AI’s Taiwan problem: Why Anthropic is worried about its chip supply

Artificial intelligence company Anthropic has warned that its dependence on Taiwan for advanced chips could threaten its ability to develop and deploy AI models, highlighting a supply chain risk that money and investment in data centres alone may not resolve.

In its registration statement filed with the US Securities and Exchange Commission (SEC) on Monday, Anthropic identified the concentration of advanced semiconductor manufacturing in Taiwan, primarily at Taiwan Semiconductor Manufacturing Company (TSMC), as a material risk to its business.

The company said disruptions caused by geopolitical tensions, military conflict, natural disasters or interruptions to critical infrastructure could severely affect the global supply of advanced chips. The impact could persist for years because establishing and qualifying alternative manufacturing facilities takes time.

The warning comes as AI companies compete for computing power to train increasingly capable models and serve a growing customer base. For Anthropic, which develops the Claude family of AI models, access to advanced chips is essential to sustaining that expansion.

“The majority of the world’s most advanced chip manufacturing capacity is concentrated in Taiwan, primarily at Taiwan Semiconductor Manufacturing Company,” Anthropic said in its filing.

The company also disclosed that TSMC has at times been unable to fully meet demand for its chip manufacturing services, leaving Anthropic unable to obtain sufficient quantities of chips within the required timeframe.

More from Business

Why Anthropic is worried about TSMC

TSMC is a central supplier in the global semiconductor industry, manufacturing advanced chips designed by companies that serve the AI computing market. Its manufacturing capacity is therefore a critical link between the companies designing AI processors and the infrastructure needed to run AI models.

Anthropic said this concentration creates several risks. These include tensions between Taiwan and China, military conflict, trade restrictions, tariffs, economic coercion, a blockade and natural disasters such as earthquakes and typhoons.

Interruptions to Taiwan’s electricity supply, water resources or other critical infrastructure could also affect chip production.

The concern extends beyond a temporary factory shutdown. Anthropic warned that a major disruption could have lasting consequences because advanced semiconductor manufacturing facilities require substantial time to build and qualify before they can produce chips at the required scale and specifications.

“Any military conflict, tariffs and trade restrictions, economic coercion, blockade, natural disaster, or other incident affecting Taiwan could severely disrupt the global supply of advanced chips,” the company said.

Why moving chip production outside Taiwan will take time

The US and other economies have sought to expand semiconductor manufacturing outside Taiwan to diversify supply chains and reduce geographic concentration. However, Anthropic said these efforts remain at an early stage and may not provide enough additional capacity to materially reduce the risks in the foreseeable future.

“Efforts to establish sufficient advanced chip manufacturing capacity outside of Taiwan, including in the United States, remain in early stages,” the company said in the filing.

Until alternative capacity becomes available at scale, companies that rely on advanced processors remain exposed to disruptions affecting Taiwan and TSMC.

Anthropic also flagged an existing capacity constraint. It said strong demand for advanced chip manufacturing has meant TSMC has sometimes been unable to fulfil the market’s requirements, affecting the quantities Anthropic could obtain on time.

The company warned that restrictions or delays in supplies from Taiwan or TSMC could adversely affect its business, financial condition and operating results.

Chips are not the only constraint on AI growth

Anthropic’s filing also highlights the broader infrastructure requirements behind AI development. Computing capacity depends not only on advanced processors but also on memory chips, information technology equipment, electricity and data centre infrastructure.

The company said the supply of high-bandwidth memory (HBM), a key component in high-performance AI accelerators, is concentrated among SK hynix, Samsung Electronics and Micron Technology.

These manufacturers are expanding production, but Anthropic warned that there is no guarantee supply will grow quickly enough to meet demand. Large customers or competitors securing substantial production capacity in advance could further limit availability for other buyers.

This creates a second potential bottleneck alongside advanced chip manufacturing. Even if AI processors are available, shortages of memory or other essential components could restrict the amount of computing infrastructure companies can deploy.

Electricity and data centre capacity also affect how quickly computing resources can be brought online. Anthropic said constraints across these interconnected parts of the supply chain could limit its ability to secure computing power at the scale, cost and speed it requires.

What the warning means for Anthropic’s IPO

Anthropic filed its registration statement on October 26 for a proposed initial public offering (IPO) of its Class A common stock. The company has applied to list on the Nasdaq Global Select Market under the ticker symbol ANT. The number of shares to be offered and the price range had not been determined in the initial filing.

The risk disclosures give prospective investors a view of the operational challenges that could affect Anthropic’s growth and financial performance.

Access to computing power is particularly important for an AI developer because shortages or higher infrastructure costs could limit the pace at which it develops models, expands services and meets customer demand.

Anthropic said constraints affecting chips, memory, electricity and data centre infrastructure could increase the cost of securing computing resources and limit investment in its models, platform and applications.