Anthropic appears to push its IPO toward November 2026

What the numbers say

Anthropic initially targeted October 2026. The offering is now expected no earlier than late October, with November considered more likely. Advisers say the delay is meant to give investors a stronger third-quarter result.

That rationale is not entirely convincing. If the second quarter is already strong enough for an IPO, a better third quarter could support the stock after listing rather than simply justify the listing itself. Investors expect Anthropic to reach a valuation of about $2 trillion and raise up to $100 billion—both figures would exceed records set by SpaceX after its IPO in June.

According to The Information, Anthropic’s revenue more than doubled between the first and second quarters of 2026, while infrastructure spending increased by 65%:

The company’s so-called adjusted operating margin rose from negative 13% in the first quarter to a single-digit positive figure in the second. That measure, however, comes with important qualifications.

Growth is creating a funding problem

Anthropic’s enterprise business is expanding quickly. By the end of 2025, about 1,500 companies had spent more than $100,000 each on its software during the previous 12 months. By the end of the second quarter of 2026, that number had risen to roughly 6,000.

Over the same 12-month period:

Anthropic expects roughly $190–200 billion in revenue in 2028, according to The Information. That forecast may help explain the scale of the expected offering, but it does not mean the money will flow directly to investors. An IPO would primarily give Anthropic continued access to debt and equity markets.

The company said in early August that it held $120–130 billion in cash. Even so, it will probably need additional financing as data-center and model-training costs consume its reserves. The computing infrastructure agreement with SpaceX, signed in May, costs Anthropic $1.25 billion per month alone.

My read is that the IPO is less a finish line than a financing mechanism for an unusually expensive business. Strong revenue growth makes the story investable; it does not settle whether the underlying economics will work once the cost of producing that growth is fully counted.

Competition is making the timing harder

Several forces could be pushing Anthropic toward a later offering.

Anthropic recently claimed profitability, but excluded major costs such as stock-based compensation and did not use standard accounting rules in the calculation. That gives investors a number to debate before they even reach the broader question of whether the company can fund its expansion.

Meanwhile, OpenAI regained some of its advantage with Astra and overtook Anthropic on the OpenRouter AI model platform. Both companies need to maintain rapid growth in the coming months to justify their large-scale data-center expansion. Higher interest rates make those facilities more expensive to build, while cheaper open models add further competitive pressure.

The other risk is harder to put into a financial model. During testing, the cybersecurity capabilities of models considered dangerously powerful led to “unintended hacks” in trials involving OpenAI, Google, Anthropic and Meta.

The incidents have not been treated as attacks or linked to sanctions and lawsuits. They have mostly been viewed as a “new development in AI.” That response may be inadequate once the companies are public and subject to greater scrutiny. The operational risk is already visible, while insurance does not yet cover the resulting damage.

OpenAI has also moved its IPO to 2027. Sam Altman explained the decision by citing security risks and said a company building AGI has reasons to remain private during a period of sweeping change. Given the other pressures facing both companies, that explanation accounts for only part of the delay.

The question I would want answered before either listing is not whether revenue is growing. It is whether AI spending is producing more revenue, or simply replacing the cost of having people do the same work. Those outcomes remain difficult to measure, and a public listing will make that ambiguity harder—not easier—to ignore.