Hardly anyone talks about Instinct anymore.

Few people are talking about Instinct anymore.

Just a month ago, this AI startup founded by 23-year-old founder Noah Shinn (with the main entity named Spear Street Technology) was the hottest rising star in Silicon Valley. It just completed its Series B financing at a valuation of 2.5 billion US dollars at the end of August, and immediately after that, on September 28, it announced the completion of a 1 billion US dollar Series C financing, with its valuation soaring directly to 10 billion US dollars. Top-tier institutions including HSG, Benchmark and Coatue all participated in the round.

In just one month, its valuation quadrupled.

However, after a short period of viral popularity, discussions about it are visibly fading away. Instead, there are news of Microsoft and Meta adjusting their capital expenditures, rumors that Anthropic has lowered its expected IPO valuation from about 2 trillion US dollars to about 1.5 trillion US dollars, and even the argument that "Apple has become an AI winner because it has never really participated in the large model competition".

The situation of Instinct is like a punctured slice, which accurately reflects the profound shift that is taking place in the AI industry right now.

14 People, 100,000 Users and a Valuation of 10 Billion US Dollars

If we want to select the most extreme bubble sample in the current AI venture capital circle, Instinct is definitely on the list.

What exactly has it achieved? According to public information, it is a product billed as a "personal AI butler". It has no independent app, not even an official public launch version, and is still in the early invitation-only internal beta stage, with the number of users just exceeding 100,000. Users interact with this AI agent via text messages or phone calls, asking it to help reply to emails, book flights, and buy groceries. The founding team has only 14 people.

But such an extremely early-stage product has propped up a valuation of 10 billion US dollars. This means that the valuation corresponding to each internal beta user is as high as 100,000 US dollars. Moreover, the founder has clearly stated that he does not want to charge users at this stage.

What supports this sky-high valuation is the "category option" and investors' "fear of missing out (FOMO)". Under the grand narrative that AI Agent is regarded as the next-generation computing platform, capital is unwilling to miss any project that may become a super entry point. Benchmark entered in the Series B round at a valuation of 2.5 billion US dollars, and the valuation rose to 10 billion in the Series C round one month later, with the book return quadrupling instantly. This "two-tier financing" game driven by leading funds essentially creates value rather than discovers it.

This is the most widely accepted doctrine in Silicon Valley — a variant of Founder's Instinct in the AI era. Investors do not need to see mature products, nor do they need to verify the revenue model. They only need to believe in the founder's resume, the popularity of the track, and the fanaticism of "I think this can work".

From "Telling Stories" to "Calculating Accounts"

However, the macro environment no longer pays for pure "instinct" and "stories".

In mid-October, Anthropic delivered a seemingly impressive result at its investor day: revenue in the third quarter reached 16.8 billion US dollars, and it achieved profitability after adjustment. Even so, investors are still disappointed, because its month-on-month growth rate has slowed down from more than doubling in the previous quarter compared with the earlier one to a range of about 45%-50%. Subsequently, the market's expected valuation slipped from about 2 trillion US dollars to the range of about 1.5 trillion US dollars.

Microsoft and Meta are also sending similar signals. Microsoft has adjusted its capital expenditure expectation for the 2026 calendar year from about 1900 billion US dollars to about 1750 billion US dollars. This change mainly stems from the accounting estimate change of the depreciation period of data centers — more future leases are converted from finance leases to operating leases and are no longer included in the capital expenditure caliber. At the same time, independent of this accounting adjustment, previous channel surveys claimed that Microsoft had canceled data center leases with a capacity of "hundreds of megawatts" in the United States, but Microsoft's official did not comment on the survey and publicly stated that its infrastructure investment plan is still proceeding as scheduled. As for Meta, its 2026 capital expenditure guidance remains at the 1250-1450 billion US dollar level, and it launched layoffs involving about 8,000 positions in the first half of this year to hedge the financial pressure of AI infrastructure.

The actions of these giants and leading model companies point to the same signal: the focus of AI narrative is shifting from the fanaticism of "who dares to spend money and who dares to shout high valuations" to the rationality of "who can recover the money spent".

At this stage, the first to be rewarded are no longer the companies that shout the largest investment figures, nor the start-ups with only 14 people and 100,000 users but a valuation of tens of billions. The market begins to favor companies that can draw a clear path between input and return again. Even companies like Apple, which do not participate in the underlying computing power arms race at all and only do end-user application distribution, have been re-priced by the capital market as "AI winners" precisely because they do not need to pay for expensive computing power.

Silicon Valley Has Lost Its "Instinct"

A dozen years ago, Silicon Valley was a place that trusted instinct extremely. Steve Jobs cut off the physical keyboard by instinct, and Mark Zuckerberg wrote Facebook in his dormitory by instinct.

But in today's large model era, this instinct-dependent entrepreneurial narrative has failed. The current AI game is an infrastructure movement that relies heavily on heavy assets, computing power monopoly and engineering capabilities. When the trial and error cost of one model training is as high as tens of millions of US dollars, and when tech giants spend hundreds of billions of US dollars on GPU procurement at every turn, investors' patience threshold has undergone fundamental changes.

The reason why Instinct's popularity has plummeted is precisely because it is stuck in the fault of this logic. It tells a very attractive story: an AI butler without an app, whose annualized GMV has reached about 1 billion US dollars. But when the story runs far faster than the product, the valuation becomes a castle in the air. When the capital environment tightens, those start-ups with no actual hematopoietic capacity naturally become the first batch of objects to be re-examined.

People no longer discuss Instinct frequently, not because AI is no longer viable. On the contrary, it is because AI is becoming more and more pragmatic. The industry has passed the wild stage where you can get a ten-billion-dollar valuation just with the resumes of a few big names and a grand vision.

After the bubble recedes, the next competition is for engineering capabilities, reasoning cost control, and the implementation ability to integrate models into specific business scenarios. The frantic era belonging to "instinct" is coming to an end, and now everyone has to speak with products and financial reports.

This article is from the WeChat official account "Tech Business", author: Tech Business, published with authorization from 36Kr.