Unlock Editor’s Digest Lock for Free
FT editor Roula Khalaf will select your favorite stories in this weekly newsletter.
On the hammer, everything looks like a claw. For the manufacturer of chips that power artificial intelligence, everything looks like a “token.” This is an industry term for the unit of information that AI models incorporate and expose. Nvidia founder Jensen Huang believes the world is at the pinnacle of a token explosion. His company’s $3.3TN market capitalization depends on it.
According to LSEG, Nvidia reported revenue of $44 billion in the three months ended in April. The company remains the biggest beneficiary of the AI boom, with little competition when it comes to cutting-edge chips, reflecting its gross profit of 71%.
It’s not without hiccups. Nvidia has won a $4.5 billion prize as a result of the frozen curb from China. Nvidia had already been forced by the White House to sell only a smaller version of the AI chip to the People’s Republic. If Nvidia could not reuse parts that weren’t sold elsewhere, Writedown would have been a big deal.
Luckily, when the door closes, the window opens. Sales of the company’s new Blackwell AI chips are expanding and appear to be faster than expected. For example, analysts at Morgan Stanley had expected the current quarter, which ends in July, to bring in $43.5 billion in revenue after losing $5 billion in forgotten shipments to China. Instead, the company says it expects $45 billion even after stripping its $8 billion of mass sales.
Furthermore, new waves of AI “factories” are erupting, filling holes in the Chinese shaped form. Large tech companies, from Alphabet to Oracle, are already planning mammoth data centres, and the government is currently in place with the law. Saudi Arabia, the United Arab Emirates and Taiwan are all loaded into state-backed “sovereignty” AI. Huang is traveling to Europe next week. This is a trip that may produce more grand declarations.
Next is token growth (meaning even more chip demand) from companies finding practical uses for increasingly complex models. The data center construction, strategist at Apollo Global Management Note, added 1 percent point to GDP in the first quarter. The launch of a “inference” bot that thinks, checks, and double-checks before answering will use 1,000 times more question data than the simpler previous AI models.
Would they all do so using NVIDIA chips? For now, it’s almost the only game in town. The closest US rival AMD has only earned a sixth of revenue. A startup backed by ambitious Abu Dhabi, Celebras starts off at a low base. And that fat gross profit gives Nvidia plenty of firepower to compete for the price if it becomes necessary. If all tokenizations are still at their current pace, Nvidia will need to do more.
john.foley@ft.com
