Gavin Baker, a seasoned tech investment expert, highlighted that Google currently holds a cost - effective edge in the AI training arena with its TPU chips. In fact, Google can even run its AI business at a negative 30% profit margin. Nevertheless, the tables might turn soon. NVIDIA is set to deploy its Blackwell chip clusters for training purposes in early 2026, followed by the launch of the GB300 chips.

The shift from the Hopper architecture to Blackwell is no easy feat; it's a complex process. Deployment delays have temporarily given Google an upper hand. Meanwhile, strides in inference technology have bridged the gap until the new chips arrive.

It's anticipated that xAI will roll out the first models trained on Blackwell in early 2026. The GB300 chips will boast “plug - and - play” compatibility. Moreover, Google's cautious approach in TPU design and supply chain strategy could curb its long - term competitiveness. While bringing in MediaTek as a supplier is a wake - up call for Broadcom, diversifying the supply might slow down the evolution of TPUs.

Once Google relinquishes its title as the lowest - cost producer, its strategic computing plans will need a rethink. Continuing to operate at negative profit margins could take a toll on its stock price. After NVIDIA's next - generation chips hit the market, the gap between them and competitors like TPUs is likely to widen even further.

Finally, a word of caution: the market is full of uncertainties, so invest with prudence.