Qualcomm buys Modular for $4B, acquires Mojo language

Key Takeaways

• Qualcomm is paying nearly $4 billion in stock for Modular, a 150% valuation increase from nine months ago
• Modular's Mojo language lets developers write AI code once and deploy across multiple chip architectures
• The deal signals Qualcomm's aggressive push beyond mobile chips into data centers and edge AI devices

Qualcomm is acquiring Modular, the Silicon Valley startup behind the Mojo programming language, for nearly $4 billion. The deal, announced Wednesday, will see Qualcomm issue up to 19.2 million shares of common stock. Nine months ago, Modular raised $250 million at a $1.6 billion valuation. The acquisition price represents a 150% jump.

Modular's entire team of roughly 150 employees, including co-founders Chris Lattner and Tim Davis, will join Qualcomm. The deal is expected to close in the second half of 2025.

What does Modular actually do?

Modular sells a chip software platform and its proprietary Mojo programming language. The pitch is simple: write AI software once, run it on different chips without rewriting code for each architecture. This directly challenges NVIDIA's CUDA, a closed software system that has locked developers into NVIDIA's GPU ecosystem for over a decade.

AMD's open-source ROCm offers an alternative, but porting code to other chips remains difficult. Modular positioned itself as the Switzerland of AI infrastructure, promising to work across GPUs, CPUs, and custom accelerators from any vendor.

Why Qualcomm wants this technology

Qualcomm generates most of its revenue from mobile device chips. But CEO Cristiano Amon has been pushing hard into new markets. The company is developing 40 different chip designs for AI gadgets: smart glasses, jewelry, earbuds, pins, and watches.

“We believe the future belongs to developer-friendly, horizontal platforms that can run across diverse compute environments and give customers real choice in how and where they deploy AI.”

— Cristiano Amon, Qualcomm CEO

Qualcomm also has data center ambitions. Late last year, it acquired Ventana Micro Systems, a startup building server CPUs based on the open-standard RISC-V architecture. The company is working on custom ASIC designs for data centers, with ByteDance reportedly an early customer.

Modular gives Qualcomm a software layer that could make its chips more attractive to developers. Without strong software tools, even excellent hardware struggles to gain adoption. NVIDIA learned this lesson and built CUDA into an ecosystem that now represents a massive competitive moat.

The founders behind Modular

Chris Lattner and Tim Davis founded Modular in 2022 after leaving Google, where both worked on TPU chips. Lattner's resume reads like a greatest hits of developer infrastructure: he created LLVM, the open-source compiler infrastructure used across the industry; he designed Swift, Apple's programming language; and he briefly led Tesla's Autopilot software program before Andrej Karpathy took over. Karpathy recently joined Anthropic.

Davis co-created TensorFlow Lite at Google, which let machine learning models run on devices with limited computing power. The pairing of compiler expertise with ML deployment experience shaped Modular's approach.

"What makes this team truly exceptional is the complementary partnership between Chris and Tim," said Dave Munichiello, a managing partner at GV, an early Modular investor. "Chris is an N-of-1 human, in that he's bold, visionary, and technically uncompromising."

The awkward position Modular was in

Modular's pitch required it to both partner with and compete against the same companies. It secured partnerships with NVIDIA, AMD, Amazon, and Apple. Simultaneously, it competed with their in-house software stacks.

Lattner told WIRED last year that he believed the software fragmentation problem had to be solved outside Big Tech because it was "structural." The big chipmakers had no incentive to make their chips interchangeable with competitors.

That strategy hit a ceiling. Building a universal software layer requires massive distribution, and distribution in enterprise infrastructure means either becoming a platform yourself or joining one. Modular chose the latter.

What this means for the AI chip market

NVIDIA's dominance in AI hardware rests on two pillars: the raw performance of its GPUs and the stickiness of its CUDA software ecosystem. Competitors have struggled to dislodge either advantage.

Qualcomm buying Modular suggests a new approach: don't try to out-CUDA NVIDIA. Instead, build a software abstraction layer that makes the underlying chip less important. If developers can write code once and deploy anywhere, the chip becomes a commodity decision based on price, power consumption, and availability.

This matters most for edge AI, where Qualcomm already has strong positions in mobile and IoT. A unified software platform across Qualcomm's 40-plus AI gadget designs could simplify development significantly.

Frequently Asked Questions

How much is Qualcomm paying for Modular?

Qualcomm is paying nearly $4 billion in stock, issuing up to 19.2 million shares. This represents a 150% increase from Modular's $1.6 billion valuation nine months earlier.

What is the Mojo programming language?

Mojo is a proprietary coding language developed by Modular that allows developers to write AI software once and deploy it across different chip architectures without rewriting code for each chip.

Who founded Modular?

Chris Lattner and Tim Davis founded Modular in 2022. Lattner previously created LLVM and Apple's Swift programming language. Davis co-created TensorFlow Lite at Google.

Why is Qualcomm acquiring Modular?

Qualcomm wants to expand beyond mobile chips into data centers and edge AI devices. Modular's software platform could help developers build for Qualcomm's chips more easily, competing against NVIDIA's CUDA ecosystem.

When will the Qualcomm-Modular deal close?

The acquisition is expected to close in the second half of 2025.

Source: Feed: Artificial Intelligence Latest / Lauren Goode