Google’s AI Chip Finally Leaves Basement, Threatens Nvidia’s Lunch Money
KEY POINTS
- •Google has developed its Tensor Processing Units (TPUs) over more than a decade for AI workloads.
- •Morgan Stanley projects sales of 5 million TPUs in 2027 and 7 million in 2028, signaling increased market presence.
- •Major companies like Apple and Anthropic are using TPUs, with Anthropic placing $21 billion in TPU orders recently.
After over a decade of lovingly crafting AI chips like some bedtime project, Google is now letting its awkward tech toddler, the Tensor Processing Unit (TPU), run around the market playground. Led by Groq CEO-turned-TPU-dad Jonathan Ross, these specialized chips are apparently four times better than their puny predecessors and have the cheek to undercut Nvidia's GPUs—originally born for gaming, not world AI domination. With tech giants like Apple quietly training AI models on TPUs and Anthropic dropping a $21 billion order (yes, billion with a B) for Ironwood TPUs, Google might actually pull off a chip coup. Meta’s flirting with TPUs too, but might stay in GPU-land because Nvidia still owns the CUDA software magic that keeps machines less confused. Morgan Stanley predicts Google could sell 7 million by 2028, each batch adding a casual $13 billion in revenue. Looks like Nvidia’s monopoly is being challenged, but tech world insiders expect a GPU-TPU love triangle for a while yet—because nobody's ready to commit, not even Google.
Share the Story
Source: Businessinsider | Published: 12/16/2025 | Author: Hugh Langley