Piper Sandler initiates coverage of the AI chip sector: Nvidia (NVDA.US) and Broadcom (AVGO.US) as "top picks", with the computing power market expected to reach $2.2 trillion by 2030.

According to Golden Ten Data APP, Piper Sandler has released a research report, initiating coverage on the artificial intelligence (AI) chip and semiconductor sector for the first time. The firm rated seven major semiconductor stocks, giving “Overweight” ratings to AMD (AMD.US), Arm (ARM.US), Broadcom (AVGO.US), Marvell Technology (MRVL.US), and NVIDIA (NVDA.US), and “Neutral” ratings to Intel (INTC.US) and Qualcomm (QCOM.US). Piper Sandler noted that Broadcom and NVIDIA were named the firm’s “top picks” in the AI accelerator space, while AMD and Arm received particular attention for gaining share in the server CPU market.

In terms of price targets, Piper Sandler set AMD at $600, Arm at $320, Broadcom at $460, Intel at $110, Marvell Technology at $270, NVIDIA at $300, and Qualcomm at $190.

Analysts David O'Connor and Zackary Altman stated in the report that computing power is truly the cornerstone of AI development, and with the rapid emergence of agentic AI, the industry is in urgent need of more computing power support.

The firm pointed out that capacity for 2027 is already sold out, and the order book for 2028 is also full. Piper Sandler estimates that by 2030, the computing power market will reach $2.2 trillion, with the accelerator market accounting for $2 trillion and the CPU market for $200 billion; this means the computing power market will expand about fivefold by 2030 and drive revenue growth across the industry chain, translating into an average annual earnings per share compound growth rate of about 45% for these stocks.

Piper Sandler believes these companies are the most important participants in the construction of hyperscale and new cloud providers. The report also notes that current computing power scarcity is clearly visible in the GPU hourly pricing of new cloud providers. According to Silicon Data’s H100 price index, NVIDIA’s H100 GPU is currently priced at $2.60 per hour, compared with $2.00 in December, prior to the appearance of OpenClaw.

The report also mentions that there was a surge in demand in February 2025 due to increased demand from DeepSeek deployments; as computing power costs drop, usage will further increase, in turn continuing to drive up demand for computing power. Piper Sandler stressed that it’s worth noting NVIDIA began mass production of the H100 as early as Q4 2022, which means this chip’s technology iteration is nearing four years. In addition, component costs are also rising, pushing up hardware vendor revenue; the firm found that the cost of GPU chips in the past two generations has climbed from 30% to 40%.

At the individual stock level, O’Connor called NVIDIA “the absolute leader in AI computing power” with an 80% market share. He said the explosion of agentic AI workloads this year has led to a step-change in demand, with supply likely to remain tight for the next two to three years, adding that, at roughly 14 times projected FY2028 earnings, this stock is “one of the cheapest in the AI space.”

O’Connor described Broadcom as the leader in custom ASIC chips with a 75% share and pointed out ongoing supply constraints. He wrote, “Demand is twice the supply, making additional supply assurance critical.” He also noted the company has visible market demand of 12 gigawatts for FY2027.

O’Connor called AMD “the best beneficiary of agentic AI,” noting that AMD continues to gain share in the server CPU market, with its Helios GPU also scaling up, with anchor clients including OpenAI, Meta (META.US), and Anthropic. He expects AMD earnings to grow at a compound annual rate of 65% through 2030.

The analyst is also bullish on Marvell Technology’s data center business, calling its $120 billion deal with Google (GOOGL.US) transformative. He also suggested that the October 6 Analyst Day could serve as a potential catalyst.

Lastly, O’Connor highlighted Arm’s dominance in CPU IP and its expansion into custom chip and accelerator IP, which he believes could enable Arm to double its earnings.