Silicon shortage reading (chips bubble)¶

Or, how the world discovered that putting all its eggs in a few Taiwanese baskets might have been optimistic.

Welcome to the semiconductor circus¶

The global semiconductor industry resembles nothing so much as Ankh-Morpork’s clacks system after the Grand Trunk Company discovered that making promises was easier than making infrastructure. Everyone needs chips. Nobody can get enough of them. The supply chain spans continents and relies on technologies so complex that only a handful of companies in the world can manufacture them. What could possibly go wrong?

Turns out: quite a lot. This guide collects resources for understanding the great chips bubble, from supply chain fragilities through AI-driven demand insanity to the uncomfortable reality that your smartphone’s price might rise because data centres are hogging all the memory.

The supply chain: fragile by design¶

These resources examine how the semiconductor supply chain went from “mostly functional” to “one factory fire away from global chaos.”

Understanding the disruptions¶

The Five Biggest Disruptions to the Semiconductor Supply Chain in 2024, Z2Data, 2024. Comprehensive overview of climate-driven disasters, geopolitical tensions, single points of failure, and the uncomfortable reality that 90 percent of goods travel by sea through a handful of critical chokepoints. Discusses fires at Renesas Electronics, ASML, and Wuxi Welnew that triggered supply chain disruptions rippling through multiple industries. Notes that semiconductor lifecycles have shrunk from 30 years in 1970 to a mere decade by 2014, causing 474,000 parts to reach end-of-life in 2023 alone.

A Look at the Semiconductor Shortage Going into 2024, Applied Energy Systems, October 2023. Tracks the shortage from pandemic origins through recovery predictions. Details the CHIPs Act funding and its impact, noting that despite 80 percent increase in job applications and 40 percent increase in internships, three-fifths of needed jobs may stay unfilled. Predicts revenue increase of 98.7 billion dollars in 2024 after a 67.4 billion dollar decrease from 2022 to 2023.

Semiconductor in 2025: Challenges and Opportunities, Gocar Electronic, June 2025. Examines supply uncertainty leading Micron to receive 600 million dollars and NVIDIA to invest 775 million dollars in customer prepayments just to secure components. Notes that fab delays will continue exacerbating supply challenges particularly for sub-11nm chips used in AI and high-end computing. TSMC reported 17 percent revenue increase in Q1 2024, signalling recovery whilst acknowledging ongoing constraints.

2025 Semiconductor Industry Outlook, Deloitte Insights, July 2025. Industry analysis covering generative AI accelerator chips, shift-left chip design approaches, global talent shortage, and need for resilient supply chains amid geopolitical tensions. Discusses TSMC’s Arizona facility delays (originally 2027, now pushed to 2028), CHIPs Act funding, and impact of Trump administration policies on semiconductor manufacturing.

What to Expect in the 2025 Semiconductor Supply Chain, Supply & Demand Chain Executive, August 2024. Notes that whilst industry insiders anticipate normalisation by early 2025, this “new normal” will feature rolling periods of constraint for certain node sizes. Discusses fab construction delays, supply visibility improvements, and the reality that market recovery doesn’t equal market resilience. TSMC’s 40 billion dollar Arizona facility has experienced multiple delays despite receiving 6.6 billion dollars in CHIPs Act funding.

How and When the Chip Shortage Will End, in 4 Charts, IEEE Spectrum, March 2023. Historical perspective showing more than 40 companies increasing capacity by 750,000 wafers-per-month between 2020 and 2022. Notes 17 percent capacity increase for 200-mm facilities through 2024, whilst 300-mm fab equipment spending hit 78 billion dollars in 2021. Points out automotive industry makes up less than 9 percent of chip demand by revenue but employs over 10 million people globally, making shortages politically sensitive.

The geopolitical dimension¶

The Nexperia Chip Crisis, Explained, Z2Data, 2024. Case study in how quickly geopolitics can disrupt supply chains. U.S. added Nexperia to Entity List in September, China retaliated with export controls, and within weeks major automakers faced production stoppages. Nexperia supplies 60 percent of its products to automotive manufacturers including Volkswagen, Toyota, BMW, and Mercedes-Benz. European Automobile Manufacturers Association stated it could take months to find alternative suppliers.

The Semiconductor Market Going into 2025: A Landscape of Opportunity and Challenges, Sourceability, 2025. Examines China’s antimony export restrictions (U.S. companies had only weeks of supply left by September 2024), Netherlands’ semiconductor equipment export limits, and Trump administration’s proposed 10-20 percent tax on imports with possible 60 percent on Chinese goods. Notes potential CHIPS Act overhaul despite bringing over 400 billion dollars in investments to U.S. semiconductor industry.

Will We See Another Semiconductor Shortage in 2025?, Procurement Pro, five days ago. Points out Europe’s dependence on Chinese rare earth elements (79 percent of gallium from China in 2023), with germanium costs increasing 115 percent from 945 pounds per kilogram in Q1 2023 to 2,050 pounds in Q2 2024 after China imposed export controls. Notes Intel put Polish and German factories on hold after 1.26 billion pounds in Q2 2024 losses, pushing supply chain resilience further into future.

The mature node problem¶

Another Semiconductor Shortage May Be Coming, S&P Global Mobility, July 2025. Examines looming shortage in mature nodes (40 nanometres and above) caused by underinvestment. Industry is focusing on advanced nodes (5nm, 3nm) for consumer electronics and datacentres, whilst automotive, industrial, and consumer electronics still need mature process chips. Analog chip demand increasing 23 percent per vehicle by 2026 compared to 2022. Risk of constraints resurfaces by 2025 despite low inventory levels.

The AI gold rush: when demand meets delusion¶

Resources examining how artificial intelligence turned chip demand from “high” to “completely unhinged.”

Market explosion and hype¶

The AI Chip Market Explosion: Key Stats on Nvidia, AMD, and Intel’s AI Dominance, PatentPC, three weeks ago. NVIDIA controls approximately 80 percent of AI accelerator market through CUDA software ecosystem. AMD making strides with MI300 series, whilst Intel’s Gaudi platform faces software challenges. Notes High Bandwidth Memory (HBM) chip demand surging as AI models require massive memory bandwidth. Businesses anticipate rising costs as demand outstrips supply.

AI Chip Statistics 2025: Funding, Startups & Industry Giants, SQ Magazine, October 2025. Global AI chip market projected to grow from 29.65 billion dollars in 2024 to 164.07 billion dollars by 2029 at 41.60 percent CAGR. NVIDIA expected to report 49 billion dollars in AI-related revenue in 2025 (39 percent increase). Data centres expected to consume 52 percent of all AI chips in 2025. Training chips cost fell from 29,000 dollars per unit in 2024 to 25,000 dollars in 2025, whilst inference chips dropped to 470 dollars per unit.

AI Chips in 2020-2030: How Nvidia, AMD, and Google Are Dominating, PatentPC, one month ago. AI chip energy efficiency improved 40 percent year over year between 2020 and 2025. AMD grew AI accelerator market share from 5 percent in 2022 to 15 percent in 2024. Google Cloud TPU adoption grew 120 percent between 2020 and 2024. Microsoft integrated AMD’s MI300 GPU in 2024 to reduce reliance on NVIDIA and gain control over AI processing costs.

Data-Center AI Chip Market – Q1 2024 Update, TechInsights, 2024. Industry analysis noting AI demand fuelling DRAM and NAND shortages whilst major chip mergers reshape landscape. Discusses NVIDIA, AMD, Intel positioning and global policies reshaping semiconductor industry around AI infrastructure.

The circular money problem¶

The AI Ouroboros, The American Prospect, October 2025. Devastating analysis of vendor financing in AI chip market. NVIDIA invests equity in startups whose number one cost is computing power, which they purchase from clouds running mostly NVIDIA chips, returning NVIDIA’s investment whilst booking revenue that wouldn’t otherwise exist. OpenAI secured 850 billion dollars in data centre investments (17 GW capacity, more power than New York City) but doesn’t have money to pay for it. Oracle commitment alone is 60 billion dollars annually for five years. Notes AI bubble is 17 times bigger than dot-com bubble and 4 times the 2008 housing bubble, with 80 percent of 2025 stock market gains attributable to AI spending.

A ‘Seismic’ Nvidia Shift, AI Chip Shortages and How It’s Threatening to Hike Gadget Prices, CNBC, five days ago. NVIDIA’s pivot to LPDDR memory makes it a customer on scale of major smartphone maker, creating “seismic shift for supply chain which can’t easily absorb this scale of demand.” Memory prices expected to rise 30 percent in Q4 2025 and another 20 percent in early 2026. DRAM imbalances of 1-2 percent trigger sharp price increases; current figure hitting 3 percent levels. Xiaomi warns consumers will see “sizeable rise in product retail prices.”

Market leadership and competition¶

Meet the Monster Artificial Intelligence AI Chip Stock That’s Crushing Nvidia and Broadcom in 2025, The Globe and Mail, 2025. AMD stock up 99 percent in 2025 versus NVIDIA’s 39 percent and Broadcom’s 48 percent. AMD expects 40 percent revenue share of server CPU market by end of 2025 with clear path to exceeding 50 percent. Data centre GPU business projected for major performance bump in 2026 with MI450 series. Next-generation Venice CPUs will be 1.7 times more powerful and efficient than current offerings.

The AI Chip War 2025: NVIDIA vs AMD - Complete Market Analysis & Predictions, Sezar Overseas, two weeks ago. Comprehensive analysis of NVIDIA’s Blackwell architecture, AMD’s MI350 series, and competitive landscape. Discusses SK Hynix 12-layer HBM3E achieving 36 GB per stack with 1.2 TB/s bandwidth. TSMC’s CoWoS production expanding from 35,000-40,000 wafers monthly in 2024 to targeted 75,000 in 2025 and 90,000-95,000 by 2026. China’s November 2025 retaliation mandating state-funded data centres eliminate foreign AI chips by 2027.

AI Chip Market Soars: NVIDIA Leads Race, Design News, September 2025. Notes AI infrastructure boom grew over 250 percent year-on-year between 2022 to 2024, but growth from 2024 to 2025 projected at around 67 percent. Omdia expects AI infrastructure spending to peak as share of data centre infrastructure. Market projected to grow from 94.31 billion dollars in 2025 to 931.26 billion dollars by 2034 (28 percent CAGR).

Power consumption: the inconvenient physics¶

Resources examining the uncomfortable reality that AI chips consume staggering amounts of electricity and nobody seems to know where it is coming from.

Data centre energy crisis¶

DOE Releases New Report Evaluating Increase in Electricity Demand from Data Centers, U.S. Department of Energy, 2024. Official report noting data centre load growth tripled over past decade and projected to double or triple by 2028. Lawrence Berkeley National Laboratory analysis shows U.S. electricity demand accounting for data centre expansion, AI applications, manufacturing growth, and electrification.

Crisis Ahead: Power Consumption In AI Data Centers, Semiconductor Engineering, August 2025. AI data centres consuming energy at roughly four times the rate that electricity is being added to grids. U.S. data centres consumed 176 terawatt hours in 2023 (4.4 percent of total), expected to reach 325 to 580 TWh by 2028 (6.7 to 12 percent of all U.S. electricity). Discusses transmission losses, cooling requirements, and four main target areas for improvement.

The State of AI: Global Energy Consumption from Data Centers Is Forecast to Break 1 Petawatt-Hour by 2026, Navitas Semiconductor, November 2024. International Energy Agency predicts total global electricity consumption of data centres will top 1 PWh in 2026, more than doubling 2022 value and equivalent to Japan’s entire electricity consumption. In U.S., data centres responsible for approximately 2 percent of 4.07 PWh total consumption in 2022.

Powering the US Data Center Boom: The Challenge of Forecasting Electricity Needs, World Resources Institute, 2024. Examines wildly varying forecasts. Rystad Energy found over 100 gigawatts of data centre demand coming online between 2024 and 2035 (10 times New York City’s summer peak). Electric Power Research Institute estimates anywhere between 4.6 percent and 9.1 percent of U.S. electricity consumption by 2030, with difference of 200 TWh equivalent to energy consumption of almost 11 million homes. Grid Strategies found total five-year future demand forecasts increased from 38 GW in 2023 to 128 GW in 2024.

Data Centers’ Insatiable Hunger for Power: A Looming Energy Crisis and Economic Shake-Up, FinancialContent, one month ago. Global data centre power demand projected to grow at 16 percent CAGR from 2023 to 2028. U.S. consumption forecast to reach 426 TWh by 2030 (133 percent increase), potentially over 600 TWh accounting for up to 60 percent of total U.S. load growth. Carnegie Mellon study estimates average 8 percent increase in U.S. electricity bills by 2030, with potential increases exceeding 25 percent in high-demand markets. PJM electricity market capacity prices soared to 329.17 dollars per megawatt-day for 2026-2027, over ten times the 2024-2025 price.

Consumer and environmental impact¶

US Data Centers’ Energy Use Amid the Artificial Intelligence Boom, Pew Research Center, October 2025. Data centres accounted for 4 percent of total U.S. electricity use in 2024, expected to more than double by 2030. U.S. data centres consumed 183 TWh in 2024, projected to grow 133 percent to 426 TWh by 2030. Natural gas supplied over 40 percent of electricity, renewables 24 percent, nuclear 20 percent, coal 15 percent. Data centres directly consumed about 17 billion gallons of water in 2023, with hyperscale facilities expected to consume 16 to 33 billion gallons annually by 2028.

Power Hungry Data Centers: The Energy Crisis Ahead, AKCP, April 2025. Global data centres consumed 460 TWh in 2022 (2 percent of total energy usage), predicted to reach 1,000 TWh by 2026. AI servers can consume up to four times power of traditional servers. With average PUE of 1.55 and average U.S. industrial power cost of 0.079 dollars per kWh, estimates 32.785 billion dollars USD annual power costs for data centres (true cost closer to 120 billion USD using global average of 0.12 dollars per kWh).

2024 United States Data Center Energy Usage Report, Lawrence Berkeley National Laboratory, January 2025. Official DOE-commissioned update estimating historical data centre electricity consumption from 2014 with scenario range of future demand to 2028. Provides estimate of total U.S. data centre electricity use including servers, storage, network equipment, and infrastructure.

Data Center Energy Needs Could Upend Power Grids and Threaten the Climate, Environmental and Energy Study Institute, 2024. Notes U.S. held flat electricity consumption from 2009 despite GDP growth, meaning country became more energy efficient. Data centre development is upending this trend with demand climbing 1.8 percent from 2023 to 2024. EPRI estimates data centres could consume up to 9 percent of U.S. electricity generation by 2030. 2024 study found data centres emitted 105 million metric tons of carbon emissions, equivalent to about 2 percent of all U.S. emissions.

The silicon is expensive. The supply chains are fragile. The energy requirements are staggering.