Insights

Anthropic could raise up to $100 billion, more than ten times the value of the entire Macedonian Stock Exchange. What will your company keep?

The planned Anthropic listing shows how much money now flows to the companies building the leading AI models. A Macedonian company can still keep its data and decisions for itself.

In September, at the annual conference of the Macedonian Stock Exchange in Ohrid, Data Masters’ founder said that Macedonia does not have the resources to build a new Anthropic. On 26 October, Anthropic, the company behind Claude, showed what that sentence means in money. It filed to list on Nasdaq and, according to Reuters, is considering raising up to $100 billion, more than ten times the roughly 5.3 billion euros that all the companies on the Macedonian Stock Exchange are worth together.

$100Bwhat Anthropic may raise, according to Reuters
10x+the value of all companies on the Macedonian Stock Exchange (about €5.3B)
$895Bcommitments in the 2 November filing, half due by the end of 2030
4%+Nasdaq drop on 29 and 30 October over AI spending doubts

Value gathers at both ends

In the first half of October, a free replacement for Adobe’s design suite, which designers pay for every month, appeared for download, built from scratch. It is an early test version that its author says was written with the help of AI. Meanwhile, Anthropic’s AI assistant Claude started turning plain questions into live data dashboards: it connects to the company’s data and decides on its own which numbers to pull. The software in between is getting cheaper to build, and value gathers at both ends of the same chain: with the few companies that build the leading AI models, and with the companies that own their data and the decisions they make on top of it. The first end takes money on the scale of this listing, while the second is within reach of every company here.

The price of dependence

A supplier listed on the stock market must answer to its shareholders, and the bill, which will be partly covered by the money raised in the listing, is large: in its amended filing of 2 November, Anthropic reports commitments of about $895 billion, almost nine times the amount it could raise on the market. Nearly all of it is for computing capacity and infrastructure, and half of it is due by the end of 2030. Investors are already asking who will pay such bills. On 29 and 30 October, after the results of Google’s owner Alphabet, as well as Amazon and Intel, the Nasdaq index lost more than four percent over doubts about whether investments in AI will pay off, and how quickly. Like any supplier, a company that builds models recovers those costs from the businesses that work with its models every day, which is why it matters under what terms they become dependent on it.

In August, Data Masters wrote about the 20-dollar illusion, and two sentences from that article weigh more today than when they were published:

The license may cost 20 dollars. But dependence is paid for with data, processes, integrations and the lost freedom to change the supplier.

Companies here still pay that price in data without noticing: they enter real numbers into a chatbot to save an afternoon, and do not ask where those numbers are now kept. An individual can hand everything to the model and lose little, while a company that does the same hands over the business itself.

The end you can decide

Some companies in the region keep their most important number in the spreadsheet of a single analyst, and their first impulse, the week a model with such capabilities appeared, was to give it everything they have. Behind that impulse is an understandable line of thinking that most owners here would share: they want an answer to every question from the whole company, based on all their data, so it seems they must hand all of it to the model. There is a choice, though, and it starts at the place where the answer is put together.

The complete answer is built on a foundation set up where your data already is, under your own roof. The article on trust in AI described the rule by which Data Masters builds systems: the rules and definitions are separate from the model and give the same result whether or not AI is involved. The model reads the descriptions of the data and creates a map of what you have. With that map, the model can decide which numbers it needs, while the calculation stays with you. Set up the foundation once and every department will work from shared definitions, so one answer can cover the whole company, and when a definition changes, every answer that uses it changes with it.

1Your datastays where it already is, under your roof
2Shared definitionsrules kept separate from the model
3The model readsthe descriptions and maps what you have
4The calculationstays with you, every answer the same

In its own guide to Claude Dashboards, Anthropic advises users to check every dashboard to see how Claude arrived at the number:

If it is not calculating what you had in mind, tell Claude what to change.

That means the number must be defined by the company, so the model relies on a foundation that someone has to build, and without that foundation even the best model can only guess.

One company was setting up an AI agent over its data so that people could ask it questions. The plan was to save every question, have the IT department check how the answer was calculated, and from then on always answer that question the same way. The obvious question is how this replaces thousands of reports, when in the end there would again be thousands of saved questions. The hardest questions, those that link several calculations, demand the most work from the model, and all of that work is paid for. The article on Power BI put it this way: “Every new question creates another report.” A collection of saved questions is a collection of reports under a new name and, like any collection of reports, it needs a foundation.

That is the uncomfortable part for the many Macedonian companies that still do not have such a foundation. A company without a number it trusts will most readily reach for the shortcut straight to the model, and it is exactly that company that risks the most, because it has nothing to check the answer against and, to get it, has already taken its data out of the company. But building the foundation is no longer a job of several years, because the model can now read how your systems are organized and complete a large part of the assembly on its own, while your people keep the part only they can finish: deciding what each number means. So the model appears twice and works for you both times: first it helps build the foundation, and then it works on top of it.

What to do

In July, Data Masters wrote that the model is rented, but the knowledge must be yours, and this listing is the strongest reason so far to act on that: use the cheap tools and the rented models, but keep your data and decisions. Use the foundation for the questions you ask week after week, with the same answer every time and at a price you have already paid, and call the model for the rest, because you pay again for its answer every time the same question comes back.

Rent

  • The cheap tools
  • The rented models
  • The model for the questions that are new

Keep

  • Your data
  • Your decisions
  • Your definitions
  • A named owner and a written rationale behind every number

Behind every number and every decision the model touches, there should be a person with a name and surname and a written rationale. That record is the memory of the company and, with a supplier that answers to its shareholders, it matters even more that it stays yours, no matter which model you use. When a tool promises that everything will be easy if you give it everything you have, remember that the freedom to change the supplier is part of the price too. Let the model read your definitions, never your customers.

A seven-day test

Every owner can run a simple check: ask three people in the company what they entered into a chatbot in the last seven days, and if the answers include real customer names or real figures, the business has already started to walk out of the door, day by day.

The price the market sets for Anthropic at the listing, and afterwards, will show how much investors think the end of the chain where models are built is worth. For a company in this region, the assessment closer to home matters more. If you hand your business over to a model, what exactly are you giving it, and will you ever be able to take it back?

Value Discovery

Bring the decision that is late.

A free Value Discovery, sixty to ninety minutes, built around your industry and your Monday morning.