Qualcomm Takes Aim at Nvidia's AI Throne with Custom Data Center Chips

Qualcomm is taking its fight from smartphones to data centers.  According to the Wall Street Journal, CEO Cristiano Amon revealed that Qualcomm is developing custom chips with a hyperscale cloud provider, with first shipments expected in the December quarter. On the surface, this looks like a mobile chip giant seeking new growth. But the real story is that Qualcomm is directly targeting Nvidia's most lucrative turf—AI inference—with three chip types: CPUs, inference accelerators, and custom ASICs. **Why this matters:** Nvidia dominates training, but the inference market is fragmenting. Hyperscalers are demanding lower power and better cost-performance. Qualcomm's mobile low-power DNA, combined with IP from its Alphawave acquisition, gives it an edge in squeezing performance within tight power budgets. Qualcomm isn't starting from zero. Last year, it landed Saudi AI startup Humain as its first data center client. But this hyperscaler deal is the real ticket. Once locked in, hyperscalers bring long-term orders, scale, and ecosystem moats. If Qualcomm ships on time in December, it transitions from challenger to serious contender. Of course, Qualcomm is still small. Amon admits the company is in early expansion, far behind Nvidia. But investors should watch growth, not size. Qualcomm's Q2 revenue hit $10.6 billion, in line with estimates, but Q3 guidance missed. Data center revenue isn't yet visible in earnings, but once shipments begin, it could become Qualcomm's most reliable growth driver in years. Meanwhile, Qualcomm isn't ignoring the edge. OpenAI recently partnered with Qualcomm to develop AI chips for smartphones. Smart move: edge AI is a space Nvidia can't easily reach, and Qualcomm's mobile channels and power efficiency are unmatched. If edge AI takes off, Qualcomm stands to benefit massively. **What to watch next:** - **Short-term:** December shipments. Any delay will test market confidence in execution. - **Medium-term:** Follow-on orders from the hyperscaler. This determines whether Qualcomm moves from pilot to primary supplier. - **Long-term:** Can Qualcomm replicate its "Qualcomm tax" model in inference—locking clients with custom chips and collecting IP royalties? **Key signals for investors:** 1. Customer feedback after December shipments. 2. Data center strategy details at the June Investor Day. If Qualcomm proves its chips are more power-efficient and cost-effective than Nvidia's GPUs for inference, Nvidia's moat will show real cracks. Nvidia's pie won't vanish overnight, but Qualcomm has picked up the knife.