Anthropic’s IPO Reveals the High Cost of Controlling AI’s Future

Anthropic’s public debut was built around a striking contradiction: the company needed 22 underwriters and more than $500 million in bank fees, yet it tried to keep the people handling the deal and the documents on a tight leash.
The IPO, known internally as Project Lantern, followed a compressed schedule filled with late changes, delayed financial projections and rushed document approvals. Anthropic capped some deal teams at four senior bankers, asked certain bank officers to accept six-month restrictions on work with AI rivals, and shared sensitive materials through a system designed to track every page they viewed.
The completed listing gives investors access to the company behind Claude, an enterprise-focused large language model similar to OpenAI’s ChatGPT, Meta AI and Google’s Gemini. It also offers a close look at how much money Anthropic needed to fund its plans, how much control it wanted to keep, and how expensive an AI IPO can become.
A tightly controlled offering
The four-banker limit was designed to prevent leaks, according to a person familiar with the matter, though banks found ways to work around it. The restrictions created a difficult trade-off for the financial firms involved: they were weighing more than $500 million in fees against less control over staffing, documents and the way their teams handled the transaction.
Anthropic also sought legal documents covering core transaction-team members and naming OpenAI, Cursor, DeepSeek and Hugging Face. Bankers and lawyers routinely manage conflicts and protect client confidentiality, but broad exclusivity rules are rare in an IPO when they bind individual bankers.
One banker who declined to sign said it was difficult to find staff who could promise not to cover or work with Anthropic’s peers. Several of those companies were pursuing their own initial or secondary offerings, making the restriction harder to accept.
Anthropic needed a large group of underwriters because of the capital it sought. That made the arrangement different from smaller head-to-head offerings, where a bank might choose one competing company and avoid the other.
The company also moved its documents away from ordinary PDF files. Most underwriting banks received materials through NetRoadshow, which recorded who opened each document and which pages they viewed, placed individual bankers’ names as watermarks and blocked users from saving copies.
That system complicated bankers’ work, but leaks still reached the press. The caution followed an earlier disclosure of details about Anthropic’s finances, risk factors and dealings with other technology companies nearly four weeks before the IPO registration statement became public.
Late documents and an expensive fundraising drive
The document controls arrived alongside a late approval cycle. Before the planned November 10, 2026 marketing launch, a law firm sent banks a PDF about Canadian share sales, then told them to disregard it because it contained an incorrect price range and share information. The corrected version of the updated registration statement reached banks through NetRoadshow at about 2 a.m. New York time, leaving only several hours for approval before filing with the Securities and Exchange Commi.
The financial incentive remained hard for banks to ignore. Coverage put the total bank fees at more than $500 million, while the syndicate led by Goldman Sachs was set to receive $550 million. Anthropic paid underwriters a fee of 0.676%, compared with 0.666% for SpaceX’s IPO.
Anthropic’s headline gross proceeds from shares sold to the public reached $81.34 billion, against $75 billion for SpaceX. The company also followed SpaceX by paying no fees on the 15% over-allotment shares reserved for underwriters to sell at the offering price when demand exceeds the initial supply.
The capital target was enormous. Anthropic was set to raise $90 billion for the development of cutting-edge AI, while estimates placed its market capitalization as high as $1.4 trillion. That figure fell short of the $2 trillion valuation some estimates had floated a few weeks earlier, but it still would have made Anthropic one of the largest stock market debuts on record.
The company also completed a private placement selling $26 billion of nonvoting N shares to a group including Alphabet, Amazon and Nvidia. Those shares convert to the regular A voting category. Investors paid 17.1% less than the market judged the shares were worth at the end of the day, creating an extra $4.4 billion that did not go to Anthropic’s treasury.
Investors saw a pop, but Anthropic left money behind
Anthropic shares rose from the offering price of $83 to $97.17 at the close, a 17.1% jump. Another account put the first-day rise at 17%, followed by an 8.5% drop at the close on November 20, 2026, as a broader reversal in AI trading cut short the opening rally.
That early jump became central to the debate over pricing. IPO investors often hope for a “pop,” but a sharp first-day rise can also mean the company sold shares too cheaply. IPO specialist Jay Ritter said Anthropic left roughly $20 billion on the table, money that could have gone toward AI infrastructure but instead went to large institutions and privileged retail investors.
Ritter calculated that the amount left behind was $14.337 billion from the public offering, only $80 million below the comparable figure for SpaceX. SpaceX shares jumped 19.2% on opening day and collected $14.417 billion less than they might have raised at the price investors were willing to pay.
Anthropic’s unusual share structure added another discount. Employees and investors held a large amount of stock without any lockup period, meaning those shares could be sold immediately at the market price. Selling another 15% of the shares at a 17.1% discount to the 4 p.m. price on November 19 created an additional $2 billion discount.
“When you include everything, it’s the largest number left on the table in the history of IPOs,” Ritter said.
Claude’s promise and the risks ahead
Anthropic’s main product, Claude, is built for enterprise customers. Depending on the package, it can draft emails, legal documents and contracts, generate code, analyze data and create detailed reports. Supporters say that speed can make businesses far more efficient, while critics warn that many human responsibilities and jobs could disappear.
Morningstar analyst Ahmed Khan said AI had become a cultural phenomenon rather than a subject limited to technology or finance specialists. He also warned that training larger language models requires expensive near-term decisions with uncertain outcomes.
“It is going to be a foundational shift in how work gets done,” Khan said. He described Anthropic’s strategy as the opposite of ChatGPT’s consumer focus, saying the company “went to the enterprise,” where customers will pay for contracts that address liability, uptime requirements, service-level agreements and security.
That business focus does not remove the financial risk. Philip Peterson, chief investment strategist at IG Wealth Management, placed Anthropic in the speculative category because it is only about five years old, uses a new technology and is still trying to work out how to make money from it.
Anthropic also remains cash-burning and has not found a path to profitability. Its seven co-founders collectively control a “god share” through a separate entity, giving them 50.1 per cent of the voting power. A simple majority of the seven founders decides how that share votes, allowing them to resist a proposal demanding that the company stop training large models and focus on profitability.
CEO and co-founder Dario Amodei summed up the company’s IPO attitude while speaking with JPMorgan Chase chief executive Jamie Dimon: “underpromise and overdeliver.” The public market will now test whether Anthropic can turn that promise into durable revenue while paying for the vast cost of building its models.
Based on
- Anthropic Sought Limits on IPO Deal Teams and Work with Rivals — unite.ai
- What makes the Anthropic IPO so significant, and what is the ‘god share’? – National | Globalnews.ca — globalnews.ca
- How big is Anthropic’s record setting IPO — cnbc.com
- Anthropic’s IPO: How do you price the future of AI? | AFP.com — origin-zh9p7m-www.afp.com
- Anthropic IPO left a record $20 billion on the table, says IPO expert Jay Ritter | Fortune — fortune.com




