Nvidia Weighs $10B for Anthropic’s $2T IPO — Here’s the Real Bet
Nvidia may put up to $10 billion into Anthropic’s IPO. That does not prove circular AI demand — but it should make every investor ask who is funding the next wave of spending.
Nvidia potentially putting up to $10 billion into Anthropic’s IPO should make every investor uncomfortable.
Not because it proves anything crooked. It does not. But it would put Nvidia in the position of backing a major customer whose growth could drive more demand for Nvidia’s own infrastructure — and that is a relationship worth examining properly.
Reuters reported on September 11 that Nvidia is in talks to become an anchor investor in Anthropic’s prospective listing. Anthropic is reportedly seeking to raise as much as $100 billion at a valuation of roughly $2 trillion; Nvidia is considering an investment of up to $10 billion. None of that is final. The discussions can still change or disappear. But the idea alone tells you exactly where this AI boom is heading. ([au.marketscreener.com](https://au.marketscreener.com/news/nvidia-in-talks-to-invest-in-anthropic-s-mega-ipo-sources-say-ce785bdfd18ff127?utm_source=openai))
This is not just an IPO. It is a demand test.
Let’s call the numbers what they are.
A $100 billion float would be enormous. A $2 trillion valuation for Anthropic, the maker of Claude, would put a five-year-old AI company in the same broad valuation conversation as the world’s most established industrial empires. And Nvidia potentially taking 10% of that headline fundraise would make it more than a passive spectator.
Nvidia sells the computing infrastructure that turns AI ambition into a very expensive monthly bill. Anthropic needs staggering amounts of compute to train models, serve customers and remain in the fight against OpenAI, Google, Meta and whoever turns up next with a clever model and a bottomless sovereign fund.
That makes Nvidia’s interest perfectly rational. If Anthropic becomes a bigger, better-capitalised buyer of AI infrastructure, Nvidia wins twice: first through a valuable equity holding if Anthropic lists well, and again if Anthropic spends heavily on the hardware and cloud capacity required to keep Claude competitive.
In November 2025, Nvidia said it would invest up to $10 billion in Anthropic under a broader arrangement that included Anthropic committing to buy $30 billion of Microsoft Azure capacity powered by Nvidia technology, Reuters reported. This proposed IPO investment would not be happening in a vacuum. It would extend an already deeply connected commercial relationship. ([au.marketscreener.com](https://au.marketscreener.com/news/nvidia-in-talks-to-invest-in-anthropic-s-mega-ipo-sources-say-ce785bdfd18ff127?utm_source=openai))
That does not mean the revenue is fake, or that the arrangement is improper. It means investors should be clear about what they are looking at: a supplier investing in a customer that may also be a major buyer of the supplier’s technology.
When a supplier invests in its customer, and the customer uses the capital to buy more from the supplier, revenue can be real, useful and perfectly legal — while still being less independent than it looks from a distance.
Anthropic has earned the right to be taken seriously
Before the usual crowd starts screaming “bubble” like it is an insight, let’s be fair.
Anthropic is not a bloke with a slide deck and a hoodie asking for $20 million to disrupt spreadsheets. It has built one of the few AI products businesses and developers genuinely use at scale. Claude has become a serious competitor in coding, analysis and enterprise workflows. The company has raised more than $125 billion privately, according to Forbes, and reported revenue momentum has been extraordinary. Forbes said Anthropic’s annualised revenue run rate had passed $65 billion ahead of a potential IPO. ([forbes.com](https://www.forbes.com/sites/jonmarkman/2026/09/02/anthropic-plans-to-release-ipo-prospectus-after-labor-day/?utm_source=openai))
That is real commercial traction. It matters.
It also explains why investors are willing to entertain a valuation that would have sounded like satire two years ago. If AI models become a core layer of business infrastructure — something companies use for software development, customer service, research, legal work, finance, operations and scientific discovery — the prize is not small.
The prize is colossal.
But here is the snag: colossal markets do not automatically create colossal shareholder returns.
Railways transformed economies and bankrupted plenty of railway investors. The internet changed civilisation and wiped out a generation of dot-com shareholders. The question is never whether a technology matters. The question is who captures the profit after everyone has paid for the machines, electricity, staff, data, distribution and inevitable price wars.
Anthropic’s prospective valuation assumes it can do more than build an admired model. It assumes it can keep customers, defend pricing, manage eye-watering infrastructure costs and build a durable business while well-funded rivals attack from every direction.
That is a bloody big assumption, even for a company doing serious revenue.
The circular-money worry is not paranoia
The lazy version of this argument is: “Nvidia investing in an AI company means the whole thing is fake.”
That is nonsense.
The smarter argument is not that this proposed IPO investment proves circular demand. It does not. The argument is that arrangements like this can make it harder to judge how much demand is truly independent. Nvidia has every incentive to encourage an ecosystem that requires more Nvidia compute. Cloud providers have every incentive to fund AI labs that commit to long-term cloud contracts. AI labs have every incentive to raise more capital because scale remains the game.
Everybody has a reason to keep the flywheel spinning.
That does not mean the flywheel is imaginary. It means outsiders must distinguish between end-customer demand and capital-funded capacity demand.
They are not the same thing.
If a bank, fund or strategic investor gives an AI company billions, and that company uses the cash to lock in billions of compute, the infrastructure supplier can report genuine sales. But as an investor or operator, you still need to ask the adult question: are ordinary customers buying enough AI products at healthy margins to justify the whole stack?
That is what Anthropic’s IPO prospectus will matter for. Not the glossy AI vision. Not the revenue run-rate headline. Not the cult of the founder.
Show me customer concentration. Show me gross margins after serving inference-heavy workloads. Show me contract duration. Show me how much growth comes from usage that customers fund from their own operating budgets rather than experimental innovation budgets. Show me what happens to margins if models become cheaper, more interchangeable or open-source alternatives improve.
Forbes correctly identified the central tension: frontier AI is not traditional software. Higher usage can create more revenue, but it also brings substantial ongoing computing cost. A software company that scales without equivalent cost growth deserves software-style multiples. A company whose costs rise aggressively with every successful customer deserves a much harder look. ([forbes.com](https://www.forbes.com/sites/ronschmelzer/2026/08/14/anthropic-at-2-trillion-is-ai-entering-bubble-territory/?utm_source=openai))
The overlooked angle: Nvidia may be buying protection, not just upside
Most people will read this as Nvidia being bullish on Anthropic.
It is that. But it may also be Nvidia reducing risk.
Nvidia has become the central tollbooth of the AI boom. That is wonderful while demand outruns supply. It becomes more complicated if the largest AI labs start building custom chips, squeezing suppliers, slowing capacity expansion or shifting workloads across competing clouds.
A strategic stake in a major AI customer helps keep Nvidia close to the decision-making. It makes commercial alignment stickier. It gives Nvidia a chance to benefit from the model layer, not merely the hardware layer.
That is smart business. Jensen Huang did not build Nvidia into a giant by waiting politely for the industry to decide his fate.
But strategic alignment cuts both ways. The closer the AI ecosystem becomes, the more investors need to watch for the difference between a thriving market and a highly concentrated circle of giants underwriting one another’s expansion.
There is another wrinkle. Nvidia’s public messaging has recently pushed back against some of the louder existential-risk rhetoric around AI. Axios reported that Huang argued AI fears were being used to create demand for cybersecurity business. ([axios.com](https://www.axios.com/2026/09/10/nvidia-ceo-jensen-huang-ai-anthropic?utm_source=openai))
Whether you agree with him is beside the point. The commercial reality is simple: Nvidia is making a very large wager that AI keeps moving from experiment to essential infrastructure. Its potential Anthropic investment would make that wager even more direct.
Do not confuse a great company with a great entry price
This is where investors usually make a mess of it.
They see a company growing at ridiculous speed and assume buying exposure at any price is intelligence. It is not. It is excitement wearing a tie.
Anthropic could become one of the most important companies in the world and still be a poor investment at $2 trillion. Nvidia could be right to invest strategically at $10 billion and the public market could still overpay after the opening bell. Both things can be true.
At the headline figures, a $100 billion raise on a $2 trillion valuation is roughly 5% of the valuation. If Nvidia bought the full $10 billion allocation, it would account for about 10% of the capital raised. That does not prove anything sinister. It does mean the IPO would arrive with one of the industry’s most powerful players visibly invested in the outcome.
That is not a reason to panic. It is a reason to read the fine print.
What this means for you
If you are a founder, stop telling yourself that “AI strategy” means buying the most expensive model available.
Your advantage will not come from renting intelligence by the token. Everyone can do that. Your advantage comes from proprietary data, a painful customer problem, workflow integration, distribution and the discipline to measure whether the AI actually improves margin, speed or retention.
If you are an operator, treat AI spend like a serious capital project. Demand a baseline before rollout. Measure hours saved, errors reduced, revenue created and customer experience improved. If the numbers are fuzzy after 90 days, cut it. A chatbot that makes your team feel modern is not a business case.
If you are an investor, separate the layers. Chip suppliers, cloud providers, foundation-model labs and vertical software companies do not have the same economics or risks. Do not value them as if they do.
And if Anthropic’s prospectus lands, ignore the theatre and make a short list: revenue quality, gross margin, infrastructure commitments, customer concentration, cash burn, related-party commercial arrangements and the pathway to free cash flow.
That is the work.
Nvidia’s possible $10 billion cheque is not proof that AI is a bubble. It is proof that the next phase of AI will be financed, interconnected and far less simple than the headlines suggest.
The winners will be the people who can tell the difference between genuine customer value and a very expensive game of pass-the-parcel.