Raquel Ditchfield@redinvestments · 6mo Nvidia’s earnings reinforced one major theme: AI demand is still massive. But the market is starting to look beyond just “more GPUs.”
A few key takeaways:
• Nvidia acknowledging ASIC competition is important. Companies like Google, Amazon, Microsoft, and Meta are increasingly designing custom AI chips to reduce dependence on Nvidia and lower inference costs.
• That said, Nvidia still has a huge moat through CUDA, networking, software ecosystems, and developer adoption. GPUs remain the gold standard for training frontier AI models.
• The shift from “training AI” to “running AI” (inference) could reshape the semiconductor landscape. Efficiency and cost-per-query matter more during inference, which may benefit CPUs, ASICs, and lower-power accelerators.
• Nvidia entering the CPU/server ecosystem is bullish long term. It positions them to become a full-stack AI infrastructure company rather than just a GPU seller.
• AMD and Intel could benefit if inference workloads become more distributed and cost-sensitive, but Nvidia is unlikely to give up dominance easily.
• Physical AI / robotics may be the next multi-trillion-dollar narrative. Autonomous systems, humanoid robots, factories, and edge AI all require massive compute infrastructure — an area where Nvidia is positioning aggressively.
• One thing investors should watch closely: margins. Competition may not kill Nvidia’s growth, but it could pressure pricing power over time.
My overall view: The AI boom is probably entering its second phase. The first phase rewarded GPU suppliers. The next phase may reward the companies enabling scalable, cheaper, real-world AI deployment.
Potential beneficiaries beyond Nvidia:
Advanced Micro Devices, Inc.
Intel Corporation
Broadcom Inc.
Taiwan Semiconductor Manufacturing Company Limited
Super Micro Computer, Inc.
Marvell Technology, Inc.
Serve Robotics Inc.
SCHQ (if rates fall and risk appetite broadens)
One thing I’d disagree with slightly in the post: CPUs probably won’t “replace” GPUs for AI inference broadly. More likely, we’ll see a hybrid ecosystem of GPUs + ASICs + CPUs depending on the workload.