Week in Review
The semiconductor supply chain entered a new phase of constraint this week: not a shortage of chips, but a stranglehold on advanced node capacity. TSMC's dominance at 5nm and below has become the binding constraint for the entire AI infrastructure stack. As demand for AI accelerators surges, TSMC controls the supply, leaving AMD and Intel competing for allocation on scarce 3nm wafers. The bottleneck is structural: equipment constraints, particularly the limited supply of EUV tools, will persist through 2027. Meanwhile, hyperscalers are doubling down on infrastructure spend. Combined capex from the four major cloud providers is now forecast at $725 billion for 2026, with Alphabet raising its 2026 capex guidance to as much as $190 billion and Meta committing $125–$145 billion. The capital is flowing, but the chips cannot follow.
Earnings & Capex Watch
Alphabet: Raised 2026 capex guidance to as much as $190 billion; expects to "significantly increase" capex in 2027.
Meta: Raised 2026 capex forecast to $125–$145 billion (up $10 billion at both ends).
Hyperscaler aggregate: Combined capex from Alphabet, Amazon, Meta, and Microsoft now forecast at $725 billion for 2026.
Intel: Reported "best month ever" following years of foundry repositioning; White House acquired 10% stake to secure domestic advanced node capacity.
Supply Chain Signals
TSMC 3nm allocation: AMD competes directly with Nvidia and Apple for scarce 3nm wafer capacity; any allocation imbalance can disproportionately limit AMD's ability to convert demand into revenue.
EUV tool bottleneck: Limited supply of extreme ultraviolet lithography equipment will constrain advanced node production through 2027; TSMC investing in advanced packaging (CoPoS) to support AI workloads.
Memory shortage deepening: Samsung predicts memory supply shortage will deepen as 2027 orders surge; CPU prices forecast to increase due to low fabrication capacity and prioritization of server CPUs.
China pricing premium: Nvidia B300 server prices in China nearly doubled to $1 million amid supply crunch following chip smuggling crackdown and export curbs; strong AI demand from Chinese firms drives scarcity premium.
Notable Signals
Cooling emerges as secondary constraint: Cooling is reshaping AI deployment timelines, facility design, and site selection; liquid cooling adoption (direct and immersion) accelerating among US colo operators to meet AI density requirements.
China circumvents export controls: Beijing reroutes chip tool imports through Southeast Asia to evade U.S. export restrictions; Washington has imposed increasingly strict controls on semiconductor equipment exports to China.
Energy firm enters AI infrastructure: Major energy company commits $1.5 billion to develop large-scale AI and data center campus (expected to close H2 2026 pending regulatory approval), signaling energy providers' pivot to compute infrastructure.
Logistics volatility, not component shortage: Lead time instability—not component shortages—is the defining cost pressure in Q1 2026; when supply chain timing breaks down, freight shifts to expedited and premium modes.
Bitcoin Macro
Bitcoin's macro backdrop remains anchored to energy and infrastructure spend. Hyperscaler capex hitting $725 billion in 2026 signals sustained demand for power infrastructure and cooling systems—both critical to mining economics. The geopolitical fragmentation of chip supply (U.S. vs. China) and the energy firm's $1.5 billion move into AI infrastructure underscore the structural shift toward compute-intensive economies. Mining profitability remains sensitive to electricity costs and grid access; the data center buildout competing for power will shape miner margins through 2027.
The bottleneck has moved from fabs to foundries. TSMC now controls the pace of AI, and money alone cannot accelerate it.