Cathie Wood, founder of Ark Invest, has rolled out a round of portfolio adjustments lately. She reduced her holdings in Advanced Micro Devices (AMD) while increasing positions in Cerebras Systems (CBRS), an AI chip maker that just completed its IPO. The trading activity has drawn wide market attention. Rather than turning bearish on AMD, the move represents a differentiated layout targeting the booming AI inference track.
The portfolio shifts were carried out across Ark’s core exchange-traded funds in late May. Wood’s reduction of AMD shares was purely partial profit-taking, and the chip giant remains the second-largest holding across her portfolios, a clear sign that she has not abandoned the stock.
Meanwhile, Wood seized a pullback in Cerebras’ share price to build new positions. The company went public on May 14, notching a 68% rally on its trading debut and claiming the title of the year’s largest IPO. From its IPO date through May 27, its stock retreated 14%, creating an entry window for Wood. After the additions, Cerebras holds a notable weighting in both the ARK Innovation ETF and the ARK Next Generation Internet ETF, with its position size roughly matching that of Nvidia, underscoring Wood’s optimism toward the newly listed AI chip player.
Wood’s dual trading decisions stem from her bullish long-term outlook for the AI inference market. She has built a portfolio structure pairing a steady core holding with a promising emerging player, based on the two companies’ vastly different technical routes and risk profiles.
In the AI chip sector, Nvidia dominates the AI training market thanks to its robust CUDA ecosystem, while AMD has carved out clear advantages in inference scenarios. Its ROCm software platform has seen substantial upgrades over the years. Paired with its proprietary chiplet architecture that accommodates more high-bandwidth memory, AMD’s products are highly optimized for inference workloads. The company has secured two large-scale GPU orders, and Anthropic also plans to deploy AMD’s latest GPUs for inference tasks.
The rapid rise of agentic AI is reshaping the ratio of GPUs to CPUs in data centers, which has driven surging demand for CPUs in inference and agentic AI applications. As the leader in this segment, AMD stands to tap into a trillion-dollar market, delivering highly visible growth — the key reason why Wood keeps AMD as a core long-term holding.
Cerebras takes an entirely different technical path. Since inference operations rely heavily on memory resources, the firm has shunned the mainstream high-bandwidth memory solution adopted by most peers. Instead, it integrates static random access memory directly onto its wafer-scale chips, delivering inference speeds far exceeding those of leading conventional GPUs.
That said, Cerebras’ oversized chips come with drawbacks: they are more complex and costly to manufacture, and require dedicated cooling and power management systems. For this reason, the company only sells or leases its products as complete server rack systems. Backed by major orders from OpenAI and newly available on Amazon Web Services, Cerebras has posted robust revenue growth. Even so, the firm has yet to achieve operational profitability. Coupled with its steep valuation based on trailing sales and high barriers to large-scale deployment, Cerebras is classified as a high-growth yet high-risk investment.
In essence, Wood retains AMD as a core position for its established business and reliable growth potential. Her modest allocation to Cerebras is a bet on the breakthrough of disruptive next-generation technology. This portfolio strategy balances steady returns and exposure to cutting-edge innovation, which has long defined Wood’s investment style.
