Qualcomm Inc.’s stock jumped 14% in after-hours trading today after it shared a series of updates about its artificial intelligence roadmap.

The company announced plans to acquire an inference software startup called Modular Inc. and previewed two upcoming AI chips. Additionally, Qualcomm significantly raised its fiscal 2029 guidance. The chipmaker now expects its non-handset revenue segment, which includes its data center business, to reach $40 billion.

The company announced most of the updates during an investor event in New York.

Qualcomm plans to finance the Modular acquisition with 19.2 million shares, which are worth $3.92 billion at the chipmaker’s last unaffected closing price. That sum is more than double the valuation Modular received after its most recent funding round. The company has raised $380 million from Alphabet Inc.’s GV, Greylock and others. 

Porting an AI model from one chip to another historically required developers to make extensive code modifications. Modular has built a software platform that automates the task. Qualcomm, which offers a growing lineup of AI chips, might be looking to use the technology to ease the task of porting AI models to its silicon. Making it easier for customers to switch from Nvidia Corp. accelerators could help the company boost product adoption.

Modular’s software also automates several related tasks. It includes AI building blocks that remove the need for engineers to write everything from scratch, which speeds up development. Software teams that don’t require a fully custom AI model have access to several hundred pre-packaged neural networks.

“As agentic AI scales across data centers and edge environments, the industry is moving toward disaggregated, multi-vendor architectures that demand a more open and modern software foundation,” said Qualcomm Chief Executive Officer Cristiano Amon.

At the company’s investor event today, Amon debuted two new data center chips called the Dragonfly C1000 and Dragonfly AI300. The C1000 is a central processing unit designed to power servers that run AI workloads. The AI300, in turn, is a machine learning accelerator compatible with both air- and liquid-cooled racks.

Qualcomm shared few details about the AI300’s architecture. It did, however, specify that the chip is expected to provide significantly higher performance per watt than existing AI accelerators. The company shared a more detailed overview of the C1000, which it says will include more than 250 cores with top speeds in excess of 5 gigahertz.

Qualcomm says the CPU will provide double the performance per watt of existing processors with comparable features. Additionally, the chip is expected to make AI clusters more cost-efficient. That value proposition has won over Meta Platforms Inc., which plans to implement the C1000 in its servers. 

The Facebook parent is adopting the chip through a CPU supply agreement the companies describe as a “multi-year, multi-generation” contract. The wording suggests Meta will also buy the next-generation processors that Qualcomm plans to launch after the C1000. The financial terms of the deal were not disclosed. 

The Meta contract may be one of the reasons Qualcomm raised its sales forecast. The company stated today that its non-handset revenue segment will reach $40 billion in fiscal 2029, an $18 billion increase over its previous guidance. Qualcomm also boosted its adjusted earnings forecast to $18 per share, which is well ahead of the $15.26 predicted by analysts.  

Photo: Qualcomm