Qualcomm bets its AI inference chips can chip away at Nvidia before 2029 – Startup Fortune

Qualcomm bets its AI inference chips can chip away at Nvidia before 2029 – Startup Fortune

By Dave Barr
Publication Date: 2026-10-30 12:50:00

Qualcomm is building a data center business around its AI accelerators, Dragonfly CPUs, custom silicon and connectivity products, backed by commercial relationships with HUMAIN, Amazon and Meta. The next test comes November 4, when Qualcomm reports its fiscal fourth-quarter results.

Qualcomm reports its fiscal fourth-quarter and full-year 2026 results after the close on November 4, and a meaningful part of the discussion is likely to focus on its expansion beyond phones. Investors will be watching its developing data center business and whether its commercial relationships can support the revenue targets it set in June in a market Nvidia still dominates.

The chips in question are the AI200, due in 2026, and the AI250, slated for 2027, both unveiled back on October 27, 2025, alongside server racks built around 768GB of memory per card, more than Nvidia or AMD currently offer on comparable hardware. Qualcomm isn’t chasing the training workloads that made Nvidia’s H100 and Blackwell chips indispensable. It’s going after inference, the business of running an already-trained model in production, generating text, powering chatbots, serving real-time applications. That’s a deliberate, narrower bet. It’s also the part of the AI compute market growing fastest as more companies move from building models to actually deploying them.

The first announced customer was HUMAIN, the Saudi state-backed AI venture, which committed to deploying 200 megawatts of Qualcomm AI systems…