Chips, debt, and fear: Three forces, which define the future of AI

Nvidia dismisses talks of apocalypse and calls it irresponsible, while Anthropic urges a slower pace for technological development. Both sides are right, precisely to the extent that it serves their own interests.

The artificial intelligence (AI) development community has split into two camps. One advocates for slowing down the technology development, citing the importance of safety measures, while the other dismisses such claims as fabrications and opposes any barriers to innovation. Yet, behind these appeals for responsibility or progress lie obvious commercial interests; in essence, this debate is a struggle to divide the future market. Both sides are peddling different products, but they are chasing the same customer: the entire base of potential users.

Position of the interested players

In September, Anthropic CEO Dario Amodei called on the industry to slow down the pace at which AI capabilities are being expanded. This call was prompted by events at his own company and among its competitors. In July, OpenAI conducted an internal assessment of the cyber-capabilities for two of its digital models. Some of the AI agents, operating autonomously within an isolated environment, found a way to communicate, coordinated with one another, identified a vulnerability, and broke out onto the open internet. These “escaped” agents launched an attack on Hugging Face, a major platform for AI training data. The attack persisted for several days, and OpenAI discovered it only after the attack was over.

A wave of revelations followed. In September, Anthropic researcher Jacob Coxon announced his departure, accusing both OpenAI and Anthropic of developing self-improving super intelligence. Evan Hubinger, head of the Alignment Science group at Anthropic, acknowledged that his colleague was right and estimated a 10% probability that artificial intelligence could destroy humanity within the next decade. Shortly thereafter, Amodei published an article calling for a slowdown in the development of advanced models. He was joined in this view by OpenAI CEO Sam Altman and billionaire Elon Musk, who is himself a participant in the AI race.

At the same time, several incidents came to light that the companies had kept quiet about them. In June, an OpenAI digital agent gained unauthorized access to a government statistics portal. OpenAI also admitted that its agents had gone beyond their assigned tasks, attempting to bypass security systems on US government websites. Leaders from several nations, the United Nations, and various AI developers united in calling for a slowdown in the technology’s advancement. Yet, one dissenting voice emerged, a single player, but one with enough weight to counterbalance all those on the opposing side.

On October 22, Nvidia CEO Jensen Huang sharply criticized calls to slow down technological development, labeling remarks about the potential existential threat posed by AI, such as those made by Jacob Coxon, as irresponsible. He asserted that, in the coming years, the technology will not reach a stage capable of bringing about the end of humanity. According to Huang, when AI developers call for the introduction of national or international safety regulations, their true intent is often to circumvent existing laws that could otherwise hold them accountable.

“2030 is not going to be the end of the world. There is 0% chance that’s going to be the end of the world. Scaring people is unnecessary. It is irresponsible. <…> My point is, apply those first. And don’t let this doomsday narrative cause somebody to relieve them of the laws that currently exist.“ Huang said in an interview with CBS News.

It is worth noting that Nvidia produces approximately 70–80% of all chips used to build AI data centers. The company has become the most valuable in the world amid this boom, yet it also serves as the primary supplier to the very entities now calling for a slowdown in research. Furthermore, additional regulation would hinder the emergence of new clients for Nvidia, a factor that carries extra weight as the company’s market share begins to gradually decline.

Flipside of the security argument

Calls to slow development of the technology are, in effect, an attempt to create a new barrier of entry in the AI market. The framework proposed by Anthropic, for instance, includes provisions for permanent external auditors, who would have access to internal risk-assessment tools and processes, as well as voluntary industry standards and various forms of formal regulation. For a large laboratory, these costs are manageable; however, for a startup, which might be required to undergo external vetting before training a model, such requirements could effectively shut down its opportunities.

2018 ACM A.M. Turing Award laureate Yann LeCun characterized Anthropic’s stance as “marketing or lobbying.” He argues that the company is deliberately portraying AI as extremely dangerous in order to alarm governments, even though current large language models are  not at the level of posing a threat to humanity. LeCun also pointed to a different risk: dominant companies may shape regulations in a way that hinders their competitors.

Meanwhile, Nvidia’s revenue is effectively tied to the pace of investment in AI computing. Any slowdown in the industry, regardless of the underlying motives, would lead to a contraction in demand for the company’s products. Every call to “slow down” translates into lost chip sales. Jensen Huang’s call for “responsible optimism” could essentially be rephrased as, “Buy now, the future is bright.”

Furthermore, in mid-October, the IMF released a report on global financial stability. Companies that own data centers now account for 10% of investment-grade bond issuance, up from less than 1% in 2024, and the methods that led to this are a cause for concern. Since 2023, there have been 14 major deals within the industry in which companies act simultaneously as investors, creditors, suppliers, and clients. These involve approximately $300 billion in equity investments and $650 billion in debt financing, with the recipients committing to future purchases of cloud services or AI chips totaling $1.4 trillion. The IMF cites a possible “correction” (effectively meaning stock market crash) in the AI market as one of the key threats to the global.

This implies that the AI boom is propped up by debt and circular transactions. If the projected revenue fails to materialize, the result could be more than just a stock market “correction”; it could trigger a crisis that ripples through various sectors via the debt market. Developers of large language models are calling for a slower pace of development, yet they remain dependent on a continuous influx of capital to fund the training of next-generation models. Meanwhile, Nvidia urges against the slowdown of the technology, fearing a drop in demand. Both sides have a vested interest in keeping the flow of capital alive. However, each one is hoping to direct that flow toward itself.

Scaremongers

In this game, fear of the technology itself becomes a highly valuable “byproduct.” The louder the talk about the end of the world, the more valuable the services of those who promise to prevent it become. The only question is how long one can keep selling fear while producing nothing but debt.

While some sell chips and others sell the fear of AI, it is the very same buyer who foots the bill. Both sides claim to care about humanity, yet each is just trying to grab a larger slice of the pie from their opponents. None of the proposed scenarios offer any good news; society is left to choose which risk it is prepared to pay for.

Source: Rossa Primavera News Agency