The AI wave is still surging! Morgan Stanley: The revenue upside potential of NVIDIA, AMD, and Broadcom is underestimated

According to Zhitong Finance APP, JP Morgan stated that driven by the recent surge in artificial intelligence (AI) related transactions, NVIDIA (NVDA.US), AMD (AMD.US), and Broadcom (AVGO.US) may have greater revenue upside potential than the market expects.

JP Morgan analyst Harlan Sur mentioned in a client report: "In short, although the market's forward earnings expectations for NVIDIA, AMD, and Broadcom have been significantly raised in recent weeks, analysts still maintain a degree of conservatism in their forecasts due to concerns surrounding infrastructure financing, competitive landscape, and execution capabilities."

The analyst added: "From the perspective of data center capital expenditures, we believe there is still ample growth space in the coming years, which will support a 40%-50% compound annual growth rate (CAGR) in the AI accelerator market, as well as the continuously expanding AI 'cake', ultimately driving the joint growth of the GPU and XPU markets (we expect the share of XPU in the overall market to gradually increase). Based on the current level of market consensus, we anticipate that as visibility improves and more capacity agreements are announced, AI-related revenue expectations will enter a sustained upward adjustment period, thereby driving further upward momentum for these stocks."

NVIDIA

In further analysis, Harlan Sur pointed out that the market may underestimate the growth in power scale of AI data center deployments by 2027. NVIDIA's management has repeatedly stated that every 1 gigawatt (GW) of AI data center capacity can generate $35 billion to $40 billion in revenue, so NVIDIA's revenue is still expected to maintain strong growth.

Harlan Sur stated: "According to our calculations, the current platform (i.e., Blackwell + Blackwell Ultra) has not fully reflected this revenue potential, but future platforms will be closer to this level." He predicts that with the upcoming Rubin platform, NVIDIA's revenue per gigawatt may approach $30 billion, while the revenue per gigawatt for the Rubin Ultra platform is expected to reach the mid-range of $30 billion.

He added: "More importantly, in terms of per-die revenue, NVIDIA's revenue potential on the Rubin Ultra platform is nearly three times that of the Blackwell platform (Rubin Ultra is $70,000 to $80,000 per chip, while Blackwell is about $26,000), which will be a key driver of NVIDIA's revenue growth per gigawatt in the coming years (rather than the growth in GPU chip shipments)."

AMD

Harlan Sur stated that AMD's goal of achieving $20 billion in revenue per gigawatt on its Helios platform, and subsequent platforms reaching $25 billion or higher in revenue, is "entirely achievable."

AMD's management previously stated that the scale of the latest deal with OpenAI "reaches hundreds of billions of dollars in revenue per gigawatt." Harlan Sur pointed out that this means at least about $15 billion per gigawatt. Although AMD has traditionally priced lower than NVIDIA and has less proprietary content in network components, he believes that the revenue level of $20 billion per gigawatt for the Helios platform is still reasonable He stated: "In terms of the deal with OpenAI, although strictly speaking, OpenAI has only committed to the first 1 gigawatt of computing power, we believe that the risk of deploying the full 6 gigawatts of capacity within the four-year period set by the agreement is very limited." "On an annualized basis, this deal alone means that AMD will generate approximately $30 billion to $35 billion in revenue each year, while the market currently expects AMD's total revenue from data center GPUs to be only about $31 billion in 2027. In other words, there is still significant room for upward adjustment in analysts' expectations."

Broadcom

Harlan Sur pointed out that after recently reaching a partnership with OpenAI, achieving over $100 billion in AI-related revenue by 2027 for Broadcom is "not a pipe dream." Assuming Broadcom's XPU is about 30% more economical than NVIDIA's GPU, this means its revenue per gigawatt is approximately $27 billion, while NVIDIA's is $35 billion. Considering an additional rate of about 30% for network business (similar to Broadcom's existing AI XPU business), the total revenue per gigawatt would reach about $28 billion.

If all 10 gigawatts of capacity are deployed within the next three years, the OpenAI-related deal could bring Broadcom $70 billion to $90 billion in revenue between 2027 and 2029, while Wall Street's current estimate for Broadcom's AI revenue in 2027 is only about $60 billion.

Harlan Sur stated: "In addition, we expect Google's TPU business to bring Broadcom over $20 billion in revenue by 2027. Combined with OpenAI-related income, Broadcom's total AI revenue in 2027 will approach $100 billion (compared to about $21 billion in 2025), which is 50%-70% higher than market consensus."