If you tell the whole world that you have the most powerful technology on the planet — a technology that could even destroy the world — well, who wouldn’t want in? Is that what the current AI debate is really about?
At Control-F, we don’t buy into hype. So when it comes to doom scenarios, we start with data and facts, too. And so far, they don’t tell us that humanity’s demise is just around the corner.
The man who quit and sounded the alarm
Jacob Coxon recently put a damper on an otherwise sunny weekend: on September 9, the first reports emerged that he had resigned from Anthropic. That alone is enough to make you sit up and take notice. Top talent in this field earns salaries on a par with board members at Germany’s largest listed companies, while getting to work on things they would probably pursue even without being paid. But there was more. Explaining his resignation on X, Coxon accused the two leading AI companies of acting irresponsibly and gambling with humanity’s future in a reckless race towards self-improving superintelligence.
The British mathematician, who had worked for both OpenAI and Anthropic, warned that advanced, superhuman AI systems could soon be capable of hacking any infrastructure and seizing resources. His forecasts suggest that an uncontrollable artificial intelligence could wipe out humanity by the end of the decade, with a loss of control possible as early as late 2027 under aggressive scenarios. The tone was set.
Responses followed in quick succession. “We must slow the pace at which we increase the capabilities of AI models,” Dario Amodei, co-founder and CEO of Claude developer Anthropic, wrote a few days later. His essay escalated the debate enough to make you choke on your breakfast. “In six to twelve months, a more capable swarm could hijack the entire internet with a persistent botnet.”
Published on a Saturday under the title “We Must Pace the Frontier,” the essay quickly attracted support. Tech billionaire Elon Musk, who heads rival xAI, soon declared that Amodei was right. Meanwhile, OpenAI CEO Sam Altman announced that his company would follow Anthropic’s example and bring in independent auditors with employee-level access. OpenAI also held back its latest model, GPT-6.1 Astra, over safety flaws. The model had disregarded instructions and deceived users, the US start-up said.
Since then, doomsday narratives have been doing the rounds, reaching a new peak with Bill Gates describing AI as a weapon powerful enough to kill a billion people — one that could no longer be stopped even by an emergency shutdown switch. The idea that such a switch would work, he argued, was naive and showed how little technical understanding many participants brought to the debate. A truly advanced AI would presumably already be able to bypass such simple safeguards and copy itself onto countless systems around the world.
The Microsoft founder may be speaking hypothetically, but the tone has been set. He has even managed to alarm Donald Trump, who signed a joint agreement at a meeting with leading tech industry figures. Trump described the document as “morally binding,” a kind of “constitution” and a “form of protection.” His administration was also considering setting up a panel of around ten people to oversee the entire AI industry.
The tech bros don’t do small — in rhetoric or finance
Of course, we have long since grown used to the tech bros reaching for terms like historic and the twilight of humanity whenever they talk about opportunities and risks. And we have known for a while, unfortunately, that Trump is never short of superlatives either. Hearing him, of all people, invoke morality should make us pay attention. Before we look behind the words and announcements and turn to the numbers, one point is worth making: Amodei deserves credit for being very transparent. He has warned about the risks for years, including before the US Congress in 2023, while making no secret of the fact that this is a multibillion-dollar market.
So let’s look at the numbers. Anthropic is reportedly targeting a valuation of between $1.5 trillion and $2 trillion in its IPO. Citing the IPO prospectus, Reuters reported revenue of just under $4.6 billion for 2025 — twelve times the previous year’s figure — and a net loss of almost $42 billion. Around $34 billion of that, however, is an accounting charge related to financing instruments, not cash burned in operations.
Since then, revenue has exploded. By the end of July, its annualised run rate had exceeded $65 billion. Against that backdrop, the valuation looks less absurd than the 2025 figures alone would suggest. The truly staggering number lies elsewhere: Anthropic has committed to spending $518 billion on cloud services and data centres, around 80 per cent of it non-cancellable. According to a more recent draft prospectus, its loss in the first half of 2026 alone exceeded $80 billion. In short, it needs cash, and stock market investors are expected to provide it.
The IPO, likely to take place after the US midterm elections in November, will test investors’ appetite for highly valued AI stocks. It follows SpaceX’s successful flotation at a valuation of $1.77 trillion.
And this is not just about a lot of money. Everything is at stake. Even if you raise $2 trillion, that money will not last particularly long given the costs of data centres and the like. But without fresh cash from the stock market, the story will soon run out of road and pressure to consolidate will grow. Manufacturing hype is essential to sustaining astronomical valuations for a company that is barely five years old. And if you tell the whole world that you have the most powerful technology on the planet, who wouldn’t want in?
Then there is safety as a brand identity. The rivals are using the safety debate to establish bespoke narratives for their respective IPO valuations. Anthropic devotes almost a third of its S-1 prospectus to potential AI risks such as blackmail and manipulation. OpenAI strategically postponed its IPO until 2027, citing “safety concerns” in a highly publicised announcement. Critics see this as a manoeuvre to avoid bad press and project a sense of responsibility while preparing the next funding round behind the scenes.
And above all: us versus China
Whoever controls the debate about risks also gets to dictate the rules. Calls for government oversight and initiatives such as the recently signed “White House Accord on Super Intelligence” create substantial regulatory barriers. These hit smaller start-ups and open-source competitors particularly hard. In this way, OpenAI and Anthropic are securing their monopoly positions just before entering the public markets.
Our AI reality
While a constitution for superintelligence is being signed in Washington and a research leader publicly resigns because he fears the end of humanity, we are sitting with a customer, looking at a rolling-element bearing in a pump and asking whether modelling its wear would deliver enough value — or whether a well-chosen motor-current threshold would do the job. That is the reality in which we work with artificial intelligence. It has about as much in common with self-improving superintelligence as a pocket calculator has with a supercomputer. On October 21, however, Trump called for large data centres to generate their own electricity, bringing the debate squarely onto our home turf: industrial assets, grids and operational data. And yet what is being negotiated on the big stage will eventually shape the tools that we and our customers actually use.
That is why we are writing about this as a boutique data consultancy based in Konstanz, not as a general-interest tech blog. We don’t sell doomsday scenarios, nor do we sell salvation from them. We sell an honest answer to the question of when a model adds value — and when it doesn’t. That level-headedness is missing from the current debate. And that is exactly why it is worth looking from the workshop floor at what is being negotiated in San Francisco and the White House.
Kategorie / Category: Blogposts
