Qualcomm buys the runtime, then hands the keys to five rival chipmakers

Modular and Qualcomm used ModCon 2026 in San Francisco to place the Mojo language and its compiler under an Apache 2.0 license, effective immediately, and to extend platform support to six kinds of chips from six different vendors ^[1][2][4]. Qualcomm paid roughly $3.1 billion in stock for a runtime that gets more valuable every time a competitor's accelerator wins the workload, which is the part worth studying ^[8][18].

The sequencing is tight. The deal was announced on June 24 at roughly $3.9 billion in stock and closed on July 28, 34 days later ^[7][23]. Qualcomm's quarterly SEC filing puts the actual value at about $3.1 billion for 18 million shares, since all-stock math moves with the share price ^[8], an implied roughly $172 per share and about 21 percent below the announced headline ^[9][10]. The license came three weeks after close ^[24]. Modular's founder Chris Lattner, who co-created LLVM and Apple's Swift, is now a Qualcomm executive vice president running advanced AI software and platforms ^[11].

The tests had been set in public. Matt Kimball of Moor Insights & Strategy warned that if Modular's openness narrowed toward Qualcomm silicon it would become "a lock-in play wearing open-source clothing" ^[12]. His colleague Bill Curtis named two checkpoints: a shipping Hexagon backend, and whether MAX stayed open to competing silicon ^[13]. According to Patrick Moorhead, who attended at Qualcomm's invitation and whose firm counts Qualcomm, AMD, AWS, Google, Microsoft and Nvidia as clients, both were met ^[21]. Qualcomm's new datacenter parts are built on the Hexagon lineage and now run the Modular stack, with a Google Gemma model working on the AI200 within two weeks of hardware access ^[14]. The MAX license dropped its device usage restrictions, and Modular's announcement commits in writing to optimizing for hardware "including hardware that competes directly with Qualcomm Technologies' platforms" ^[5][15]. Chief executive Cristiano Amon told Moorhead directly: "We did not acquire Modular to make it a Qualcomm-only software" ^[16]. Amon's framing was Android and Kubernetes arriving for AI at the same time ^[17].

Strip the framing and the arithmetic is plain. Five of the six supported vendors are not Qualcomm ^[22]: Nvidia, AMD, Apple Silicon, AWS Trainium and Google TPUs ^[4]. Qualcomm bought day-one software for its own chips from datacenter to edge, a services business, and a seat in the layer that decides where workloads run even when a rival's chip wins ^[18]. Modular Cloud, the inference service, is now generally available ^[3]. Moorhead argues the platform lowers the cost of bringing a new AI chip to market and lowers compute costs for token buyers who can now arbitrate across chips as well as models ^[25][20]. An Apache 2.0 release is also hard to quietly walk back ^[26].

Three things to watch. The alliance program Amon committed to launching by year end is where neutrality either acquires governance or stays a promise ^[5]. Mojo on Windows, with Microsoft's help, tests whether the portability story extends past the datacenter ^[6]. And the honest caveat belongs to the reporting itself: this is one paid analyst's account of a vendor event, and he says real unknowns remain ^[19][21].