NVL72 rack packs 2+ miles of copper — analyst says the copper-to-fiber inflection is near
BenBajarin · x · 2026-09-19
Ben Bajarin points to his March report "Copper to Fiber: The Connectivity Inflection in AI Infrastructure," arguing NPO (network processing optics) will ramp hard first and that this is becoming consensus.
Key data: a single NVIDIA NVL72 rack carries over 2 miles of copper across 5,000 cable runs, 3,000 lbs of cable and 100+ lbs of steel reinforcement. Selling 60,000 NVL72 racks would mean 120,000+ miles of copper — nearly five laps around Earth. Next-gen training requires rack-to-rack distances where copper physically can't carry signals fast enough, and NVIDIA's roadmap has a generation where scale-up goes optical. Network saturation is already wasting GPU time worth billions of dollars; the debate is only about timing.
More from Infra
- ProgramAsWeights: describe an AI function in English, run it offline on CPU — yuntiandeng · 2026-09-19
- Magnitude: free open-source desktop engine profiles your hardware, picks and tunes local models — nickbaumann_ · 2026-09-19
- ContinuityBench: stateful LLM failover keeps 99.2% context vs near-0% stateless — its_vayishu · 2026-09-19
- AI Data Center Power Bottleneck Makes Bloom Energy This Analyst's Top 2026 Stock Pick — Beth_Kindig · 2026-09-19
- Jev classifies the entire Bluesky firehose for animal posts in near real time — danshipper · 2026-09-19
- 28 hashes logged: crowd audits Gensyn's open-1b, first verifiable training run — benfielding · 2026-09-19