Anthropic’s bulls and bears
Well, if you didn’t know it already, Anthropic, the startup behind the Claude chatbot ecosystem, filed a Form S-1 with the Securities and Exchange Commission this week, officially launching what may be the most-watched initial public offering (IPO) in history.
The company hopes to raise roughly US$100 billion and target a valuation between US$1.5 trillion and US$2 trillion. In so doing, it has sparked intense debate among investors, analysts and portfolio managers.
For the bulls, Anthropic’s listing proves the multi-billion-dollar artificial intelligence (AI) boom is real, highly profitable, and set to transform the world economy in ways larger than industrialisation, electricity, semiconductors, or the internet.
For sceptics, the S-1 IPO documents reveal a flimsy financial picture built on massive multi-year compute debt, huge non-cash accounting losses, complex circular deals with cloud partners who are also direct rivals, and a strange corporate structure that deprives public shareholders of normal voting rights.
Massive growth
The company’s revenue has grown staggeringly over the last three years, rising from US$40 million in 2023 to US$4.6 billion in 2025.
And from that US$4.6 billion last year, first half 2026 revenue hit US$16.3 billion.
For perspective, the second quarter of 2026 alone generated US$11.6 billion in revenue, a 140 per cent sequential jump over the previous quarter.
Importantly, that single quarter allowed Anthropic to post its first-ever quarterly operating profit of US$974 million, translating to an eight per cent operating margin.
Gross margin also surged, reaching roughly 85 per cent on paper in the first half of 2026, up from just 12 per cent a year earlier and negative 95 per cent in 2024.
Compute efficiency explains that operational turnaround. By that I mean that every dollar spent on computing infrastructure yielded US$2.08 in revenue by mid-2026, up from US$0.40 per dollar in early 2025. That represents a five-point-two-fold jump in efficiency over five quarters.
At the same time, customer expansion metrics look unprecedented. Anthropic reported a net revenue retention rate of roughly nine hundred per cent, meaning existing enterprise accounts spent nine times more this year than they did last year. Corporate customer adoption has surged alongside it, as the number of clients spending over US$100,000 annually exploded – there isn’t a better word to describe it – from 950 to nearly 12,000 in twelve months.
According to Anthropic’s S-1, there are roughly 250 customers spending US$10 million or more annually, and a tier of 1,900 spending over US$1 million.
When you look at workforce efficiency, the numbers are enviable. Trailing twelve-month revenue per employee shot up to US$6.6 million, compared to US$1.4 million a year earlier, even as total headcount expanded from 940 employees at the end of 2024 to about 5,300 today. That kind of productivity dwarfs standard software industry benchmarks, where $450,000 per employee is considered very healthy.
The bulls argue these figures prove frontier AI models are turning into an indispensable economic engine, one where unit costs drop rapidly for users while overall platform usage scales exponentially.
The sceptics
When critics read the same financial statements, they see a more alarming picture.
For example, while Anthropic did indeed grow its base of clients spending over US$100,000 to nearly 12,000, Ed Zitron – whose views we discuss below – reveals that 80 per cent of OpenAI and Anthropic’s enterprise revenues originate from just one per cent of their customer base. In other words, the vast majority of the cash driving Anthropic’s growth comes from a very small number of “whale” clients spending tens or hundreds of millions.
Zitron also notes these high-rolling, top-tier enterprise clients aren’t traditional Fortune 500 companies (like banks or retailers) using AI for standard business operations. Instead, he says they’re VC-backed AI startups, wrapper apps, and developer platforms (e.g., companies like Anysphere/Cursor) that burn through hundreds of millions of dollars in VC funding specifically to buy Anthropic API tokens so they can fuel their own software.
But the big number the bears are focused on is the headline net loss of US$80.7 billion for the first half of 2026.
The company’s defenders correctly point out that US$70.4 billion of that loss stems from non-cash fair-value accounting adjustments on convertible notes as the company’s private valuation climbed. However, cash-flow pressures remain enormous.
Operating cash flow for the first half was US$3.9 billion, but the S-1 notes a successful IPO will immediately trigger US$2.9 billion in stock compensation expense, with another US$23.7 billion to be recognised over the next two years. That means roughly US$11 billion in annual non-cash stock compensation will weigh on an operating profit of around US$4 billion, pulling GAAP earnings back into the red.
Beyond the stock compensation, sceptics also worry about the hundreds of billions of dollars tied up in long-term compute commitments running through 2045. While Anthropic reported over US$16 billion in revenue for the first half of 2026 and narrowed its operating loss, it also disclosed well over US$500 billion in contractual commitments for data centres and computing infrastructure.
Procuring compute capacity means committing capital 12 to 18 months before data centres come online, locking Anthropic into multi-year commitments, including 20-year leases, with partners like Nscale, Lambda, AMD, Microsoft, and SpaceXAI.
Consider what happens if enterprise software adoption slows down, or if competitors drive model pricing down faster than usage expands. In that scenario, Anthropic could find itself stuck paying massive fixed infrastructure bills on stranded capacity – a white elephant.
Furthermore, there’s the frequently reported web of counterparty risk. Nearly half of Anthropic’s first-half revenue flowed to cloud cluster providers like Amazon Web Services, Google Cloud, and Microsoft Azure. These companies also account for 58 per cent of Anthropic’s receivables. In other words, Anthropic relies heavily on strategic partners who are also as its lenders, equity holders, and direct rivals selling competing models on those very same clouds.
Governance
Anthropic’s IPO has also sparked debate about corporate control, regulatory exposure, and public governance.
Supporters adore Anthropic’s public benefit corporation setup and its explicit approach to safety. By keeping strategic decisions under the guidance of a five-member trustee board drawn from national security, public policy, and AI safety circles, it helps ensure AI technology is managed responsibly.
In their view, this unconventional governance structure frees the leadership team from quarterly stock market pressures and lets them focus on long-term R&D bets, such as curing diseases or solving complex scientific challenges.
From an ordinary investor’s perspective, however, the same governance setup looks like a total loss of traditional shareholder rights.
Under Anthropic’s multi-class share structure, public Class A buyers are capped at 49.9 per cent of total voting power. Meanwhile, a single Class F share held by an entity tied to the seven co-founders retains fifty-one per cent voting control in perpetuity, while a single Class T share held by the Long-Term Benefit Trust selects the majority of the board of directors.
The setup removes traditional investor protections, leaving individual investors with virtually zero ability to influence board seats, guide strategy, or vote on change-of-control deals.
The short sellers
Adding to the debate, prominent tech critic Ed Zitron and ‘The Big Short’ investor Michael Burry have emerged as two of the most vocal voices warning that Anthropic’s IPO is precarious at best.
Zitron reckons the company’s S-1 filing is evidence of a fundamentally broken business model, pointing out that in 2025 Anthropic spent two dollars and seventy-five cents for every single dollar of revenue it generated – worse even than OpenAI’s burn rate at the time – while racking up massive net losses and locking itself into US$400-US$500 billion in total compute and cloud lease obligations.
He argues that the company’s reliance on adjusted profitability ignores real-world training and inference costs, and he warns investors that a massive share of its revenue depends on a circular ecosystem where smaller AI startups trade money within the same tech bubble.
Michael Burry is also sounding the siren. He insists the official prospectus changes nothing about his fundamental bear case for the artificial intelligence trade.
While frontier models like Claude are improving rapidly and operating more efficiently, technological brilliance doesn’t solve the underlying financial and leverage risks, and Anthropic’s disclosures reveal massive multi-year infrastructure commitments that could destroy capital if token pricing commoditises or if Anthropic’s cloud suppliers, who are also its big tech rivals, shift entirely to their own proprietary models.
Both Zitron and Burry believe the rush to list on the Nasdaq is ultimately an attempt to unload heavy financial risks onto everyday retail investors before the broader AI infrastructure boom hits a wall.
The test
The bull case fundamentally rests on the revenue efficiency stats. Recall, net revenue retention is about 900 per cent, meaning customers are soending nine times as much this year as they were last year. When Snowflake listed its revenue efficiency was 158 per cent. When GitLab filed its revenue efficiency was at 152 per cent. At the time, investors considered those stats spectacular.
It’s worth noting GitLab rose about eight per cent after its IPO before its shares collapsed 78 per cent from its high in November 2021 to May 2023. Snowflake investors experienced similar volatility after its IPO in late 2020. Its shares initially rose 19 per cent, then fell 47 per cent, before rising 90 per cent into November 2021 and subsequently falling 72 per cent into September 2024.
The bulls note, Anthropic is growing at more than 10x annually with 900 per cent net revenue retention and is priced at between 20x to 30x its revenue run rate, which is less than what CrowdStrike and Cloudflare are priced at on forward revenue with a fraction of the growth.
The bears insist Anthropic investors are paying something like 95 times trailing GAAP revenue, more than Palantir, for a company that buys compute from Amazon, Google, and Microsoft, who sell competing models on the same clouds that resell Claude.
For what it’s worth, I think it all hinges on my number one question: whether use cases keep increasing as fast as the price per task keeps falling.
If nothing else, Anthropic’s IPO will test whether massive growth outweighs fundamental long-term risks and circular financing.
Roger Montgomery is the Founder and Chairman of Montgomery Investment Management. Roger has over three decades of experience in funds management and related activities, including equities analysis, equity and derivatives strategy, trading and stockbroking. Prior to establishing Montgomery, Roger held positions at Ord Minnett Jardine Fleming, BT (Australia) Limited and Merrill Lynch.
He is also author of best-selling investment guide-book for the stock market, Value.able – how to value the best stocks and buy them for less than they are worth.
Roger appears regularly on television and radio, and in the press, including ABC radio and TV, The Australian and Ausbiz. View upcoming media appearances.
This post was contributed by a representative of Montgomery Investment Management Pty Limited (AFSL No. 354564). The principal purpose of this post is to provide factual information and not provide financial product advice. Additionally, the information provided is not intended to provide any recommendation or opinion about any financial product. Any commentary and statements of opinion however may contain general advice only that is prepared without taking into account your personal objectives, financial circumstances or needs. Because of this, before acting on any of the information provided, you should always consider its appropriateness in light of your personal objectives, financial circumstances and needs and should consider seeking independent advice from a financial advisor if necessary before making any decisions. This post specifically excludes personal advice.
