Anthropic’s IPO prospectus gives investors one of the clearest pictures yet of the economics behind the frontier-AI boom — and the numbers contain both an extraordinary growth story and a major warning about the cost of competing at the cutting edge.
For UK retail investors, the key question is not simply whether Anthropic can become a huge technology company. It is what its flotation says about the sustainability of the wider AI investment cycle, from Nvidia (NASDAQ:NVDA) and Broadcom (NASDAQ:AVGO) to Amazon (NASDAQ:AMZN), Alphabet (NASDAQ:GOOG), data centres, networking, power and the investment trusts already holding Anthropic privately.
A spectacular growth story — but with spectacular costs
Anthropic’s reported 2025 numbers illustrate the extraordinary speed at which frontier AI is scaling.
| Anthropic | 2024 | 2025 |
| Revenue | $386m | $4.59bn |
| Operating loss | $2.98bn | $8.06bn |
| Net loss | $8.31bn | $41.97bn |
| Compute & infrastructure costs | ~$2.5bn | $7.33bn |
| Cash + short-term investments | — | $20.28bn |
Revenue increased almost twelvefold, while the operating loss more than doubled. The headline $42bn net loss is particularly misleading because approximately $34bn related to non-cash accounting adjustments to financing liabilities rather than money consumed by the underlying business.
The more important number for investors is therefore the $8.1bn operating loss.
Compute and infrastructure represented around 58% of operating expenses in 2025, demonstrating the unusual economics of frontier AI: generating revenue requires enormous amounts of computing power before the company can achieve software-like margins.
And Anthropic is not planning to slow investment dramatically.
The prospectus reportedly identifies approximately $518bn of future cloud, computing and infrastructure commitments. This is a multiyear figure rather than an amount Anthropic intends to spend in one year, but it nevertheless illustrates the scale of infrastructure required to support its ambitions.
The bull case: AI could be much bigger than market expects
Anthropic’s central investment thesis is exceptionally ambitious.
The prospectus argues that AI could transform the global economy more profoundly than industrialisation, electricity or the internet. That is effectively the foundation on which investors are being asked to value the company.
The financial forecasts explain why.
Anthropic was reported in August to be targeting approximately $190bn-$200bn of revenue in 2028, compared with 2025 revenue of $4.6bn. Its revenue run-rate had already climbed from around $9bn at the end of 2025 to more than $47bn by May and subsequently reached roughly $65bn by July, according to reports.
That means the investment case depends heavily on future revenue rather than current earnings.
Earlier projections suggested Anthropic could generate its first quarterly operating profit in 2026, with Reuters reporting a forecast $559m operating profit for the second quarter and at least $10.9bn of revenue.
This is the crucial transition investors will be watching:
rapid revenue growth → increasing utilisation → operating leverage → sustainable margins
If Anthropic can demonstrate that progression, the IPO could provide powerful evidence that today’s enormous AI infrastructure spending is ultimately producing highly valuable software businesses.
But there is a second interpretation
The prospectus also provides ammunition for investors questioning the AI boom.
Anthropic needs enormous quantities of chips, data-centre capacity, electricity and cloud infrastructure to generate its revenues.
That creates an uncomfortable question for the entire AI ecosystem:
How much of the revenue being generated by AI ultimately flows back to the infrastructure companies supplying the AI companies?
Anthropic’s $7.33bn 2025 compute and infrastructure expenditure is already greater than its $4.59bn revenue.
And the company says its customers are not necessarily locked into long-term commitments. Around a quarter of 2025 revenue came from just two customers, according to Reuters’ reporting on the prospectus.
That concentration creates another risk: Anthropic’s rapid growth needs to translate into durable enterprise relationships rather than simply enormous consumption of computing capacity.
PitchBook senior analyst Harrison Rolfes highlighted three hurdles: compute and energy requirements, maintaining trust and accuracy, and embedding Anthropic deeply enough into customers’ businesses that switching becomes difficult.
What it means for Nvidia, Broadcom, Amazon and Alphabet
For investors in the AI trade, Anthropic is potentially as important for what it buys as what it sells.
Its planned infrastructure commitments provide evidence that demand for AI compute is not simply a short-term experiment. If Anthropic and its rivals continue scaling models, demand should remain substantial for:
- AI accelerators and networking
- high-bandwidth memory
- advanced semiconductors
- data centre capacity
- electricity and grid infrastructure
- cloud computing
- cooling and power-management equipment.
That supports the structural investment thesis behind companies such as Nvidia and Broadcom and the hyperscalers Amazon and Alphabet.
But the prospectus also raises the bar.
Investors can no longer simply assume that more AI spending = higher returns for every AI supplier.
The market increasingly needs evidence that AI customers can generate sufficient economic value to justify the enormous infrastructure investment.
This distinction is already visible in equity markets. Recent analysis from Investors’ Business Daily found that investors had begun moving away from some capital-intensive AI infrastructure businesses towards companies demonstrating clearer revenue growth and profitability.
The fund manager view
Anthropic’s existing private market investors provide an interesting perspective for UK investors because several investment trusts already provide indirect exposure.
Scottish Mortgage (LON:SMT) manager Tom Slater has described Anthropic as one of the world’s most important AI companies. The trust has argued that its revenue trajectory has been extraordinary, while recognising that forecasting such a rapidly evolving business involves substantial uncertainty.
Scottish Mortgage’s exposure was about 2.9% of the portfolio at the end of August, while Anthropic represented approximately 6.8% of Baillie Gifford US Growth (LON:USA) and 6.1% of Schiehallion (LON:MNTN), according to reports.
That matters because a successful IPO could make Anthropic’s valuation much easier for public-market investors to scrutinise — and potentially create a readily observable market price for a company that has until now been valued through private funding rounds.
The sceptical view
There is also a clear counterargument.
Panmure Liberum has argued that current AI valuations may be assuming the wrong technological path, with smaller language models running locally eventually handling many workloads currently expected to require enormous data centres. It also highlighted the potential vulnerability created by what it described as circular financing across AI companies, chipmakers and hyperscalers.
David Merkel of Aleph Investments similarly questioned whether a $2tn valuation would prove durable, asking whether AI would ultimately generate enough additional productivity to justify the valuations being placed on the industry.
These are particularly important questions because Anthropic’s reported valuation target of more than $2tn would represent more than twice its $965bn valuation following its May funding round.
Anthropic’s unusual approach to risk
There is another important distinction between Anthropic and many conventional technology companies.
How to invest in technology stocks: A UK beginner’s guide 2026
The company describes itself as a safety-focused AI research business and a Public Benefit Corporation. Its stated purpose is to develop AI that is reliable, interpretable and steerable.
Its prospectus reportedly devotes roughly 80 pages of 261 to risk factors, including warnings that increasingly capable models could behave unpredictably or potentially cause catastrophic harm.
Anthropic itself acknowledges the tension between safety and commercial competition. Its filing says that a continuous cadence of model releases is inherent to remaining at the frontier, while safety work is resource-intensive and competes for computing power and talent.
That is a remarkable disclosure for investors: the company believes it must spend aggressively to remain technologically competitive while simultaneously acknowledging that the technology itself creates risks that could undermine its development.
What UK retail investors should watch
The Anthropic IPO could therefore become an important real-world test of the AI investment thesis.
The numbers to watch are less about the headline $2tn valuation and more about:
- Revenue growth — can extraordinary growth continue as the base becomes larger?
- Gross margin — can AI economics increasingly resemble software economics?
- Compute costs per unit of revenue — is efficiency improving?
- Operating cash flow — when does revenue growth translate into cash generation?
- Enterprise retention — are customers embedding Claude deeply into their operations?
- Infrastructure commitments — how much future spending is required to sustain growth?
- Model economics — does greater capability create enough additional revenue to justify higher computing costs?
- Competition — particularly from OpenAI, Google and other frontier-model developers.
Sharesify investor verdict
Anthropic’s prospectus does not provide a simple confirmation or rejection of the AI boom.
Instead, it exposes the central economic debate – an extraordinary growth story and a major warning.
On one side is a company growing revenue at an extraordinary rate and forecasting a potentially enormous future market. On the other is a business that spent more on computing and infrastructure than it generated in revenue in 2025 and is committing hundreds of billions of dollars to future infrastructure.
For UK investors, that makes the IPO potentially more significant than simply another technology listing. Anthropic could become a public-market measuring stick for the entire AI ecosystem.
If investors accept its growth forecasts and believe margins can eventually expand, the listing could reinforce the long-duration AI infrastructure investment thesis. If investors instead focus on capital intensity, customer concentration and the distance between today’s revenue and tomorrow’s valuation, it could force a much tougher reassessment of AI valuations across the market.
As Scottish Mortgage’s Tom Slater has effectively acknowledged, the further investors look into the future, the wider the range of possible outcomes becomes.
For Sharesify readers, that may be the most important lesson from the prospectus: the next phase of the AI trade is likely to be less about proving that AI works and more about proving that the economics work.
Disclaimer: The author Steven Frazer has a personal interest in Scottish Mortgage, Nvidia, Broadcom.
You might also like:
