By Akash Sriram
Oct 29 (Reuters) - Alphabet is making billions of dollars selling homegrown AI chips it once reserved for internal use, but building inventory to meet surging demand is draining cash as the tech giant spends heavily on data centres.
Unlike its traditional search and cloud businesses, selling AI chips that compete with Nvidia's requires Alphabet to spend on hardware well before customers pay, putting pressure on cash flow.
The trade-off, outlined by finance chief Anat Ashkenazi on Alphabet's
post-earnings call on Wednesday, offers an early look at the costs of a business that has emerged as one of the company's biggest growth areas as investors fret over AI's threat to its lucrative search advertising business.
"Where previously inventory balances were primarily some of our hardware products, now those TPU (custom chip) systems that are not yet sold will be showing up. So you see cash going out ... it's not the traditional kind of capex but rather a different cycle of cash flow generation," Ashkenazi said.
That shift could pressure Alphabet just as investors grow more concerned about the mounting cost of AI. Its shares fell nearly 11% on Thursday after the company reported another quarter of cash burn and reiterated that spending on AI infrastructure would rise sharply next year.
Still, the processors are helping drive growth.
Alphabet does not break out revenue or margins from the chips, called tensor processing units, but said their sales made a "meaningful" contribution as Google Cloud revenue more than doubled in the September quarter.
Sales of TPU systems, which bundle the chips with networking, storage and software, are expected to rise further in the current quarter and ramp up significantly in 2027, Google Cloud chief Thomas Kurian said.
Alphabet last year said it would supply Anthropic with TPUs worth tens of billions of dollars. Media reports have said the company is also in talks on a similar deal with OpenAI.
"We completely think these are short-term gyrations," said Jeremy Saltzberg, senior investment officer at Check Capital.
"We don't have reason to believe that the margins would be significantly different than what an Nvidia GPU (graphic processing unit) would sell for. And of course, Nvidia GPU margin is incredible."
(Reporting by Akash Sriram, Deborah Sophia in Bengaluru; Editing by Aditya Soni and Anil D'Silva)
