Instinct Raises $1B and Anthropic Delays IPO as AI-Agent Markets Diverge
Instinct raised $1B at $10B on September 28, days after Anthropic reportedly pushed its $2T IPO to November for a clean Q3. The private anchor remains Sierra AI's May $950M Series E at $15.8B.
September 28, 2026

Two capital-markets events a few days apart show a gap between private and public investors. Instinct raised $1B at a $10B valuation from Sequoia, Benchmark, and Coatue. By September 25, Anthropic's rumored $2 trillion IPO had reportedly been pushed from October to November. Sierra AI's May Series E at $15.8B, on $950M raised, is the private-market benchmark both are measured against. Private investors are valuing AI-agent companies at multiples the last cycle would have refused, while the public window has slipped by a month. The reported reason is timing: an October listing would come before Anthropic could put audited Q3 numbers in the prospectus, and November lets a clean quarter do the talking. We'd add, as our own read rather than the reporting, that the safety headlines since Sierra priced make that clean quarter worth more than usual.
The New York Times framed the delay bluntly, asking whether safety risks could derail the sector's IPO prospects. Whatever the timing reason, that's a question Anthropic's bankers still have to answer, because it's the one mutual-fund managers face when asked why a $2T AI IPO belongs in a pension plan's balance sheet. Sierra and Instinct don't need to answer it. Their money comes from venture investors like Coatue and Sequoia, whose conversations with their own investors are different. Anthropic does need the answer, because its exit has to survive public disclosure.
Private and public markets price the same risk differently
Private money didn't slow when Anthropic's IPO did: Instinct's round closed days later. That's because private and public investors are asking different questions about the same kind of company. The private question is "will this company reach $500M in ARR in the next three years?" Sierra's 2026 customer list makes a yes plausible enough for a $950M check. The public question is "if this company reaches $500M in ARR, could a class action or a regulatory action cost more than that?" This question carries a different discount rate, which is how the same kind of asset can be valued at $15.8B in the private window and at a discount publicly.
Instinct’s round is the clearest signal. A $1B raise at $10B for a consumer personal-agent company from three of the most disciplined funds in the sector shows the fundraising market concentrating on companies with a defensible position in a category with perhaps three names left worth funding at scale. Every other AI-agent company raising a Series C in 2027 will quote Sierra and Instinct comps in its deck, and the money will show up if the numbers support it.
The counter-case worth taking seriously
The clean bear view is that private investors are misreading the same safety signal the IPO window is pricing in. Sierra will need a public exit within about three years to return its investors’ money, and every AI safety incident between now and then shrinks the exit multiple. On that view, investors should price today's private multiples as if the public discount is coming, with Anthropic's IPO delay as a possible leading indicator and Sierra's Series E as the lagging one.
That view carries real weight, but it assumes the safety risk only moves in one direction. This week, Nvidia shipped guardrail software and enterprise procurement started writing safety into RFPs. Anthropic's constitutional AI approach also gives it policy language that fits a mutual-fund allocation memo better than the pure-model-training vocabulary. The market isn't sure the safety discount will hold once the products start answering the safety question, and private investors are effectively buying an option on that happening. That’s a bet with a payoff structure the public window can't match, and it may well be a reasonable one.
What this means for founders and CROs
First, if you're raising, your comparables now start with Sierra and Instinct, and last quarter's rounds count for less. Bring the numbers that make a yes to the private question plausible: contracted ARR, expansion cohorts, and gross-margin trajectory. The Sierra deal closed because its numbers were easy to follow in a 30-minute meeting.
Second, if you sell to enterprise buyers, expect your customers to cut their lists of AI vendors. Sierra can win a bake-off against three smaller vendors on procurement risk alone, because a $15.8B company passes the "will it still be around in 2029?" test that a $400M vendor may not. Factor that into your competitive positioning this quarter.
Third, if you're close to a public exit yourself, get ahead of the safety-disclosure conversation before an investor raises it. The S-1 language that works on a November roadshow will differ from what worked six months ago. Anthropic's reported delay, timed so audited Q3 numbers can go in the prospectus, also previews the disclosure standard that every AI-agent IPO is likely to face for the next 18 months.
Reuters first reported on a leaked draft of Anthropic's prospectus on September 28, and the draft gives about 80 of its 261 pages to risk factors, including a warning that advanced AI could pose “catastrophic or existential risks to humanity.” The same draft shows a 2025 net loss of $42 billion, more than $8 billion of it from operations, on revenue that grew 12-fold to nearly $4.6 billion. Those revenue figures also anchor our look at why Cursor, Attio, and Linear still charge per seat for AI. For a founder planning a roadshow, Anthropic's risk language is now the standard an AI-agent S-1 gets read against.
Private and public markets will eventually agree on a price for AI-agent risk. Until they do, the gap between them is the opportunity.
Sources
- Sierra AI closes a $950M Series E at a $15.8B valuation
- Instinct raises $1B at a $10B valuation
- Anthropic reportedly pushes its $2T IPO to November
For a founder planning a roadshow, Anthropic's risk language is now the standard an AI-agent S-1 gets read against.
If you're raising, bring contracted ARR, expansion cohorts, and gross-margin trajectory, because your comparables now start with Sierra and Instinct.



