- Anthropic’s regulatory filing is its first step to joining fellow Artificial Intelligence (AI) developer SpaceX on the US stock market
- The prospectus shows that Anthropic – like Grok – is still losing a lot of money and consuming cash, thanks to its hefty research and investment plans
- Spending on AI at the hyperscalers continues to grow all the same and that is providing opportunities to those who are on the receiving end of the capital expenditure splurge
- This covers everything from silicon chips to energy providers to earthmoving equipment
- The ultimate question is when – or whether – the AI labs will generate a profit and cashflow sufficient to justify their massive early-stage investment
“Anthropic’s regulatory filing in the USA formally fires the starting gun on the company’s initial public offering (IPO) and its path to becoming a publicly owned firm whose shares trade on a US stock exchange,” says AJ Bell investment director Russ Mould.
“This deal comes hot on the heels of SpaceX’s listing in June, when the sale of $75 billion in shares put a valuation of $1.75 trillion on Elon Musk’s rockets-to-AI-to-satellites firm. Anthropic seems determined to come with an even bigger price tag, but investors need to ask themselves when – or whether – the company can generate enough profits and cash flow to justify what would be the seventh-largest stock market capitalisation of any company in the world, based on current valuations.
Source: Marketscreener, LSEG Refinitiv data, consensus analysts’ forecasts
“The spending by the large language model (LLM) developers and AI labs is clearly filtering down to a whole host of companies through a growing and complex food chain, that stretches from:
- firms such as OpenAI and Anthropic through to data centre and cloud computing giants Microsoft, Amazon and Alphabet (who are also developing their own AI and LLM offerings), as well as Oracle;
Source: LSEG Refinitiv data
- to fledgling so-called neocloud providers of AI infrastructure and data centres such as Coreweave, Nebius and Iren;
Source: LSEG Refinitiv data
- to the providers of construction equipment for the building of the data centres, such as Caterpillar;
Source: LSEG Refinitiv data
- to experts in areas such as water and power supply, cooling systems and connectivity to ensure the data centres run as efficiently as possible, such as Vertiv, Eaton, Schneider, Siemens, XP Power and Computacenter;
Source: LSEG Refinitiv data
- to baseload power providers in new areas such as small modular nuclear reactors that could ease the strain on existing energy supplies and grids, including Oklo and NuScale, or the UK’s Rolls-Royce;
Source: LSEG Refinitiv data
- to flash memory and data storage drives and cards, as made by SanDisk, Seagate, Western Digital;
Source: LSEG Refinitiv data
- to a range of silicon chip makers including, graphics processing unit (GPU) designer NVIDIA and its key foundry manufacturing partner TSMC, customer accelerator (XPU) expert Broadcom, central processing unit (CPU) makers Intel and AMD or memory chip giants Micron, Samsung Electronics and SK Hynix;
Source: LSEG Refinitiv data
- right along to semiconductor production equipment (SPE) specialists such as ASML, Applied Materials, KLA and Tokyo Electron at the very back end.
Source: LSEG Refinitiv data
“China has its own, rival AI developers, such as DeepSeek, Moonshot AI and Alibaba, who can draw on their own food chain of suppliers and whose DeepSeek, Kimi and Qwen models are looking to challenge OpenAI’s ChatGPT, Anthropic’s Claude, Meta’s Muse, SpaceX’s Grok, Microsoft’s Copilot and Amazon’s Nova.
“The spending arms race between the major AI labs has gone up several gears since OpenAI launched ChatGPT in November 2022 and DeepSeek launched R1 in January 2025.
Source: Company accounts, Marketscreener, analysts’ consensus forecasts. *Alphabet, Amazon.com, Meta Platforms, Microsoft and Oracle
“As a result, David Roche of Quantum Strategy & Geonomics asserts that US companies alone have invested $3.1 trillion in AI since 2013, including off balance-sheet commitments such as leases on data centres, a figure that, in his words, ‘is bigger than the combined cost of the Vietnam War, the Interstate Highway System, the Apollo program, the Marshall Plan, and the eradication of polio.’
“Moreover, the AI labs and hyperscalers are committed to a further $3 trillion in spending by 2030.
“All of that cash is flooding through the food chain, to the particular benefit of the enablers of AI and the providers of modern-day picks and shovels, rather in the same way that it was the sellers of mining equipment who did rather better than the prospectors themselves during legendary American gold rushes of the nineteenth century.
“This can be seen in the booming profits generated by the memory providers, silicon chip makers and SPE producers in particular, and also how well their share prices have done, especially in the USA.
“The list of the twenty best performing stocks in 2026 in the S&P 500 index to date is dominated by these firms.
Source: Marketscreener, LSEG Refinitiv data
“The Philadelphia Semiconductor Index, or SOX, a thirty-stock basket of leading silicon chip designers and makers and SPE manufacturers has also gone into orbit.
Source: LSEG Refinitiv data
“The benchmark has been buoyed by rapid revenue and profits growth across the industry, where annual sales could hit $1.5 trillion for the first time ever in 2026, according to leading consultants.
Source: WSTS, SIA, Gartner
“The power of AI spending can be seen in the UK equity market too, given the lofty positions in the FTSE 350’s share price performance rankings of XP Power, Raspberry Pi, Computacenter and Renishaw. Computacenter currently tops the FTSE 100 this year, while XP Power and Raspberry Pi rank in the top three for the FTSE 250.
Source: Marketscreener, LSEG Refinitiv data
“By contrast not one of the Magnificent Seven of Alphabet, Amazon, Apple, Meta Platforms, Microsoft, NVIDIA and Tesla features in the list of the top performers in the USA.
“Only three of those seven have managed to outperform the S&P 500 in 2026 to date and Oracle’s shares have been the worst of the lot. Its newly issued bonds are faltering, too.
Source: LSEG Refinitiv data
“Oracle’s woes relate to the substantial amount of borrowing it is using to fund data centre construction, while the Mag7 seems to be weighed down by concerns over competition from China, whether stories of ‘rogue’ AI activity will put off potential consumer and corporate users and how the hyperscalers’ massive expenditure is to be funded. Even some of the Mag7 are now either raising equity, debt or both as their previously prodigious cash flow wilts in the face of their substantial up-front spending commitments.
Source: Company accounts for Alphabet, Amazon, Apple, Meta Platforms, Microsoft, NVIDIA and Tesla
“Not one of Microsoft, Alphabet, Amazon or Meta do a particularly good job at disclosing how much money they are making (or losing) as they develop their AI offerings, although Alphabet, Amazon, Tesla, Microsoft and NVIDIA are booking capital gains on their equity investments in OpenAI, SpaceX and other developers as funding rounds boost their respective valuations.
“SpaceX does disclose how much Grok is losing. In the first six months of 2026, SpaceX’s AI business generated $1.5 billion in sales and $2.5 billion in operating losses.
“Reports suggest Anthropic’s S1 filing for 2025 also reveals a sea of red ink, in the form of an $8 billion operating loss on $4.6 billion of sales, according to unofficial reports on, and readings of, the document.
“Anthropic’s sales have rocketed since then, judging by management’s statements that July’s annualised run rate (ARR) for sales was $65 billion, but ARR is not recognised under generally accepted accounting principles (GAAP) and the company’s claims it made a profit in the second quarter of 2026 appear to rely on adjusted earnings which exclude several key lines of costs, rather than GAAP standards.
“As such, the further away companies are from the actual spending on AI and LLM development, and the closer they are to receiving that money, the better they seem to be doing, at least for now.
“For the developers and labs to justify their stock market valuations, actual in the case of SpaceX or putative in the case of Anthropic and OpenAI, they will need to start to show a return of some kind on their investment at some stage.
“The question is how patient investors are prepared to be and what may trigger either a fresh round of enthusiasm for LLM developers – and thus perhaps even higher valuations – or a first wave of disappointment, which could have the opposite effect, and perhaps do so right across the food chain from top to bottom.
“It may be that LLMs are just one facet of AI, too, as companies develop their own small language models (SLMs), to preserve their data integrity and maintain their proprietary edge, while areas such as energy, healthcare and materials also offer huge potential for AI-fired research to deliver astonishing results in areas yet to be considered from an investment point of view.
“After all, anyone looking at the stock market landscape in 2000, just as the technology, media and telecoms bubble burst, would have had a hard time looking beyond Cisco, Intel, Nokia, Lucent and Nortel as the most likely future winners. By the same token, it would have been awfully difficult to pick out Amazon, Alphabet or Meta as this century’s biggest winners, at least to date.
“Amazon was a loss-making bookseller, Google (as Alphabet was then) was just two years old and Meta Platforms (Facebook back then) did not come into existence in 2004.”
