Qualcomm acquires Modular: AI infrastructure competition moving towards hardware neutrality and the era of CUDA unlocking?

On June 24, 2026, Qualcomm announced that it had reached an agreement to acquire AI software startup Modular Inc. The all-stock transaction is valued at approximately $3.92 billion to $4 billion. Qualcomm expects to issue up to 19.2 million shares of common stock to Modular equity holders. The transaction is expected to close in the second half of 2026, subject to customary closing conditions and regulatory approvals.

The core significance of this acquisition is that Qualcomm is transforming from a hardware company centered on smartphone chips into a full-stack provider of enterprise AI infrastructure. And Modular's value lies precisely in the fact that it provides a skeleton key to crack NVIDIA's CUDA software lock-in.

Deconstructing Modular's technical assets: The "hardware-neutral" capability of a compiler and inference framework

To understand the strategic weight of this acquisition, it is necessary to first clarify the technical core of Modular.

Modular was co-founded by Chris Lattner — Lattner is the primary author of the LLVM compiler infrastructure and the Swift programming language, and his engineering team has participated in building most of today's AI infrastructure. Modular's core assets can be broken down into two layers:

First layer: Mojo programming language. Mojo is a high-performance programming language designed specifically for AI infrastructure, built on top of the next-generation compiler technology MLIR. It allows developers to write code once and run it efficiently on diverse hardware such as CPUs, GPUs, and TPUs. Mojo has been open-sourced in Modular Platform version 25.3, with over 450k lines of open-source code accumulated.

Second layer: MAX inference framework. MAX (Modular AI eXecution) is an end-to-end AI compiler and runtime inference framework. It supports PyTorch, ONNX, and native Mojo models, providing low-latency, high-throughput inference services on various hardware including NVIDIA, AMD, Apple Silicon, and more. MAX Engine integrates the entire inference path into a single compilation unit, eliminating the overhead introduced by traditional wrapper-based stacks.

The core value of combining these two layers of technology can be summarized in one sentence: hardware neutrality. Modular's unified platform allows AI models to run efficiently on CPUs, GPUs, NPUs, and custom ASIC architectures without rewriting code for each accelerator. For developers and enterprises, this means "build once, deploy anywhere," while reducing total cost of ownership (TCO).

Qualcomm CEO Cristiano Amon said in the announcement: "As agentic AI expands across data center and edge environments, the industry is moving toward distributed, multi-vendor architectures that require a more open and modern software foundation."

Potential impact on Arm's royalty revenue and CPU market competition

The potential impact of this acquisition on Arm needs to be examined from two dimensions.

Royalty revenue dimension: Arm's FY2026 Q4 financial report shows quarterly revenue of $1.49 billion, up 20% year-on-year, a record high. Full-year royalty revenue was $2.61 billion, up 21% year-on-year. Among this, data center royalty revenue more than doubled year-on-year, and Arm's CPU compute share among top hyperscalers has reached about 50%. UBS predicts that Arm's CPU business revenue could reach $26 billion by 2030, with approximately $10 billion being royalty revenue.

Qualcomm's Dragonfly C1000 CPU uses the Arm architecture, has over 250 cores, adopts a chiplet design, and supports PCIe Gen 7 and CXL connectivity. Qualcomm has signed multi-generation CPU contracts with Meta and Microsoft. This means that Qualcomm's expansion in the data center CPU market will directly translate into Arm's royalty revenue growth in the short term — every Dragonfly CPU shipped, Arm receives a royalty share.

CPU market competition dimension: However, in the medium to long term, Arm itself is moving from "architecture tax" to "chip platform." Just six weeks after Arm's first general-purpose AI (AGI) CPU was released, customer demand surged from $1 billion to $2 billion. Arm management predicts that "by 2030, Arm will capture the largest CPU market share."

This means that Arm is transforming from a pure IP licensor into a potential competitor of Qualcomm — in the data center CPU market, Arm both licenses its architecture to Qualcomm for royalty revenue and designs complete CPU solutions for direct sale. This "both supplier and competitor" relationship is a structural tension that cannot be ignored in the evolution of Arm's business model.

Qualcomm vs NVIDIA vs AMD: Divergent paths in the three companies' AI full-stack strategies

Comparing the AI strategies of Qualcomm, NVIDIA, and AMD together, three distinct paths can be clearly seen.

NVIDIA: CUDA ecosystem moat + full-stack vertical integration. NVIDIA's market capitalization has exceeded $5 trillion, and its core barrier is not hardware computing power but the CUDA software platform. CUDA locks millions of developers into NVIDIA's hardware ecosystem — code optimized for CUDA binds workloads to a single hardware architecture. NVIDIA is expanding from training to inference and edge computing, while entering the data center CPU market with its Arm-based Vera CPU. Its revenue visibility from Vera CPU for 2026 has already reached $20 billion.

AMD: ROCm open-source ecosystem + precision strike strategy. AMD has adopted a more targeted "precision strike" strategy, catching up with CUDA through an open ROCm ecosystem, building advantages in key battlefields such as PC, embedded, and developer ecosystems. AMD's Ryzen AI Max/Halo launched in Q2 2026, targeting the developer market at a cost significantly lower than NVIDIA DGX. AMD's stock price in 2026 has risen approximately 150%.

Qualcomm: Hardware-neutral software layer + horizontal edge-to-cloud platform. Qualcomm's strategy is the exact opposite of NVIDIA's — not to create a new software lock-in, but to break existing software lock-in. Through Modular's hardware-neutral compiler, Qualcomm provides developers with a migration path of "change chips without changing code." Combined with the Dragonfly data center product portfolio and the high-speed connectivity IP obtained from the earlier $2.3 billion acquisition of Alphawave, Qualcomm is building a complete architecture from chip to software to interconnect.

The strategic orientation of the three companies can be summarized as follows: NVIDIA does "lock-in," AMD does "replacement," and Qualcomm does "unlock." The risk of Qualcomm's path is that hardware neutrality means customers can also choose not to buy Qualcomm chips. But the opportunity is that if the AI industry indeed evolves toward multi-vendor, decoupled architectures, Qualcomm's "open layer" positioning could occupy a unique niche.

Market data and analyst views

As of June 29, 2026, Bitcoin was reported at $59,641, down 0.5% in 24 hours; Ethereum at $1,574, up 0.2% in 24 hours. Bitcoin's expected decline this quarter is 13%, which would be the third time since its inception that it has experienced two consecutive quarters of decline.

On QCOM stock, Bernstein analyst Stacy Rasgon maintained a hold rating with a price target of $235. BofA Global Research raised its price target from $165 to $195. BofA analysts expect Qualcomm's strategic shift to the data center to generate at least $2 billion in incremental revenue annually by fiscal 2027-2028. Notably, on June 26, Barclays gave Qualcomm a sell rating. Qualcomm's stock price has risen 29.7% year-to-date in 2026.

Conclusion

Qualcomm's $4 billion acquisition of Modular is a landmark event in the shift of AI infrastructure competition from the hardware layer to the software layer. When hardware computing power is no longer the only bottleneck, the deciding factors will be "who can reduce developer migration costs" and "who can provide true freedom of hardware choice." Modular's compiler and inference framework gives Qualcomm the latter, while the Dragonfly product portfolio and customer endorsements from Meta and Microsoft validate the market demand for the former.

The real test of this transaction lies in whether Modular can maintain its credibility as a "hardware-neutral" platform after being acquired by Qualcomm. If Modular gradually becomes Qualcomm's proprietary software stack, its core value will dissipate; if it can continue to serve a diverse hardware ecosystem within Qualcomm's system, it could become a lever to pry open NVIDIA's CUDA monopoly. The answer will gradually unfold after the transaction closes in the second half of 2026.

FAQ

Q1: What is the transaction amount and structure of Qualcomm's acquisition of Modular?

Qualcomm is acquiring Modular in an all-stock transaction, with an expected issuance of up to 19.2 million shares of common stock to Modular equity holders. Based on Qualcomm's recent stock price, the transaction is valued at approximately $3.92 billion to $4 billion. The transaction is expected to close in the second half of 2026, subject to customary closing conditions and regulatory approvals.

Q2: What are Modular's Mojo and MAX respectively?

Mojo is a high-performance AI programming language based on MLIR compiler technology, supporting code to run on various hardware including CPU, GPU, and TPU. MAX is Modular's AI inference framework, providing end-to-end compilation and runtime services, supporting PyTorch, ONNX, and Mojo models, achieving low-latency inference on multiple hardware platforms.

Q3: What impact does this acquisition have on NVIDIA's CUDA ecosystem?

Modular's hardware-neutral compiler allows developers to write code once and deploy it on different chips, directly challenging the software lock-in effect of NVIDIA CUDA. If Modular is widely adopted, enterprises can freely switch between different hardware without incurring high code rewrite costs.

Q4: After Qualcomm acquires Modular, is it positive or negative for Arm?

Short-term positive — Qualcomm's Dragonfly CPU uses Arm architecture, and increased shipments directly translate into Arm royalty revenue. Medium- to long-term potential competition — Arm itself is moving from an IP licensor to a chip designer, and its AGI CPU may compete with Qualcomm in the data center market.

Q5: What are the core differences in the AI strategies of Qualcomm, NVIDIA, and AMD?

NVIDIA establishes software lock-in through CUDA, binding developers to its hardware ecosystem. AMD catches up through the open-source ROCm ecosystem, making precise breakthroughs in PC and embedded markets. Qualcomm provides a hardware-neutral software layer through Modular, attempting to break single-hardware bindings and build an open horizontal platform.