Four major US companies plan to spend $725 billion in 2026 alone on data centers, chips, networks, and power plants. These companies are Microsoft, Amazon, Alphabet and Meta. The previous year, the same companies spent a record $410 billion, representing a nearly 77% increase over the past 12 months.
Hot Topics
Standard & Poor's Global estimates that the projected figure for 2027 is approximately $878 billion. This spending has one stated goal, which the industry calls achieving artificial general intelligence (AGI), meaning an intelligent system that doesn't master a specific task but rather performs general human cognitive work.
On September 6th, Jensen Huang, CEO of Nvidia, the world's largest chipmaker, announced that OpenAI had reached that milestone after the launch of its latest model, the Astra. But how do we know he's right?
The question may seem strange, but in reality, there is no universally agreed-upon answer. This isn't because definitions are nonexistent—some exist, are precise, and even have a test that can be administered on any system at any time. However, companies that spend more than others operate by a different kind of definition, one that no one can test or declare anyone a failure on.
In the absence of such a test, the volume of spending replaced the test. As long as the figure was rising, fears of the bubble bursting were postponed for a few more weeks; in other words, the proof of the road's soundness became the amount of money spent on it.
Here we are trying to understand one reason for this absence: not why the precise definition of the term artificial general intelligence has been delayed, but why that definition has not been requested at all, and why it has remained an obscure goal until now.
Around the 17-minute mark, Altman was asked about the difference between his work today and how he worked 10 years ago. He replied that what he was doing now was the most important work he could imagine, and that "we are now, sort of, at the point of technological singularity." The term singularity is borrowed from physics and mathematics, where it refers to a point at which equations cease to yield meaningful results, such as the center of a black hole. The American computer scientist and novelist Vernor Vinge coined its modern meaning in the early 1990s, defining it as the moment when a machine becomes capable of improving itself, with each generation designing a smarter one, until the pace of change exceeds human ability to predict or control it. Crucially, the original definition describes an event occurring within the machine, which, in principle, can be verified by observing it.
"Uniqueness begins when a machine writes a beautiful paragraph, then we ask when it will write a novel, and wonders turn into routine, then become the minimum acceptable."
Then he defined the word in the same article, saying that uniqueness proceeds in this way: we are impressed that a machine can write a beautiful paragraph, then we ask when it will write a novel, and wonders turn into routine, then become the minimum that can be expected or accepted.
Tags:
technology
