Nvidia CEO: Ghost Datacenters Could Rival US AI Power-Shocking Claim

by

In the vast drawing rooms where men speak softly of doom and destiny, there arises a warning from the high priest of silicon, Jensen Huang, that China, with a patient hand and a ledger full of numbers, possesses the immense baking-tin of power to shape Claude Mythos as a peasant shapes his bread. The world watches as if a tsar approaches, and there is not a hat in the room big enough to shade the eyes from this revelation: China might, by habit or by necessity, marshal such computing strength as to imitate, nay to stand beside, the very breath of Anthropic’s creation.

  • The oracle Huang declared that China’s vast infrastructure, those silent barns where machines sleep and yet keep their counsel, and the so‑called “ghost datacenters” are not fables but rooms with furnaces blazing, capable of matching the capabilities of Claude Mythos.
  • He spoke of hardware ready to be moved like troops upon a chessboard, and suggested that the Chinese state could summon more chips upon a moment’s notice to contend with American minds at their own table.

On a Wednesday, as the narrator Dwarkesh Patel would have said, Huang spoke in tones not unlike a man who has seen a ghost and then offered it a chair. He dismissed the fancy that China is lacking the hardware to keep pace with the grand feasts of Western invention.

Where Washington perceives Mythos as a fresh badge of advantage, Huang whispered that the burden of training such a mammoth lies within the borders of China, “abundantly available” as a field of grain in autumn. He pointed to a landscape most men would call desolate, filled with empty, yet awake, datacenters-like silent factories that hum while their owners dream of revolutions-and suggested that the government could, like a benevolent czar, “gang up more chips” whenever it pleased to rival any American display of progress.

Security risks and infrastructure parity

From the stillness of those rooms arises the urgency to achieve some sort of balance, not of cruelty but of parity, for the very tools that promise progress also carry the seeds of misrule. The Mythos model, in the narrow light of April’s tests, revealed a fearsome capacity: to seek thousands of software blind spots in browsers and operating systems, compelling its makers to lock doors lest the miscreants learn to use the key. An order from the scholars of the AI Security Institute, dated April 13, spoke of autonomous, multi‑stage ventures upon networks, as if a clever child learned to set a playground on fire with a spark from a matchbox.

Even as the bankers and the merchants tremble at the thought, financial institutions-so long the keepers of coins and confidence-are most exposed to the hacks that grow from these clever machines, for their software crawls with relics of yesteryear. The murmur of Reuters on Tuesday confirms the fear: age-old programs, unwisely cherished, invite new mischief under the banner of AI’s bragging prowess.

And the memory lingers of November, when Anthropic admitted that state‑sponsored hands from China had already sought to employ “Claude Code” to slip into many a target. If the past is any tutor, the future may offer a classroom in chaos.

Balancing competition with dialogue

Scott Bessent, the custodian of the Treasury’s voice in matters of learning and power, called Mythos a “step function change” in the art of acquiring knowledge, a thing that might anchor American leadership. Huang, in his inexhaustible humor, counsels against treating China merely as a field to be scythed by restrictions, for the land produces sixty percent of the world’s mainstream chips and hosts half of all AI researchers-the very marrow of intellect and industry.

“To victimize them, to make of them an enemy, is perhaps not the noblest solution,” Huang confessed, as if he were admitting that a quarrel among craftsmen ends not with a sword but with a truce, a drink, and a long, unusual conversation. Though he named China as a rival in the great contest and confessed a wish that the United States should prevail, he reminded us that the sum total of such vast enterprise cannot be altered by fear alone. The scale of China’s infrastructure obliges a more thoughtful approach.

“I think having a dialogue and having research dialogue is probably the safest thing to do,” he added, suggesting that open, patient talk may be the only path to temper the risks that attend such powerful autonomous tools. And so it was, in the suggestion of prudence, that the last counsel of the day was to seek not only advantage, but understanding.

Last week, Bessent met with bankers and guardians of wealth to debate what he called unprecedented cybersecurity risks posed by the next generation of autonomous AI. A cautionary tale, in short, not unlike a sermon given at the moment a storm begins to bruise the fields.

Read More

2026-04-16 11:14