Qualcomm forecasts $15 billion data center chip sales by 2029, shares soar (updated)

Amazon.com

Yesterday 20:57

Custom chip revenue from two hyperscaler customers will start before year-end, Qualcomm says
Microsoft will use Qualcomm's HBC chips for AI tasks 
Meta will use Dragonfly C1000 CPU designed for AI data centers
Adds details on financial forecast, context
By Anhata Rooprai and Stephen Nellis

    June 24 (Reuters) - 
    Qualcomm QCOM.O said it expects to generate $15 billion in sales from its data center business by 2029 as it moves beyond its core smartphone chips, sending shares more than 12% higher in after-hours trading on Wednesday.
  

    Chief Financial Officer Akash Palkhiwala said at an investor presentation that the data center business will bring in $5 billion for fiscal 2027, with $1 billion coming from the new custom-chip customers.
  

    Qualcomm also said it expects $40 billion in revenue from chips outside its smartphone stronghold by 2029, up from previous estimates of $22 billion, and with handsets only making up a third of its chip revenue by then.
  

    "We will be truly diversified," Palkhiwala said.
  

    Arm Holdings, which supplies underlying technology for many Qualcomm chips, also rose 5% after the forecast.   
  

    Bank of America analysts had earlier said they expect modest revenue of roughly $2 billion to $5 billion annually from Qualcomm's data center push by fiscal 2027-2028.
  

        
  

    META, MICROSOFT AMONG CUSTOMERS
  

    Earlier in the day, Qualcomm said Microsoft MSFT.O and Meta Platforms META.O will use its new AI chips and that it will make custom chips for two other unnamed "hyperscalers."
  

    Qualcomm's shift to AI chips reflects mounting pressure in the smartphone market, which has been squeezed by a memory chip shortage driven by surging demand for AI infrastructure, and major customers such as Apple AAPL.O and Samsung 005930.KS developing chips in-house.
The chipmaker on Wednesday said Microsoft will use its new category of chips that relies on cheap memory chips used in smartphones and laptops, instead of the pricey high-bandwidth chips used by Nvidia and SRAM memory used by Cerebras Systems CBRS.O.
The company calls the new category "High Bandwidth Compute" or HBC.
"That is a tremendous value that we deliver to the industry in terms of performance per cost advantage," said Tony Pialis, Qualcomm's data center chief.
Qualcomm said Meta will use its new CPU called Dragonfly C1000 that it has designed specifically for AI data centers, entering a market where both Arm Holdings and Nvidia are courting customers.
Pialis also said Qualcomm has won two major customers — called "hyperscalers" in the computing industry — for whom it will make custom chips, with revenue starting before the end of this calendar year.
"I have not had to push my way into hyperscale customers; they've been pulling us in," Pialis said, without naming the customers.
     
CROWDED DATA CENTER CHIP MARKET
Qualcomm, which has attempted to boost its data-center business multiple times, is re-entering a fast-growing, but hyper-competitive AI market full of large incumbents such as Nvidia NVDA.O, the newly minted Cerebras and other custom chip options including Amazon's AMZN.O Graviton and Google's GOOGL.O Axion, Bank of America analysts warned in a client note on Tuesday.
Qualcomm said in April that it plans to begin shipping processors and other AI chips for data centers by year-end.
It also said it was working with customers on three kinds of chips: central processing units, inference accelerators, and custom application-specific integrated circuits (ASICs), a segment that has been booming for rivals such as Broadcom AVGO.O and Marvell MRVL.O.
AI inference — running trained AI models — has emerged as a key battleground.


 (Reporting by Anhata Rooprai in Bengaluru; Editing by Sayantani Ghosh and Sahal Muhammed)
 ((Anhata.Rooprai@thomsonreuters.com))