Anthropic flags HBM supply risk from Samsung, SK hynix in IPO filing

Anthropic, the developer of AI chatbot Claude, has flagged the possibility that HBM supply from Samsung Electronics, SK hynix and US chipmaker Micron may fail to keep pace with AI demand as a key risk factor in its IPO filing. The company also warned that the concentration of advanced semiconductor manufacturing at Taiwan's TSMC could constrain its AI business expansion.

In an S-1 registration statement filed with the US Securities and Exchange Commission (SEC) on Monday (local time), Anthropic said supply bottlenecks across semiconductors, power and data centers could emerge as it works to secure the computing resources needed to develop AI models and run its services.

The filing particularly highlighted that SK hynix, Samsung Electronics and Micron produce the vast majority of HBM and other high-performance memory chips that are critical components in AI accelerators.

Anthropic said the three companies are expanding their production capacity to meet surging demand, but that it remains uncertain whether supply will grow fast enough to satisfy it.

The company also raised concerns that supply shortages could worsen if large customers or rivals lock in exclusive supply agreements or secure the bulk of chipmakers' production capacity. It added that changes by manufacturers to their allocation decisions or business strategies could also make it harder to obtain the semiconductors it needs in a timely manner.

The concentration of advanced chip manufacturing at TSMC was also cited as a risk. Anthropic warned that geopolitical tensions or natural disasters such as earthquakes and typhoons could severely disrupt the semiconductor supply chain.

While efforts are underway to expand advanced chip production capacity outside Taiwan — including in the United States — Anthropic said those initiatives remain at an early stage, making it difficult to reduce dependence on TSMC in the near term. The company also disclosed that there have been instances where TSMC was unable to fully meet demand, leaving Anthropic unable to secure sufficient semiconductors when needed.

AI infrastructure costs are also rising rapidly. Anthropic's computing and infrastructure expenses for the first half of this year reached $8.95 billion, up 283 percent from $2.33 billion in the same period last year. The increase was driven by growing AI model development activity, increased service usage and higher costs for securing computing resources.

The vertical integration of supply chains by rivals was also identified as a concern. Anthropic said that if major AI companies move directly into chip design and production or data center operations — or strike exclusive deals with hardware manufacturers — competitors could gain access to large-scale computing resources at lower cost.

The disclosures come as Anthropic pursues a NASDAQ listing targeting a valuation of as much as $2 trillion. Even as the AI industry grows rapidly, shortages of critical semiconductors and the concentration of production facilities are emerging as key variables that could determine the pace of the company's future expansion.