Cathie Wood warns high HBM prices may drive AI chip redesigns

Cathie Wood, founder of ARK Invest, has warned that the rapid price increase of High Bandwidth Memory (HBM) could act as a negative signal for the semiconductor industry. While many investors view surging component prices as a positive indicator, Wood suggests that extreme price hikes—such as those seen in the HBM market—may drive engineers to develop new architectures that reduce or eliminate reliance on these expensive components.

Wood cited companies like Groq and Cerebras as examples of firms utilizing innovative engineering to bypass traditional HBM requirements. Groq utilizes a Language Processing Unit (LPU) that integrates memory and computation on-chip, while Cerebras employs a Wafer-Scale Engine (WSE) to integrate computing cores and memory onto a single large chip. Wood compared this trend to Tesla’s previous engineering efforts to reduce cobalt usage in batteries when supply chain costs rose.

Analysts also note that the current divergence in free cash flow between chipmakers and hyperscale cloud providers may be temporary. Furthermore, market competition and optimization efforts in algorithm development are increasingly focused on minimizing memory usage to avoid supply chain vulnerabilities.