- Essays··5 min read
The Interconnect Decision That Moved Upstream
CoWoS packaging capacity — not wafer fabrication — has become the binding constraint on AI hardware, and the shortfall has reorganised how clusters are designed. Network architects who once specified fabrics after compute teams locked the BOM now sit in allocation meetings, because the interconnect bandwidth budget and the packaging timeline are the same constraint. Memory, topology, and fabric, previously sequential decisions, now collapse into one.
ai-infrastructureadvanced-packaginghbminterconnectRead - Essays··6 min read
The Interconnect Nobody Can Buy
UALink 2.0 was ratified in April 2026; no production-ready silicon exists. For procurement decisions targeting Q1 2027 delivery, that leaves NVIDIA NVLink over InfiniBand or RoCE on Ethernet — each with known performance trade-offs, HBM supply constraints, and operational costs that never appear on the CapEx comparison slide. The interconnect you want and the silicon you can order are not the same thing.
ai-infrastructureinterconnectnvidiaualinkRead - Essays··10 min read
The Specification You Can Download Is Not the System You Can Buy
A ratified specification is not a commissioned cluster. UALink 2.0 promises 1,024-accelerator pods at half the TCO of NVSwitch, but evaluation hardware is not expected until late 2026, and your purchase order is due in August. The more productive question for enterprise finance is whether InfiniBand or Ethernet closes your deployment faster than a roadmap that has not yet touched silicon.
ai-infrastructuregpu-clustersinterconnectdata-centre-networkingRead - Essays··9 min read
China's exascale push against GPUs · what LineShine's interconnect says about post-NVLink fragmentation
LineShine, which debuted atop the June 2026 TOP500, is the first system to exceed two exaflops of sustained double-precision performance without GPU accelerators. It was built under US export controls from indigenous Arm processors, on-package HBM, and a custom fat-tree fabric called LingQi. The ranking is the least of it: China has closed the interconnect gap that sovereign exascale programmes could not paper over, and the scale-up fabric market is now genuinely fragmented.
supercomputinginterconnectai-infrastructurechina-techRead - Essays··7 min read
Copper ran out of road, and nvidia just bought the on-ramp to what's next
Two press releases, 48 hours apart, told you more about the next decade of AI infrastructure than any keynote. Copper ran out of road, and NVIDIA just bought the on-ramp to what comes next.
ai-infrastructureco-packaged-opticsnvlink-fusioninterconnectRead