Analysts at UBS have indicated that the recent Computex exhibition in Taiwan further supports the view that demand for computing power related to artificial intelligence (AI) continues to intensify, with the rise of AI agents being a key driver. The event also highlighted that NVIDIA (NVDA) GPU shipments continue to significantly outpace its main competitor, Advanced Micro Devices (AMD).
A team of UBS analysts led by Timothy Arcuri noted in an investor report released Monday: "Regarding the latest product roadmap cadence for NVIDIA and AMD, our latest research (consistent with our models) suggests NVIDIA's Blackwell platform will dominate shipments through 2026, with Rubin Ultra gradually joining starting in Q4 2026. For AMD's Helios platform, we still expect motherboards to ship in Q4 2026, but full rack deployment is likely pushed to late 2026 due to extra time needed for dual-wide rack optimization, validation, and data center integration."
Chip designer Arm (ARM) also provided a more optimistic outlook for its CPU business prospects, having secured more large customers. Arcuri mentioned: "Arm confirmed its AGI chips are now in mass production at TSMC, with products covering 36 kW (8,160 core) and 200 kW (45,696 core) rack configurations. Importantly, ByteDance and Oracle (ORCL) were disclosed as new AGI customers, and Arm stated its previously set AGI revenue target of $15 billion for FY2031 is actually conservative. We see the hyperscale customer base expanding, which is encouraging, as we had previously noted that relying solely on Meta (META)/OpenAI might struggle to reach the $15 billion demand scale, given their historical reliance on other suppliers (like AMD, NVIDIA, Intel (INTC), and cloud service providers' own CPUs) for cloud compute."
Additionally, Marvell Technology (MRVL) has become a market focus. This follows comments from NVIDIA CEO Jensen Huang suggesting Marvell has the potential to become the next trillion-dollar company, which directly spurred a significant rise in its stock price.
Arcuri analyzed: "NVIDIA emphasized its reliance on Marvell for interconnect, silicon photonics, and optics, including collaboration in NVLink Fusion. Our research suggests this area could impact future AWS Trainium deployments. Marvell also reiterated that the 'copper bottleneck' is actually moving inside the rack at 400G rates, further supporting the future transition towards co-packaged optics (CPO) at higher data rates."
He added: "While we view the market reaction as somewhat overdone given the limited new information in this disclosure, we remain constructive on Marvell's positioning in AI networking and XPU."
Comments