<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>The Machine Behind AI</title><description>Mapping the supply chain from raw quartz to running GPU clusters.</description><link>https://the-machine-behind-ai.pages.dev/</link><item><title>Chapter 0: Executive Summary</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch00-executive-summary/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch00-executive-summary/</guid><description>Four companies and two North Carolina quartz mines gate over 70% of semiconductor production in the $725B AI infrastructure buildout.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 1: The AI Infrastructure Cycle</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch01-macro-framework/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch01-macro-framework/</guid><description>The four largest hyperscalers will spend $725 billion on capex in 2026, up 77% year-over-year, with transformer lead times reaching 210 weeks.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 2: Raw Materials &amp; Industrial Inputs</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch02-raw-materials/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch02-raw-materials/</guid><description>Two mines in Spruce Pine, NC supply 70-90% of semiconductor-grade quartz globally. Hurricane Helene proved how fast a storm can threaten chip supply.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 3: Semiconductor Materials</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch03-semiconductor-materials/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch03-semiconductor-materials/</guid><description>JSR and TOK hold 91% of the global photoresist market. Qualifying a replacement supplier at a leading-edge fab takes 2-3 years.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 4: Semiconductor Equipment</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch04-semiconductor-equipment/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch04-semiconductor-equipment/</guid><description>ASML is the only EUV lithography manufacturer. Its machines cost up to €380M, with optics and laser suppliers clustered within 200km in Germany.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 5: EDA Software &amp; Semiconductor IP</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch05-eda-semiconductor-ip/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch05-eda-semiconductor-ip/</guid><description>Synopsys, Cadence, and Siemens EDA control 75% of the global EDA market. A single 3nm chip tapeout now costs over $50 million.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 6: Chip Design</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch06-chip-design/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch06-chip-design/</guid><description>NVIDIA data center revenue hit $197.3B in FY2026, a 13x increase in three years. Agentic AI is shifting CPU-to-GPU ratios toward 1:1.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 7: Foundries &amp; Fabrication</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch07-foundries/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch07-foundries/</guid><description>TSMC fabricates an estimated 92% of all AI chips at 7nm and below, with 3nm wafers priced at $25,000-$27,000 each.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 8: Memory: HBM, DRAM, NAND</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch08-memory/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch08-memory/</guid><description>HBM needs twice the wafer area of standard DRAM per bit. SK Hynix overtook Samsung as the largest DRAM maker by revenue in Q1 2025.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 9: Advanced Packaging &amp; IC Substrates</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch09-advanced-packaging/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch09-advanced-packaging/</guid><description>TSMC CoWoS packaging is sold out through 2026. Ajinomoto remains the sole manufacturer of the ABF substrate film every AI chip requires.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 10: Networking Silicon, Switches &amp; DPUs</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch10-networking-silicon/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch10-networking-silicon/</guid><description>Broadcom supplies switch silicon in 70-90% of cloud data center Ethernet switches. Ethernet surpassed InfiniBand in AI cluster deployments in 2025.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 11: Photonics, Optical I/O &amp; CPO</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch11-photonics-optical-io/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch11-photonics-optical-io/</guid><description>Lumentum is the only supplier shipping 200G-per-lane EML lasers at volume. NVIDIA committed $4B to Coherent and Lumentum in March 2026.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 12: Connectors, Cables &amp; Network Infrastructure</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch12-connectors-cables/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch12-connectors-cables/</guid><description>AI data centers require 10-36x more fiber than traditional setups. Meta&apos;s $6B deal with Corning is the largest fiber optic agreement in history.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 13: Power Generation</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch13-power-generation/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch13-power-generation/</guid><description>US data center power demand is projected to reach 134 GW by 2030, up from 50 GW in 2024. No SMR is yet operational.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 14: Power Distribution, Transmission &amp; Grid</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch14-power-distribution/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch14-power-distribution/</guid><description>Large power transformer lead times hit 128-144 weeks in Q2 2025. The US manufactures only 20% of its transformer needs domestically.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 15: Thermal Management &amp; Cooling</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch15-thermal-management/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch15-thermal-management/</guid><description>Air cooling hits limits at 30-40 kW per rack. NVIDIA Blackwell racks generate 120 kW, making liquid cooling a physical requirement.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 16: Site Selection, Land &amp; Permitting</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch16-land-permitting/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch16-land-permitting/</guid><description>$156 billion in data center projects are blocked or delayed globally, with 200+ community opposition groups active across 24 US states.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 17: Physical Construction &amp; Commissioning</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch17-physical-construction/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch17-physical-construction/</guid><description>The global data center construction market hit $227.6B in 2025. Modular prefabrication cuts on-site labor by roughly 40%.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 18: Data Center Operators</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch18-data-center-operators/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch18-data-center-operators/</guid><description>CoreWeave IPO&apos;d in 2025 with a $30.1B revenue backlog. North American colocation vacancy fell to 2.3%, down from 9.8% in 2020.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 19: Servers, Storage &amp; Compute Platforms</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch19-servers-storage/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch19-servers-storage/</guid><description>ODM-direct server shipments to hyperscalers grew 112% YoY to $66.8B in Q3 2025, now representing 59% of the global server market.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 20: System Software, Firmware &amp; Orchestration</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch20-system-software/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch20-system-software/</guid><description>NVIDIA acquired Slurm scheduler creator SchedMD and Run:AI for $700M, completing a six-layer vertical software lock-in stack.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 21: Edge Inference, On-Device AI &amp; Defense Demand</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch21-edge-inference/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch21-edge-inference/</guid><description>Smartphones account for 29% of TSMC advanced node revenue. Every mobile chip competes with NVIDIA GPUs for the same scarce 3nm wafer starts.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 22: Cross-Cutting Analysis</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch22-cross-cutting-analysis/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch22-cross-cutting-analysis/</guid><description>Ajinomoto ABF film would halt global substrate manufacturing within 4-6 weeks of disruption, the fastest time-to-impact of any bottleneck.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item><item><title>Chapter 23: Conclusion</title><link>https://the-machine-behind-ai.pages.dev/chapters/ch23-conclusion/</link><guid isPermaLink="true">https://the-machine-behind-ai.pages.dev/chapters/ch23-conclusion/</guid><description>AI intelligence costs are collapsing, but EUV lithography and transformer manufacturing cannot. The physical layer holds durable pricing power.</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate></item></channel></rss>