The recent report of the Center for Strategic and International Studies puts into context the current war scene in the Middle East: during the first 100 hours of war with Iran, U.S. spending reached US $3.7 billion.

It's difficult to explain how much of this was dedicated to the use of IA, but considering the Pentagon's contracts with the Big Tech - plus a key player, Palanti - whose budgets for this year have broken record and considering the uses of IA at war, it's reasonable to assign about 25% of spending to IA, that's about US $925 million.

In a military operation called an Epic Fury, IA applies, inter alia, to:

  • The information referred to in Article 3 (3) of Regulation (EC) No 1065 / 2006 shall be published as:
  • automatic prioritization of targets,
  • missile path prediction.
  • an automated analysis of satellite images,
  • detection of patterns,
  • classification of targets,
  • Drones, radars, early warning systems.

The conflict between Anthropic, the company behind Claude, and the U.S. government is a result of the Defense Department's intention to automate attacks using IA under the concept "human on the loop" that is, with human supervision while Anthropic believes that the applicable ethical principle should be human intervention, or "human in the loop" that means that IA should be not only monitored but should, before applying, validate each decision with a responsible human.

The same goes for the massive surveillance that the U.S. government wants to enforce within its national territory.

Anthropic's ethical standards run into the Trump administration's initiative and that's why the company was declared "supply-chain-risk" that means that every U.S. public office should terminate its contracts with Anthropic in less than 6 months.

For the company founded by two former OpenAI, this represents an immediate loss of nearly US $300 million, plus cancellation of future contracts. That's why they're going to justice to seek compensation.

Now good.

Even though this scenario makes clear that, in the U.S., a key client of the most important technologies companies in the world is the State, even the above-mentioned amounts do not explain how the development of the IA currently gets financed.

The primordial use of communication technologies for military purposes isn't new. Indeed, the Internet was mastified as an ARPANET by-product, an encrypted military communication development, as did GPS and satellites.

But the current rate of evolution of the IA and its disruptive capacity are a result of an investment and development level that had never been seen before.

Training a border model as ChatGPT costs between US $1,000 and US $10 billion. The computation capacity that this means (as much as electricity) generates virtuous agreements with companies as Microsoft Azure and NVIDIA.

Meanwhile, this week WiRed published an interview with Nick Fox, Google's Vice President of Knowledge and Information, and asked Fox about whether Gemini will have an advertising (paid content) as a commercial policy intended to gain money for using his IA. Fox responded that they do not rule out to do so but that they are not sure that's the business model.

Similarly, today none of the models that compete with Gemini accept sponsorship. Nevertheless, so far, no IA company has been making profits.

The previous challenge, with the consolidation of the platform economy, as social networks and mobile applications became massive, became clear that the strategy was to offer free tools for users' data. The model then moved to "freemium" proposals.

But today's race to master the IA market has no comparison.

Dario and Daniela Amotii founded Anthropic in 2021 and 5 years after the company was valued at US $350 billion after the last round of investment of February this year. But the company will lose, by 2026, US $5 billion.

NVIDIA, the most valuable company in the world - US $4.39 billion, double Brazil's GDP, an onceava global economy, key to the IA for manufacturing processes - is an OpenAI partner and posted on its website that Microsoft owns, at the ChatGPT creator, a share of US $13.5 billion, though OpenAI projects US $14.000 million loss this year.

To give more dimension to these numbers: only 8 countries from Latin America generate, every year, as much or more capital as Microsoft has invested in OpenAI. The share of pesos for dollars between Argentina and the United States was performed for US $20 billion.

So, beyond the fascination of the daily advances of ChatGPT, Claude, Gemini and others, how do we explain the A's business? What's his point?

Circular arrangements for purchase of computation

As described above, Anthropic received substantial investment from Microsoft and NVIDIA.

The latter, for their part, ratify information but the arrival of money depends on compliance with contractual commitments that require Anthropic to buy billions of dollars in Microsoft Azure computation and NVIDIA hardware. It's a closed circuit: capital goes into Anthropic through the "strategic investment" door and goes out through "required infrastructure consumption."

The Amotii brothers' lab receives liquidity to train increasingly powerful models, but that training can only be bought more from their investors: cloud and hardware giants are guaranteed, for years, the sale of energy, data centers and GpU-Graphics Processing Unit, superior to standard processing with the capacity to train IA.

Finally, anything impacts on the valuation of companies, whose share keeps growing. Nevertheless, following industry specialists, the loop described above is a bubble that could be exploded at any time.

Dominating positions, vertical integration

On the other hand, the old paradigm expressed in "data are 21st century oil" also works.

The Big Tech have already consolidated their dominant position in industry. But IA's dung allows them what's known as the vertical integration of products.

Microsoft integrated Copilot into its Microsoft 365 Suite and, in each program, the role of the IA keeps growing. Google had already embedded a search engine into the browser, Chrome, and now imposed Gemini.

Claude and ChatGPT are new but so powerful that earlier tools start to include them - recently version 5.2 of ChatGPT offers to integrate with Gmail and Claude can work within Excel.

Of all agents, they are expected to become buyers of mass consumer products later this year. Maybe at that time we will start selling advertising to convince our next IA assistants about their purchase decisions.

These and other maneuvers are possible because the companies that dominate the game are so powerful that they can wait years for their bets to be profitable while they lose money as we have mentioned earlier, as long as they are gathering information as valuable - and forging a strong relationship with consumers - as to become dominant or to reaffirm their position at a new vertical.

That's new? Not at all.

Google - Alphabet- bought YouTube in 2006, for US $1.65 billion, as the platform lost money and faced huge legal risks. To put an example, statistical data indicates that YouTube closed its accounting in 2008 with a red balance of about US $400 billion.

The company took about 9 years to get profitable (just about 2015). Today, as currently reported, YouTube generates more than $50 billion a year in income and is one of Google's most valuable assets.

Just in 2018 YouTube consolidated its premium proposal with subscription. Until then, users pay only with data. A similar thing's happening with IA, but the dung's so much bigger that there's far less resistance to pay entry values.

Edoe Cohen is a system engineer who has developed several startups at Silicon Valley. His last release is an application of healthy food delivery, about Blockchain.

"We decided to do so about Blockchain because we want the value of connecting people around a service to stay in the community. The big tech's business model chokes competition and even users and providers, because the initial investment is so big and so many capital losses behind companies that centralise information, they later abuse their dominant position to get their accounts better" reflect.

A's laboratories such as Anthropic and OpenAI do not listed and private financing rounds are starring the big tech and some great known investors, such as Salesforce and Sequoia.

Thus, what's clear is that the massive adoption of IA isn't the final goal of those who are currently competing for an IA race.

Nevertheless, the fact that their development transformed once and again the society we live in remains disturbing. Especially if we believe that IA agents do not come up as solutions to pressing problems, as in health-related areas, for instance.

Thus, while we are passing through a moment of constant fascination, nobody dares to accept that at least for now, the massive adoption of IA isn't business.

None of the representatives of these companies declare the plan for these developments to be sustainable. No monetization model justifies the use of IA agents at mass consumption level.

While in the stratosphere, giants put together virtuous circles with a hype that trembles with conflicts such as Anthropic's with the U.S. government, and users have been amusing with toys that they did not demand, perhaps at corporate level automation will be a key: less human, more software, more efficiency and integration, but less work.

"Monitoring market and competition, mitigating risks, ensuring scalability and selling infrastructure" says my IA assistant.

That's another capitalism.

No, it's actually another world.