Doesn’t anyone know how to foresee the value of a tech company anymore?

The latest shock came from Oracle, the ancient—by Silicon Valley standards—database software giant. Its shares surged 36%, the most since 1992, after the company signed “significant cloud contracts with the who’s-who of AI.” Total remaining performance obligations, or contracted revenue it expects to collect, reached $455 billion, more than four times higher than a year earlier. It was a big shock even though the mega-cap is covered by 47 brokers, who presumably know it well.

This outsized earnings surprise is no doubt a win for Chairman Larry Ellison. His company has gone through the 1990s dotcom cycle, when its stock price fell by over 80% after that bubble burst in 2000. While Oracle’s share price has recovered decades later, it has lost a lot of its shine. No longer part of the Magnificent Seven, it has not been seen as a major beneficiary to the AI boom.

Unfortunately, Oracle’s drastic melt-up also raises the uncomfortable question of whether sell-side analysts know how to value tech companies in the AI era. After all, the improvement of the technology’s capabilities can be non-linear, and investment professionals need to incorporate this aspect in their models without making their forecasts look outlandish.

On the earnings call, Ellison focused on ‘AI inference,’ the process of using a pre-trained model to generate content for its existing database clients. The new applications, which make it easy for companies to learn and gain insights from their private data, have a promising future. Analysts might be aware of its market potential but can be unwilling to measure them in a meaningful way. After all, in the US, the economy-wide firm AI adoption rate is only 9.7%, according to Goldman Sachs.

Mapping out convincing future cash flows is made all the more complex by the fact that the AI world’s biggest players are private and thus don’t need to disclose their capital expenditures—unless they are fundraising.

OpenAI, for instance, reportedly projected it will burn $115 billion through 2029, about $80 billion higher than its previous estimates. But how much of that incremental spending will go to Oracle is anyone’s guess. All we know is, directionally, there will be more demand for cloud services, which Ellison’s company provides.

These are valid concerns, but one can see the danger of analysts persistently underestimating a company’s earnings power. Over time, they will lose their credibility and investors will simply discard the valuation multiples — the stock market’s most important anchor. What’s the point of looking at forward price-earning ratios if money managers believe analysts have no clue and are always late to the game? There was a sense of that during the dot-com boom.

Silicon Valley’s powerful venture capitalists are certainly more than happy to tell you valuation doesn’t matter when a disruptive technology is still at an evolutionary stage, because they can’t justify their investments otherwise. Here’s an example: Investors including SoftBank, Thrive Capital and Dragoneer are purchasing OpenAI shares at a $500 billion valuation, nearly twice the price they were paying about six months ago. Why the dramatic change?

In the past, investment analysts would use discounted cash flows or earning multiples to get a sense of whether a stock is over or underpriced. But as we enter a new world order and a revolutionary technology is upon us, how do we project the known unknowns?

For decades, academia has talked about incorporating options pricing models in enterprise valuations, essentially embedding the value of a call option to capture the blue-sky scenario. Perhaps it’s time that sell-side analysts heed their advice. ©Bloomberg

The author is a Bloomberg Opinion columnist covering Asian markets.