Should I Buy the Anthropic IPO? What you need to know

Last reviewed: 27 October 2026

Should I buy the Anthropic IPO?

That’s the question filling my inbox this morning. Anthropic filed its S-1 with the US SEC yesterday. The Claude maker is heading for a Nasdaq listing under the ticker ANT, probably the week of 16 November. Some investors are whispering valuations between US$1.5 trillion and US$2 trillion.

My short answer is no. Not at those numbers. Not for most people.

I’ve been around long enough to watch a few of these. The first-day pop is usually for the people who already own the shares. Public buyers get asked to pay a price that assumes everything goes right for years. This one is no different, only bigger.

What the filing actually shows

The growth is real. Revenue hit US$16.3 billion in the first half of 2026. That’s up roughly 1,300% on the same six months a year earlier. Second quarter alone brought in US$11.6 billion and produced an operating profit of US$974 million. At the operating line, the business is starting to work.

The giant net loss of roughly US$80.7 billion for the half is mostly non-cash. Accounting adjustments on the rising value of convertible notes as Anthropic’s own valuation climbed. Operating cash flow was positive over the period. Anyone telling you the company is burning US$80 billion in real cash hasn’t read past the first line.

Metric (H1 2026)	Figure	Notes
Revenue	US$16.3bn	+1,300% vs H1 2025
Q2 revenue	US$11.6bn	—
Q2 operating profit	US$974m	First operating profit
H1 operating loss	US$693m	Narrowed sharply
Net loss (mostly non-cash)	US$80.7bn	Valuation-driven notes
Compute & infrastructure	US$8.9bn	+283% year on year

One more number that sticks. First-half 2026 revenue is close to 400 times the company’s full-year 2023 revenue. Three years ago this was a research shop. Now it books more in a quarter than most ASX 200 companies book in a year.

The compute commitments are the real story

Here’s where it gets uncomfortable. Anthropic needs vast amounts of computing power to train and run Claude. It has locked itself into paying for that power whether customers show up or not.

These are minimums, payable even if the capacity sits idle:

Supplier	Minimum commitment	Through
Broadcom (AVGO) equipment leases	US$161.2bn	multi-year
SpaceXAI compute	up to US$120.5bn	multi-year
Alphabet / Google (GOOGL) cloud	US$114.5bn	2033
Amazon (AMZN) cloud	US$110bn	2036
Microsoft (MSFT) cloud	US$44.9bn	multi-year
Extra TPU capacity (midpoint)	US$137bn	2028–2031

Add it up and you’re looking at well over half a trillion US dollars of contracted spending against a business that books US$16 billion a half. The bet is that demand keeps compounding fast enough to grow into those commitments. It might. But if growth slows even a little, Anthropic is left paying for expensive, underused kit. The company flags exactly this risk in its own prospectus.

This is the single most important thing the filing teaches a retail investor, and it applies to the whole sector. The AI build is being funded by enormous, fixed, long-dated commitments. The revenue to cover them is real but variable. That gap is where the risk lives.

The circular financing problem

This is where it gets genuinely interesting, and a little uncomfortable. The same companies Anthropic is paying hundreds of billions for compute — Amazon, Google and Microsoft — are also three of its biggest investors and biggest sales channels. Money goes round in a circle.

Customers buy access to Claude through their existing Amazon, Google or Microsoft cloud accounts. The cloud provider bills the customer, keeps a fee and passes the rest to Anthropic. That channel accounted for 48% of Anthropic’s revenue in the first half of 2026, up from just 11% in 2023. Meanwhile Amazon has tipped in around US$18 billion (with up to US$15 billion more committed) and Google around US$23 billion (with another US$20 billion pledged). Anthropic then turns around and commits to spend even more than that buying their cloud and chips.

Anthropic’s own filing describes these partners as “investors, customers, distribution partners and competitors” all at once. It warns they have visibility into its pricing, that they could favour their own AI models, and that the arrangement has already drawn antitrust attention.

There’s concentration too. Anthropic’s five largest customers made up roughly 25% of its revenue over the first nine months of 2026. Lose one or two of those, or lose a cloud partner’s goodwill, and the picture changes quickly. None of this means the business is doomed. It means the quality of the growth is harder to judge than the raw percentage suggests.

Valuation, control and the mission

Reports point to an expected valuation of US$1.5 trillion to US$2 trillion. Using the rounded US$16.3 billion first-half figure and simply doubling it gives about US$32.6 billion annualised. That puts the indicated valuation at roughly 46 to 61 times annualised revenue. It is an illustration, not a forecast. It still shows how much future growth buyers are being asked to pay for.

Another way to look at it. A US$2 trillion equity valuation divided by a 30 times earnings multiple implies roughly US$67 billion of annual net profit. At a 20% net margin that would require about US$333 billion of annual revenue. Neither the multiple nor the margin is a prediction. The exercise just translates a large number into a business outcome you can question.

Governance is another issue. The founders will hold a single Class F share carrying 50.1% of general voting power. As a public benefit corporation, Anthropic is explicitly allowed to put its mission ahead of ordinary shareholder returns. A Long-Term Benefit Trust sits above ordinary shareholders. You can admire the intent and still recognise it as a real risk to the share price. Safety decisions can protect the business. They can also delay products and revenue. Ordinary shareholders have limited influence over those trade-offs.

The S-1 itself has been called an “anti-sell”. Its risk section runs long. In one section the company says it will “depart from profit maximization” and make forward investments that “may seem irresponsibly large”. It warns that its own AI could be misused or allowed to operate with a high degree of autonomy. You don’t often see a company try this hard to talk nervous money out of the room.

What Australian investors should actually do

Start by accepting that you’re probably already in this trade. If you own a Nasdaq-100 ETF, a global tech fund, or a typical industry super growth option, you hold Amazon, Microsoft, Alphabet and Nvidia in size. The question isn’t whether to get exposure to AI. It’s whether you’re being paid properly for the risk you already carry.

Here’s how we’re thinking about it at MPC Markets:

1. Know your real exposure first. Add up how much of your portfolio sits in the same handful of US mega-caps across your direct holdings, your ETFs and your super. Most people are more concentrated than they think.
2. Prefer the picks and shovels over the gold miners. The surest money in a build-out is often made selling the equipment, the memory and the power, not in the headline names paying the bills. Think the chip, memory and energy-infrastructure layer rather than paying any multiple for the model makers themselves.
3. Respect the cycle. Our current house view is cautious, with the economy sitting in a slowdown phase. That argues for selectivity and some tactical cash, not for chasing the most expensive growth on the board.

And a watchlist for the next twelve months:

• The Anthropic pricing and first-day trade in November. It’s a live read on how much AI risk public markets will actually swallow.
• Whether revenue growth keeps slowing. Reports of third-quarter growth cooling already fed a broader tech wobble this month.
• Hyperscaler capex discipline. If Amazon, Microsoft and Google start questioning the return on their own AI spending, the whole supply chain feels it.
• Cheaper Chinese open-weight models and what they do to pricing power.

None of this is a reason to avoid AI. It’s a reason to own it deliberately, at a sensible price, with your eyes open to the compute bill underneath. Great technology and great investment are not the same thing. The gap between them is usually just the price you paid.

Frequently Asked Questions