AI data-center spending flows beyond GPUs and operators into networking, optics, power and cooling. Nvidia, AMD, Broadcom and Marvell supply chips; Arista, Nokia and Credo connectivity; Applied Optoelectronics and Lumentum optics; Vistra, Constellation and Bloom power; Vertiv and Modine cooling. Meta's $140B spending this year illustrates the scale, per a commenter with whom the point is agreed. x.com  x.com 

Inference's move toward edge devices and dense, memory-hungry racks should drive demand for specialized chips, links, power delivery and cooling. McKinsey projects inference at 60% of AI computing demand by 2030. Ambarella makes on-device AI chips; Astera Labs links GPUs, memory and networks; Credo supplies short copper connections; Rambus earns from memory interfaces; Navitas reduces power waste; Modine cools dense racks; and Super Micro builds inference-specific server racks at volume. x.com  … Show more

Credo looks exposed as networking becomes AI's next bottleneck and Nvidia locks down optical fiber for scale-up, scale-out and scale-across links. MicroLED is readily available and a good contender for scale-up and potentially scale-out, with many Asian suppliers pivoting into it. x.com  x.com 

Arista's newly announced rack-scale AI fabric adds a catalyst to the interconnect theme. Arista says its open Ethernet platform supports up to 144 accelerators per rack and cross-rack scaling; Arista joins Marvell and Semtech on the list. Lumentum remains listed despite its sale over a personal cost-basis display issue, with a possible repurchase after the wash-sale period. x.com 

Credo's interconnect opportunity grows as campus-scale AI clusters require more links to validate and integrate. The OCP 2026 meeting takeaway is that complex requirements put interconnect at the center, with both copper and optical links set to benefit. x.com 

Arista is losing share in a record market, a bad combination. Evercore's Q2 Ethernet front-end tracker shows the market at a record $7.3B, up 34% year over year, with cloud up 54% and neoclouds up 208%, yet Arista lost 145bps of front-end share. The tracker doesn't show who gained, though IDC ranked Nvidia's own Ethernet business first in data-center switching revenue in Q1, on a different vendor cut. x.com 

Optical networking's broader rollout will be slow because scale-up and scale-out require orders of magnitude more connections than today's limited optical distance links. Supply-chain discussions echo that slow pace. NPO/CPO starts in switches and reaches XPUs and GPUs much later, with new challenges. Optical backplanes will be necessary as racks move to optical, though copper will remain in many implementations. x.com  x.com  x.com  x.com 

Broadcom faces more accelerator share risk than Nvidia, a key caveat for its AI case. Nvidia's strengths span silicon, systems and networking co-design, and software and ecosystem, while Broadcom's revolve around silicon and packaging IP, making it easier to take accelerator share from Broadcom. Broadcom's switch silicon business also faces some disruption risk from optical circuit switches. x.com 

ASIC design is a race to the bottom, very bearish for Broadcom, MediaTek and Marvell as more leading-edge wafer capacity comes online from TSMC, Intel and Samsung. Google sources IP and swaps vendors constantly, showing the design houses add little differentiated value, transceivers aside. AMD, Apple and Nvidia do not outsource because it dilutes margins and most IP is one-size-fits-all. At scale, wafer allocation is the real constraint; the shift may take a year or two. x.com  x.com 

Qualcomm is building the network layer around its HBC memory architecture as an integrated system, attacking AI data movement at several layers at once. HBC cuts movement between memory and compute, Alphawave adds SerDes, optical and die-to-die links, and per AIwire Cornelis Networks will provide networking for Qualcomm's Dragonfly HBC accelerators, with compute in the fabric to cut stalls. Qualcomm's data-center chief Tony Pialis says moving data across the rack is now as important as compute. … Show more