Is the "AI Moment" belonging to AMD here? Wall Street bullish report: Major clients are shifting from NVIDIA to AMD

According to the Zhitong Finance APP, Wall Street investment firm Truist Securities has released a new research report stating that AMD (AMD.US), a leader in the PC, HPC, and data center chip sectors, has seen stronger-than-expected demand for data center CPUs and artificial intelligence computing power chips. As a result, Truist has upgraded AMD's stock rating from "Hold" to "Buy," the most optimistic bullish rating, and significantly raised AMD's 12-month target price from $173 to $213.

In the latest report, Truist stated that some hyperscale customers focusing on AI infrastructure are now seriously considering moderately shifting from NVIDIA's (NVDA.US) AI server clusters to AMD for a more cost-effective and comprehensive performance AI computing system. This suggests that NVIDIA's market share in the AI chip and AI server cluster market, which is as high as 90%, may gradually be eroded by AMD in the future. Following this positive rating upgrade news, along with reports that IBM and AMD will jointly develop the next-generation quantum computing architecture known as "quantum-centric supercomputing," AMD's stock price rose more than 3% during early trading on Tuesday.

"In recent years, our industry contacts, as well as AI infrastructure component buyers/sellers, have told us that hyperscale customers deploying AI clusters have only been experimenting with AMD's computing infrastructure as a 'cheap supplement' to NVIDIA," said William Stein, a senior analyst at Truist, in an investor report. "However, there seems to have been a significant shift recently, as over the past month or so, our industry contacts and buyers/sellers have generally pointed out that hyperscale customers are actively collaborating with AMD as long-term partners, expressing genuine interest in scaling up the deployment of AMD AI server clusters."

Truist also raised AMD's target price from $173 to $213. The firm believes that AMD's next-generation AI GPU, the MI355 AI GPU, launched earlier this summer, could become a significant driver of performance growth for the chip giant in the coming quarters.

Also on Tuesday, reports indicated that AMD is collaborating with the established American tech giant IBM to develop the next-generation quantum computing architecture that combines quantum computing and high-performance computing, known as "quantum-centric supercomputing." This collaboration aims to help enterprises efficiently scale AI-based high-computing-intensity workloads while leveraging IBM's secure hybrid cloud computing environment.

The Chinese market is expected to become a "significant incremental benefit" for AMD

Shortly before Truist Securities released its bullish report on AMD, AMD CEO Lisa Su personally stepped in to bolster investor confidence, emphasizing that there are very positive growth signs in demand for all AMD products, with strong demand for AMD AI chips in many global markets, and that the company is accelerating the process of obtaining U.S. government approval to return to the Chinese market The stock analyst team from the Wall Street financial giant Bank of America (BofA) recently stated that after the two major chip giants in the U.S.—NVIDIA and AMD—agreed to pay 15% of their overall revenue from AI chips in the Chinese market to the U.S. government in exchange for licenses to export AI chips to China, they maintain a "buy" rating for NVIDIA and AMD stocks, with target prices remaining at $220 and $200 for the next 12 months.

The semiconductor analyst team from another Wall Street investment firm, Bernstein, indicated that Wall Street generally does not recognize the precedent set by the U.S. government in imposing penalties on chip exports, but they also acknowledge that "getting 85% is much better than 0%."

Led by Vivek Arya, Bank of America's semiconductor industry analysts wrote: "According to media reports, in exchange for the 15% penalty, the U.S. government is willing to allow NVIDIA/AMD to ship specific AI chips (H20/MI308) to the Chinese market. We believe this is a very significant incremental positive." "The core logic is that NVIDIA/AMD likely have enough pricing power to offset part of the 15% penalty; the two chip giants can utilize some of the previously impaired inventory, allowing them to recover a significant portion of gross profit even while bearing the 15% penalty; and the recovery of the Chinese market maintains the initial goal of mutually benefiting from the extremely important Chinese AI computing power ecosystem, which is expected to somewhat restrain domestic competitors in China."

"In the past 90 days, we have seen many positive signals in global markets regarding the demand for high-performance computing hardware, and the demand for AI accelerator chips remains enormous," said Lisa Su after AMD announced its earnings in early August. When discussing the progress of resuming approvals for exporting AI chips to China, Su stated: "We have several important licenses under review, and the feedback we have received is that they are progressing smoothly." Su continued to predict that the global AI chip market size will exceed $500 billion by 2028, implying a compound annual growth rate of up to 60% over the next three years.

At the Advancing AI 2025 conference held on Thursday, June 12, Eastern Time, AMD launched the new MI350 series AI GPUs—MI350X and MI355X—specifically designed for extremely complex AI training/inference workloads, claiming that the performance of the new accelerators is three times that of the previous generation MI300X, targeting NVIDIA's Blackwell architecture product line. The two new products demonstrate AMD's ambition to challenge NVIDIA's most advanced AI chips—the Blackwell architecture AI chips—in terms of technical specifications.

Benchmark performance comparison results show that the HBM3E memory capacity of the AMD MI355X is 1.5 times that of NVIDIA's GB200/B200 AI GPUs, and in peak FP64/FP32 performance, the MI355X has about a 2-fold advantage. In terms of inference throughput, which AI engineers focus on, based on Llama-3 405B inference, the MI355X (FP4) improves by 30% Token/s/$ compared to Blackwell B200, and can still break even against GB200 AMD claims to lead by up to 1.13 times in certain specific high-parameter AI training workloads