Qualcomm walks into its June 24 Investor Day with two massive acquisition bids on the table, a new chip brand aimed squarely at Nvidia’s inference dominance, and a stock that has climbed more than 6% in the past week. The smartphone giant is trying to become something else entirely, and the price tag so far sits north of $14 billion.
The Tenstorrent Deal: Buying Jim Keller’s Playbook
The headline transaction is the reported $8 billion to $10 billion pursuit of Tenstorrent, the AI chip startup founded in 2016 by silicon architect Jim Keller. Reuters reported on June 15 that the two companies are in advanced negotiations, with QCOM shares jumping more than 4% on the news.
What Qualcomm gets here is not just another chip design team. Tenstorrent builds AI inference accelerators on RISC-V, the open-standard instruction set architecture that sidesteps the Arm licensing layer Qualcomm has relied on for decades. Keller, whose resume includes stints leading chip architecture at Apple, Tesla, AMD, and Intel, has built a team focused on efficient inference silicon that competes directly with Nvidia’s data center GPU stack.
The strategic logic is clear: Qualcomm wants a data center inference product line that does not depend on Arm’s licensing terms and can be customized for the hyperscaler market. Tenstorrent’s RISC-V foundation provides that. The price, however, is steep for a startup that has not yet shipped at hyperscale volume.
Modular Adds the Software Layer
Less than a week after the Tenstorrent news broke, Bloomberg reported that Qualcomm is nearing a separate deal to acquire Modular Inc. for approximately $4 billion. Modular, founded by Chris Lattner and Tim Davis in 2022, builds AI infrastructure software, including the Mojo programming language designed to replace Python in high-performance ML workloads.
If both deals close, Qualcomm will have assembled a vertically integrated AI stack: Tenstorrent’s RISC-V hardware accelerators, Modular’s compiler and software infrastructure, and Qualcomm’s own Dragonfly-branded inference chips. That is a full-stack play that no other non-Nvidia company currently offers.
Dragonfly Takes Flight: AI200 and AI250
The Investor Day is expected to showcase the Dragonfly brand, which CEO Cristiano Amon unveiled at Computex 2026 in Taipei earlier this month. The lineup includes two inference accelerators: the AI200, featuring 768 gigabytes of low-power DRAM, and the AI250, which uses a near-memory computing architecture that Qualcomm claims improves memory bandwidth by up to 10 times while cutting power consumption by 35% to 70% compared with conventional GPUs.
The AI200 is slated for shipment in late 2026, with hyperscaler ASIC deliveries pulled forward after a major customer win that Qualcomm has not publicly named. The AI250 targets a 2027 launch. Both chips sit in the inference-only camp, leaving training workloads to Nvidia and AMD, but targeting what Qualcomm sees as a $2 billion to $5 billion near-term market opportunity in FY2027 and FY2028.
The HUMAIN Agreement and the Gigawatt Ambition
Qualcomm has already secured a supply agreement with HUMAIN, the Saudi Arabian AI infrastructure venture, to support a 200-megawatt data center deployment starting in 2026. HUMAIN plans to deploy Qualcomm-based inference capacity across hundreds of megawatts, which would make it one of the largest single-customer commitments for any non-Nvidia AI chip vendor.
The HUMAIN deal signals where Qualcomm is placing its bets geographically. While US hyperscalers remain the primary target, sovereign AI infrastructure programs in the Middle East and Asia represent greenfield demand that Nvidia cannot easily monopolize, particularly as US export controls create supply uncertainty for GPU-dependent buyers.
What the Numbers Say About the Pivot
Qualcomm’s ambition is to generate a majority of its revenue from non-smartphone segments by the end of the decade. That is an enormous shift for a company whose Snapdragon mobile chips still account for the bulk of its $40 billion-plus annual revenue. The data center push, combined with existing contracts in automotive (BMW, GM) and industrial IoT, forms the diversification thesis Amon will present to analysts.
The market is giving Qualcomm credit, but cautiously. QCOM stock has moved from roughly $195 to above $210 since the Tenstorrent news dropped, but the shares still trade at roughly 17 times forward earnings, a discount to AMD and a fraction of Nvidia’s multiple. The gap reflects Wall Street’s reasonable skepticism: Qualcomm has announced data center ambitions before, notably with its Centriq server chip in 2017, and walked away when the economics did not work.
This time looks different for two reasons. First, the AI inference market did not exist at scale in 2017. Agentic AI, retrieval-augmented generation, and always-on chatbot deployments have created sustained inference demand that favors power-efficient, purpose-built silicon over general-purpose GPUs. Second, Qualcomm is buying its way in rather than building from scratch, which compresses the timeline but raises execution risk around integration.
The Bigger Picture
Qualcomm’s Investor Day arrives at a moment when the AI chip landscape is fragmenting. Nvidia still dominates training and high-end inference, but Qualcomm’s recent deal with ByteDance for custom AI agent ASICs demonstrated that hyperscalers are actively seeking alternatives. Amazon, Google, and Microsoft all have internal chip programs. Broadcom builds custom silicon for several of them.
What Qualcomm is assembling, if it can close both Tenstorrent and Modular, is a differentiated offering: RISC-V hardware that avoids Arm’s licensing overhead, a software stack that speaks to ML engineers, and inference silicon tuned for the power-per-token economics that will define the next phase of AI deployment.
The question is whether Qualcomm can execute. Two multi-billion-dollar acquisitions, a new chip brand, and a pivot away from smartphones represent enormous operational complexity. Amon will need to convince investors that this is not Centriq 2.0 but a structural transformation backed by real customer commitments and a product roadmap that ships on time. The HUMAIN agreement and the pulled-forward hyperscaler ASIC deliveries suggest the demand is there. Whether Qualcomm can meet it is the $14 billion question.