March 4 (Reuters) - Chip designer Broadcom on Wednesday projected its artificial intelligence chip revenue would exceed $100 billion next year, signaling surging demand for custom chips in a market dominated by Nvidia.
The company's share price rose nearly 5% in extended trading after it also announced a new share repurchase program of up to $10 billion through the end of the year.
Big Tech firms such as Alphabet, Microsoft, Amazon and Meta are expected to spend at least $630 billion to build AI infrastructure this year, boosting demand for chips, servers, storage and networking equipment from companies like Broadcom.
"Our visibility in 2027 has dramatically improved. Today, in fact, we have line of sight to achieve AI revenue from chips in excess of $100 billion in 2027," CEO Hock Tan said on a post-earnings call.
The company expects second-quarter revenue to be about $22.0 billion, above analysts' average estimate of $20.56 billion, according to data compiled by LSEG. It projected AI chip revenue of $10.7 billion in the quarter.
Broadcom, which supplies semiconductors and infrastructure software, does not typically design entire AI chips itself. Instead, it works with clients such as Google to develop their tensor processing units (TPU) and ChatGPT-maker OpenAI on their in-house custom processors.
The company's engineers help translate an early design into the physical layout of a chip that can be fabricated by manufacturers like TSMC.
"Broadcom's guidance for the April quarter and into 2027 was very encouraging. While the company typically only guides one quarter at a time, the visibility into results more than a year out is an indication of significant growth in demand," said D.A. Davidson analyst Gil Luria.
Tan said Broadcom is expected to deliver 1 gigawatt's worth of TPUs for AI startup Anthropic in 2026, with demand rising to 3 gigawatts in 2027. Broadcom aimed to ship OpenAI's first AI chip in 2027 and deliver over 1 gigawatt's worth of the chips.
The chip volumes for major customers Broadcom disclosed on Wednesday showed that it is signing deals approaching the size of those at major AI chip companies such as Nvidia, which disclosed 5 gigawatts of sales to OpenAI last week, and Advanced Micro Devices, which has signed deals of up to about 6 gigawatts with both OpenAI and Meta.
Another revenue source for Broadcom is Meta, with Tan saying reports of a slowdown in the social media company's AI chip business — part of that company's moves to design more of its own custom silicon and reduce reliance on external suppliers — were mistaken.
"Meta's custom accelerator MTIA roadmap is alive and well," Tan said, referring to Meta's custom chip. "We're shipping now."
Broadcom said last month that it expected to sell at least 1 million chips by 2027 based on its stacked design tech, marking a new product and a sales target that could represent a revenue stream potentially worth billions of dollars.
Growth in Broadcom's infrastructure software segment slowed to around 1% at $6.80 billion in the first quarter, compared with analysts' expectations of 2.6% growth to $6.88 billion.
Last month, AI chip designer Nvidia posted better-than-expected results for the January quarter, and forecast current-quarter revenue above market estimates.
Nvidia dominates the accelerator market, providing the chips essential for training and running AI models. However, Broadcom has emerged as a strong contender, offering its own specialized, custom-designed chips as an alternative.
Broadcom said its first-quarter revenue rose 29% to $19.31 billion, beating analysts' average estimate of $19.18 billion. Its adjusted earnings per share of $2.05 exceeded estimates of $2.03.
Its AI revenue more than doubled to $8.4 billion in the quarter ended February 1, driven by demand for custom AI accelerators and networking.
(Reporting by Jaspreet Singh in Bengaluru; Editing by Shinjini Ganguli, Stephen Coates and Alan Barona)
