stable utilities.