Claude has won over developers. Will Anthropic win over investors?

The people around me who do the most programming with AI are clear about their tool of choice: Claude. This is no scientific survey. But when someone chooses a tool for their everyday work, and pays for it, their opinion carries weight.

The data points in the same direction. JetBrains surveyed more than 15,000 professional developers: in May–July 2026, around 39% were using Claude Code, ahead of GitHub Copilot and Codex. That measures adoption, not superiority at every task. But it confirms that the preference is real and widespread. [1]

On October 26, Anthropic filed its paperwork with the SEC to list on Nasdaq under the symbol ANT. There is no price range yet. Its future shareholders will have to decide what that position is worth and how long it can last. [2]

There are customers, and they pay

The strongest argument for Anthropic fits into one figure: more than $16 billion in revenue in the first half of 2026. A year earlier, it was $1.163 billion. Reported growth of 1,300%. [3]

That does not fit the idea of a technology that produces only flashy demonstrations. There is a business here.

A word of caution: this does not mean that every company that buys AI will make money from it. A tool saves work on some tasks and creates errors or supervision costs on others. The test takes place inside each organization, not in a sales presentation.

But the case for it is persuasive. When a company finds a profitable use, it expands it. It starts with programming, moves on to documentation or analysis, and ends up embedding it in its own processes. It stops paying out of curiosity and pays for a capability it uses.

For me, that is the most promising thing about Anthropic: it can grow within its customers, not just by winning new ones.

Research cannot be switched off

To understand its accounts, we need to separate two things. One is running the models and serving users: inference. The other is developing the models that will need to remain competitive tomorrow.

The distinction is technical and economic. In the first half, Anthropic reported almost $9.75 billion in research and development. That includes compute, personnel and other expenses; it is not just the cost of training a new version. [3]

It is tempting to imagine how much it would earn if it cut that bill. The problem is that it would be cutting precisely what makes people choose it.

My reading: much of that research is the price of holding its position. Anthropic needs today's business to finance the next advance without falling into a race where each generation eats up everything the previous one earned.

There are good signs. According to the S-1, it made an operating profit of $974 million in the second quarter, against an operating loss of $1.667 billion in the first. The half-year ended with an operating loss of $693 million. A quarter in positive territory matters. It does not guarantee that it will last. [3]

Buying capacity before you need it

Infrastructure adds another problem. Anthropic has signed major commitments with cloud providers, chip suppliers and data centers. Some contracts require it to pay for capacity even if it does not use it. [3]

The details are in the appendix, to avoid turning this article into a telephone directory of astronomical figures. [4]

The decision makes sense: running out of capacity can prevent it from serving customers. Indeed, Anthropic has acknowledged that demand growth has affected the reliability and performance of some services during peak hours. [5]

The risk arises if you contract infrastructure for growth that never materializes. Revenue can slow before the payments fall due.

That is why looking at sales is not enough. We need to look at its commitments, their flexibility and their price. The capacity that enables growth today can become a cost that is hard to cut tomorrow.

Its partners are playing their own game too

Anthropic's relationships with the tech giants are not those of a customer and an ordinary supplier. Amazon supplies infrastructure and invests in it. Microsoft and Nvidia have also combined technology agreements with investment. [5, 6]

That gives it resources, distribution and financing. It also creates dependencies.

Diversifying suppliers helps, but does not solve the problem. Anthropic negotiates with companies that have their own AI strategies. Their interests may align for a time and diverge later.

Nor should we assume that they want to harm it. They make a great deal of money selling it infrastructure. It is interdependence, not trust forever.

And an essential distinction for investors: money put in by a partner is not customer revenue. A company investing in Anthropic and Anthropic buying its services are separate transactions, even if they form part of the same relationship.

An advantage can slip away without a major defeat

The competitive risk that interests me most is OpenAI winning back those who use these tools intensively. There are reasons to watch it: in the JetBrains survey, Codex went from 3% adoption in January to 16% in May–July 2026. Claude remains strong, but the others are moving. [1]

There is another, less striking scenario: alternatives, including open-weight models developed in China, become good enough for many tasks.

Anthropic could retain customers and keep growing, yet find itself forced to charge less. With expensive research and large infrastructure commitments, margins would narrow.

Its defense cannot depend solely on the next model. It has to turn technical quality into a reliable experience: tools that work, useful integrations and processes that customers do not want to change. Professional loyalty is earned through work. And through work it is lost.

Being strategic does not make it untouchable

I would not discount political risk because Anthropic operates in a strategic sector. Its conflict with the Pentagon is already in court: on October 13, it asked for reconsideration of a ruling that upheld one of its supply chain risk designations. [7]

There are also legal risks unrelated to that dispute. In July, a judge approved its $1.5 billion settlement with authors in a copyright lawsuit. [8]

Being strategic opens doors. It also attracts political pressure and scrutiny. It is no insurance against adverse decisions.

A good company can be a bad buy

My assessment of the business is favorable. My reservation is whether it can keep its advantage without paying too much for it.

Anthropic has already demonstrated demand. Now it has to show that it can sustain it, fund research and meet its infrastructure commitments beyond a period of exceptional growth.

And anyone thinking of buying shares faces another question: at what price?

If the IPO valuation takes years of leadership, near-perfect execution and extraordinary growth for granted, the company could do reasonably well and still disappoint its shareholders.

I find it more useful to look at Anthropic from that angle than to debate whether all AI is a bubble or whether it will transform every business. Claude can be a great tool. Anthropic, a promising company. Neither makes its shares a good investment at any price.

This article is an opinion analysis. It is not an investment recommendation.

Appendix: figures and agreements

Sources consulted on October 29, 2026. Amounts in US dollars. This covers the main public agreements identified, not all private contracts.

The accounts at a glance

First half of 2026:

• Revenue: $16.286 billion.
• Consumption-based: $13.064 billion.
• Subscription-based: $3.222 billion.
• Cost of revenue: $2.380 billion.
• Research and development: $9.741 billion.
• Operating loss: $693 million.

These are unaudited half-year results; they should not be confused with annualized revenue or the second quarter's result alone. The operating result was −$1.667 billion in the first quarter and +$974 million in the second: −$693 million for the half-year. [3]

Major capacity commitments

The prospectus breakdown, also reported by GeekWire, includes: [3, 4]

• Google: $114.5 billion through July 2033.
• Broadcom: approximately $161.2 billion in five-year equipment leases, excluding interest.
• Additional TPU capacity: an estimated $125–150 billion for 2028–2031.
• SpaceX/xAI: up to $120.5 billion through 2029, with certain cancellation options.
• Amazon Web Services: at least $110 billion over ten years.
• Microsoft Azure: at least $44.9 billion through May 2033.
• Nscale: up to approximately $44.6 billion.
• Lambda: approximately $41.3 billion.

Other relevant agreements and announcements

• Fluidstack: an announced $50 billion investment program in US infrastructure, including data centers in Texas and New York. This does not necessarily amount to a separate contract of that value with a single supplier. [9]
• AMD: deployment of up to 2 gigawatts using its accelerators and a planned investment of up to $5 billion in Anthropic. Contracted capacity and investment in the company are different concepts. [10]
• Akamai: an $11.6 billion agreement, expandable by up to a further $9 billion. [11]
• Nvidia and Microsoft: the agreement announced in November 2025 envisaged investments of up to $10 billion and $5 billion, respectively, alongside technological and commercial collaboration. The initial Azure commitment was later expanded, so it should not be added to the updated amount in the previous section. [6]
• Financing linked to Broadcom: a $63 billion transaction with Bank of America, Blackstone, Citi and Morgan Stanley. It forms part of an earlier capacity agreement; it is not automatically another $63 billion in spending. [12]
• Volta/Bitdeer, in Norway: Bloomberg attributed a $10 billion contract to Anthropic. Reuters reported that attribution, but said it could not independently verify it. It is listed separately from the confirmed agreements. [13]

How to read this appendix: do not add all the amounts together. The list mixes minimum commitments, potential ceilings, estimates, investment programs and financing for transactions already included, with different terms and conditions. The figure that matters for assessing risk is how much Anthropic will have to pay, when, and how much room it has to adjust that spending if demand changes.