Warren Buffett's Berkshire Hathaway invested in Google in Q3 2025, coinciding with SemiAnalysis reporting on Anthropic's TPU purchases. Buffett cited understanding Google's TPU v5p architecture, comparing it to railroads, as key to his investment decision.
Key facts
- Berkshire Hathaway invested in Google in Q3 2025.
- TPU v5p has 8,960 chips in a 3D torus.
- ICI links run at 4.8 Tbps per chip.
- Optical circuit switches reconfigure mid-job.
- SemiAnalysis linked TPU buys to Anthropic.
Warren Buffett's Berkshire Hathaway first invested in Google in Q3 2025, coincidentally the same time that SemiAnalysis called out a huge increase in TPU purchases from Anthropic. In Q1 Berkshire added more to their position. Buffett actually said this about his own diligence into Google's AI Infrastructure supremacy: 'I don't buy what I can't understand. So Greg sat me down with the TPU v5p spec. 8,960 chips wired in a 3D torus, every chip talking to six neighbors over ICI links at 4.8 Tbps a pop, wraparound rails so no chip is ever at the end of the line, and optical circuit switches throwing the junctions mid-job to carve out whatever submesh your sharded matmul needs. Add in ring all-reduce running consists in both directions along each torus axis, collective permutes shuffling shards between sidings, and bandwidth-optimal SPMD partitioning across the data, model, and pipeline dimensions. Folks, it's just BNSF. Six neighbors, scheduled consists, a yardmaster throwing switches, trains that never stop. And I've been understanding railroads since 1942.'
The Unique Take
This is not a typical Buffett tech bet. The Oracle of Omaha, known for avoiding what he doesn't understand, dove into the technical architecture of TPU v5p—a 3D torus with 8,960 chips, 4.8 Tbps inter-chip links, and optical circuit switches. His analogy to BNSF railroad operations reveals a structural insight: Google's TPU system is a massive, deterministic network, akin to a rail yard, where data flows like trains on scheduled consists. This contrasts with the common view of AI infrastructure as abstract compute; Buffett sees it as a physical, logistical system.
Technical Depth
TPU v5p is Google's latest custom chip for AI training and inference. The 3D torus topology ensures low-latency communication, with each chip connected to six neighbors via ICI links at 4.8 Tbps. Optical circuit switches allow dynamic reconfiguration of the network mid-job, enabling efficient sharded matrix multiplications. Ring all-reduce operations run in both directions along each torus axis, optimizing gradient synchronization across distributed training. Collective permutes shuffle shards between sidings, and bandwidth-optimal SPMD partitioning handles data, model, and pipeline parallelism. This architecture is designed for massive scale, supporting up to 8,960 chips in a single pod, as [SemiAnalysis] reports.
Market Implications
Berkshire's investment signals confidence in Google's AI infrastructure leadership, particularly as Anthropic (a major TPU customer) scales its models. The timing—Q3 2025—suggests Buffett's team saw the TPU ecosystem as a moat, not just a chip. With Q1 2026 additions, the position is growing. [According to @SemiAnalysis_], the TPU purchases from Anthropic were a key signal.
Key Takeaways
- Berkshire Hathaway invested in Google in Q3 2025, after Buffett studied TPU v5p architecture.
- He compared it to railroads, citing 8,960 chips and 4.8 Tbps links.
What to watch
Watch for Berkshire's Q2 2026 13F filing to reveal further Google position changes. Also track Google's TPU v6 announcements and Anthropic's next model release, which could validate or challenge the TPU investment thesis.
