AI Supply Chain Industry Data

Measures how much HBM production is crowding out conventional DRAM capacity, the core mechanic behind the memory supercycle thesis. The supply response is now committed but slow: Micron confirmed roughly $50B for two Boise fabs on 20 August, backed by up to $6.2B in CHIPS Act funds, with ID1 targeting first wafer output in mid-2027 and ID2 in late 2028.

The single most-cited capacity constraint on how fast accelerator orders can actually ship.

Tracks the shift from merchant GPU toward in-house accelerator designs at the largest cloud buyers.

Spot pricing, not contract pricing: more volatile and the first place a supply/demand imbalance shows up.

Conventional DRAM, not HBM alone: the clearest read on capacity-reallocation pressure spilling into general-purpose memory. HBM4 economics are stepping up alongside it, at roughly $31-32/GB for NVIDIA-bound parts against $17-18/GB for HBM3e. SK hynix has begun writing long-term agreements without the industry-standard price cap, a structural break from Micron's capped approach.

Faster than the prior generation's 800G ramp at the same point in its cycle.

Higher layer count and low-loss material grade are the main drivers of AI PCB content-value growth.

Now the default assumption for any rack above roughly 40kW density, not a differentiator.

Effectively zero available capacity in the markets hyperscalers most want to be in.

Now longer than the typical data-centre shell construction timeline: power, not chips, is the pacing constraint. High-power transformer lead times have stretched further still, from 24-30 months pre-2020 to roughly five years today, and behind-the-meter gas generation has moved from backup concept to deployed primary power for AI sites.

The single most-watched number in the whole chain, and the scrutiny has moved from the size of the number to how it is funded. Alphabet's first-half capex rose 111% to $98B and the company turned free-cash-flow negative in Q2 for the first time, undercutting the argument that this buildout is comfortably covered by operating cash flow. Including Oracle, the five largest operators guide to roughly $775-800B for 2026.

Still early, but the clearest paying use case in the downstream chain by revenue evidence so far.

Silicon capability is ahead of the replacement cycle it was expected to drive.

Component suppliers are seeing order growth ahead of any single integrator's public shipment targets.