GOOGL's $250B AI Chip Opportunity Takes Aim At NVDA — Beth Kindig Says Google Is Going After ‘One Of The Most Profitable Hardware Businesses’

• Analyst Beth Kindig says Google's TPU strategy could challenge Nvidia's dominant AI hardware business. • New TPU 8i and 8t chips offer up to 80% and 170% better performance-per-dollar than prior models. • Selling TPUs to third parties like Anthropic could become a major new revenue stream, potentially worth $250B. IO Fund analyst Beth Kindig says Alphabet's TPU chips could challenge Nvidia not by being faster, but by being cheaper to run, attacking the cost-per-token economics straining Big Tech's AI spending. Google booked its first third-party TPU revenue in Q2, and its new TPU 8i and 8t systems offer up to 80% and 170% better performance-per-dollar than the prior Ironwood generation for inference and training, respectively. Kindig sees the bigger opportunity in selling TPUs directly to outside customers rather than only through Google Cloud, shifting infrastructure costs onto buyers while creating a second revenue stream. Anthropic alone is expected to deploy a gigawatt of TPUs in 2026, scaling to 15 gigawatts by 2028 — a trajectory Kindig says could eventually rival Google's Search business in scale. She also flagged Google's ability to network 134,400 TPUs across superpods, which could matter more if Nvidia faces delays with next-gen systems. GOOGL is up 2% year-to-date versus 13% for NVDA.