American artificial intelligence company Anthropic has identified its reliance on Taiwan's TSMC for manufacturing and potential shortages of high-bandwidth memory (HBM) as significant management risks. The company warned that if the supply of key semiconductors for AI model training and services becomes bottlenecked, it could jeopardize its business.
In a risk factor section of its S-1 filing submitted to the U.S. Securities and Exchange Commission on October 26, Anthropic stated that it is entirely dependent on three companies—SK Hynix, Samsung Electronics, and Micron—for HBM and high-performance memory, which are critical for AI accelerators.
While memory manufacturers are expanding production, they may struggle to keep pace with surging demand. Anthropic emphasized that if competitors or large customers secure available production capacity first, it could lead to severe setbacks.
The situation is similar in the foundry sector, where advanced manufacturing capabilities are overly concentrated at TSMC. This concentration poses risks to the entire supply chain in the event of geopolitical tensions or natural disasters such as earthquakes or typhoons.
Efforts to establish advanced chip production infrastructure outside Taiwan, including in the U.S., are still in the early stages, making it difficult to immediately alleviate these concentration risks. Anthropic noted that it has previously faced challenges in obtaining necessary chips on time due to TSMC's supply limitations.
As semiconductor supply issues persist, Anthropic's infrastructure costs have surged due to rising computing expenses. In the first half of this year, the company's computing and infrastructure costs reached $8.945 billion, a staggering 283% increase from $2.334 billion during the same period last year. Concerns have also been raised that if major competitors succeed in vertical integration from chip design to data centers and hardware, Anthropic could fall behind in cost competitiveness.
Additionally, Anthropic included the existential risk posed by AI technology as a major concern beyond semiconductor supply and financial limitations. The company explained that advanced AI developed without adequate safeguards could become uncontrollable or be misused for manufacturing bioweapons, surveillance, and oppression by authoritarian regimes. Dario Amadei, CEO of Anthropic, has also mentioned the possibility of AI systems evolving into agent clusters that could control the entire internet.
Other factors that could threaten future operations include environmental pollution from data center construction, community backlash over power and water allocation, and high revenue dependence on major clients like SpaceX and Microsoft.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.