Qualcomm hasn't been able to manufacture desktop-class ARM devices yet (see SQ1 and X13 ARM edition), and the GPU IP doesn't really hold a candle to what else is out there either. They are however using significantly less power than say, x86 SoCs and GPUs from AMD, Intel or Nvidia.
Nvidia on the other hand has plenty of IP for GPUs and AI workloads that is proven to have high performance, even on the Tegra scale of things. The Jetson product line as an example does still outperform whatever Qualcomm has come up so far. But they are consuming more power and don't even come close to the same thermal envelope.
So Nvidia needs to reduce their energy needs, while Qualcomm needs to create actually performant devices for real-time AI workloads. That's why.
Nvidia on the other hand has plenty of IP for GPUs and AI workloads that is proven to have high performance, even on the Tegra scale of things. The Jetson product line as an example does still outperform whatever Qualcomm has come up so far. But they are consuming more power and don't even come close to the same thermal envelope.
So Nvidia needs to reduce their energy needs, while Qualcomm needs to create actually performant devices for real-time AI workloads. That's why.