Миллиарды в GPU: Дженсен Хуанг объяснил, почему ИИ-пузырь не лопнет

Every time financial reports from tech giants show yet another tens of billions of dollars spent on 'hardware,' the market nervously twitches. Investors search for signs of a bubble, comparing the current AI boom to the dotcom crash of the early 2000s, and ask themselves: when will these crazy spending sprees finally end?

NVIDIA CEO Jensen Huang decided to calm the panic-stricken, claiming that we are not at the peak of madness, but at the very beginning of a large-scale restructuring of the global IT landscape that will last almost a decade.

According to Huang, the current level of capital investments in data centers is not merely justified, it is absolutely sustainable.

We are accustomed to measuring success in quarters and annual reports, but NVIDIA plays the long game.

Jensen forecasts that the construction of new AI infrastructure will stretch over the next seven to eight years.

This is not a temporary spike of interest in chatbots, but a fundamental replacement of old computing power with new power capable of working with generative models and accelerated computing.

The old world, built on central processors, gradually fades into the past, making way for an architecture where GPU reigns.

Skeptics from the banking sector often point out that NVIDIA's clients — Microsoft, Google, Meta, and Amazon — could 'overeat,' by buying too many chips in advance.

However, Huang insists that demand for AI solutions remains 'simply boundless.'

AI has ceased to be a toy for geeks and has become a powerful tool that genuinely benefits business right now.

The spread of these technologies is occurring at such a pace that current production capacity can barely keep up with orders.

According to the company's head, AI-based applications have become so functional that their deployment in the corporate sector has taken on an avalanche character.

It is important to understand the context of this statement. NVIDIA is currently in a unique and simultaneously precarious position where it is both the primary beneficiary and the chief evangelist of this new technological religion.

If servers were previously universal, they are now becoming specialized data factories.

Huang calls this shift inevitable and logical.

Those who today invest billions in Blackwell chips are not simply buying expensive hardware; they are reserving themselves a place in the future economy where computational power will become the primary resource, akin to electricity in the 20th century.

Of course, there is always a touch of marketing optimism in Huang's words — he desperately needs to maintain investors' faith in endless growth.

But if you look at the numbers, the arguments become more convincing.

When a cloud provider spends a billion on infrastructure and immediately leases these resources to startups training new models, this is not 'burning money,' but a working business model.

The entire world is now turning into one giant supercomputer, and NVIDIA has assumed the position of chief architect and supplier of building materials.

Analyzing the situation, one can conclude that Huang is essentially announcing the end of the era of 'cheap and slow' cloud.

We are entering an era where efficiency is measured not by the number of servers, but by the speed of token processing.

If the forecast of seven to eight years is correct, we will see several more generations of architectures, each more powerful than the last, and bubble talk will fade away on its own once AI services begin generating direct revenue comparable to the costs of their creation.

The key takeaway: Jensen Huang makes it clear that chip shortage and high prices are the new norm for the coming years.

Are competitors ready to wait eight years for the market to saturate, or will they find a way to throw a wrench in NVIDIA's plans sooner?

At the moment, it seems that the 'green giant' confidently controls the pace of the game.