Google's TPUs Could Dethrone Nvidia in AI Chips

Alphabet has been developing its own Tensor Processing Units (TPUs) for years, but the company is now aggressively scaling production and renting capacity to other AI firms. Google's TPUs offer up to 30% cost savings for AI workloads compared to using chips from other hyperscalers, as they are custom-designed for its Gemini AI model.

Google announced a joint venture with Blackstone to deploy 500 megawatts of TPU capacity by 2027, with plans to scale further. This neocloud business model could capture 20% of the AI cloud market by 2030. Other tech giants like Amazon, Microsoft, and even SpaceX are also developing custom processors, signaling a broader industry shift away from reliance on Nvidia's GPUs.

Nvidia currently commands an 86% market share in AI data center chips and enjoys a 74% gross profit margin. However, the rise of custom processors threatens both its market share and pricing power. If more companies adopt Google's TPUs or design their own chips, Nvidia could face significant revenue and margin compression.

Winners in this shift include Google (GOOGL) and other hyperscalers with custom chip programs, as they reduce costs and create new revenue streams. Losers include Nvidia (NVDA) and potentially other GPU-dependent companies. The neocloud model also poses a threat to traditional cloud providers like Amazon (AMZN) and Microsoft (MSFT), though they are hedging by developing their own chips.