Bank of America: NVIDIA’s valuation is at a 7-year low, with a discount of 30% to 35% to larger technology peers

Bank of America: NVIDIA’s valuation is at a 7-year low, with a discount of 30% to 35% to larger technology peers.

As the AI investment craze continues to heat up, NVIDIA's market performance does not seem to match. As of the close of trading on Wednesday Eastern Time, Nvidia's gain for the year was only 9.6%, far less than the 87% gain of the Philadelphia Semiconductor Index during the same period.

Bank of America Securities pointed out in its latest research report that Nvidia is still at the core of the AI computing power industry chain.

Market concerns around storage costs, self-developed ASIC competition, position congestion, and ecological investment efficiency have been over-amplified, and the current valuation provides a more attractive buying opportunity.

Bank of America maintains a "buy" rating on Nvidia with a target price of $350, based on a price-to-earnings ratio of 26 times calendar year 2027 earnings per share, excluding cash. This valuation is within the range of NVIDIA's historical forward price-to-earnings ratio of 25 times to 56 times.

The market is overly pessimistic about Nvidia Bank of America believes that the core problem facing Nvidia is not the deterioration of fundamentals, but that investors are beginning to re-examine the risk-return ratio of AI trading.

As the scale of AI capital expenditures continues to expand, the market is worried that cloud manufacturers are investing too quickly, storage costs are rising, and self-developed chips are replacing GPUs, which has put Nvidia's valuation under pressure. However, from a horizontal comparison, NVIDIA's valuation discount is already more obvious.

Bank of America said Nvidia is currently trading at about 18 times forward earnings, a seven-year low. Based on calendar year estimates, Nvidia's 2027/2028 price-to-earnings ratio is about 16 times/12 times, while large technology peers such as Apple, Microsoft, Google, Amazon, and Meta average 22 times/19 times.

In other words, NVIDIA has a discount of about 30% to 35% relative to large technology stocks. What makes this discount even more noteworthy is that Nvidia's growth rate is significantly higher than that of its peers. Bank of America predicts that Nvidia's sales and earnings per share compound growth rates will reach 48% and 52% from 2025 to 2028.

In contrast, non-Nvidia's large technology peers' sales and EPS compound growth rates during the same period are only 16% and 15% respectively. Nvidia's PEG (price-to-earnings ratio relative to earnings growth ratio) in 2027 is only 0.3 times, which is far lower than the peer average of 1.6 times.

(Comparison of price-to-earnings ratios, PEG and other data between NVIDIA and other technology giants) This means that on the one hand, the market recognizes the long-term trend of AI, but on the other hand, it is pricing the core beneficiaries of AI computing power at lower valuations.

Bank of America believes that this contradiction reflects investors' over-pricing of short-term risks, rather than a reversal of Nvidia's growth logic. HBM's price increase is not the decisive pressure, Nvidia's pricing power is underestimated The market’s biggest concern recently is that rising HBM costs will compress Nvidia’s gross profit margin.

Bank of America believes this concern is exaggerated. The report pointed out that from Blackwell to Rubin, the cost of a single rack HBM may increase by about US$200,000-300,000, but the price of the entire rack may increase by US$2 million-3 million.

Blackwell racks sell for about $3 million to $4 million, while Vera Rubin racks may cost $6 million to $7 million. In other words, storage costs have indeed increased, but NVIDIA has stronger pricing power through a complete set of rack architecture upgrades such as CPU, NVLink, Quantum Ethernet network, and software functions.

Specifically, the Blackwell B200 rack HBM cost is about US$156,000, accounting for about 5.2% of the rack ASP; the Blackwell Ultra B300 is about US$317,000, accounting for 7.9%; the Rubin V200 is about US$382,000, accounting for 6.4%; Rubin Ultra Although the HBM cost of the V300 has risen to US$1.534 million, the rack ASP is expected to be as high as US$21 million, accounting for approximately 7.3%.

This means that even if the cost of HBM increases, its proportion in the value of the entire machine is still relatively controllable. Bank of America expects Nvidia's overall gross profit margin to remain around 75%.

The report also emphasizes that Vera Rubin can achieve about 10 times performance/watt improvement compared to Blackwell, which means that the cost of inference tokens can be reduced by about 10 times; inference performance is improved by 3.3 times, and training performance is improved by up to 5 times.

It is difficult for cloud vendors to develop self-developed ASICs to shake the advantages of NVIDIA platform Another major market concern comes from cloud manufacturers’ self-developed chips. ASICs such as Google TPU, Amazon Trainium, and Meta MTIA are regarded as potential alternatives to Nvidia GPUs.

Especially in the context of huge AI capital expenditures and cloud manufacturers hoping to reduce costs, the market is worried that self-developed chips will weaken Nvidia's share. Bank of America believes that this concern ignores historical experience.

Google TPU was launched as early as 2015, Amazon Trainium in 2020, and Meta MTIA in 2023, but since 2015, Nvidia GPU accelerator revenue has grown approximately 700 times. In other words, ASIC has existed for a long time, but it has not stopped Nvidia from continuing to expand in the AI computing power market.

The key difference is that ASIC usually serves specific cloud vendors and specific workloads, with a narrow range of applications; NVIDIA provides a widely available, ecologically mature, software and hardware integrated platform.

For enterprises, sovereign AI, cloud service providers, emerging cloud vendors and developers, the NVIDIA platform still has advantages in compatibility, supply chain, development tools and deployment efficiency.

The latest data also shows that Nvidia’s sales to ultra-large-scale cloud vendors increased by 115% year-on-year, almost twice the growth rate of cloud capital expenditures. This shows that Nvidia has not only not lost share from cloud vendors, but is still expanding its share.

Cash flow is sufficient to support ecological investment and shareholder returns The market is also concerned that Nvidia's large-scale investments in suppliers and customers may constitute an inefficient use of capital or even be interpreted as stimulating demand.

Bank of America admitted that this may raise questions about the quality of income, the efficiency of capital allocation and the authenticity of customer demand. However, the bank believes that its scale is still controllable relative to Nvidia's cash flow.

According to report statistics, Nvidia’s recent ecological investment totals approximately US$65 billion, covering OpenAI, Anthropic, Intel, Synopsys, Nokia, CoreWeave, Nebius, Lumentum, Coherent, Marvell, Corning, IREN, Wayve, Nscale, Ayar Labs, xAI, etc.

Among them, investment in OpenAI is approximately US$30 billion, investment in Anthropic is up to US$10 billion, and investment in Intel is approximately US$5 billion.

Bank of America estimates that this US$65 billion accounts for approximately 35% of the US$187 billion expected free cash flow in the 2026 calendar year, and only 17% of the US$385 billion expected free cash flow in the 2027 calendar year. Therefore, Nvidia still has sufficient ability to promote dividends and buybacks.

In terms of profit forecasts, Bank of America expects Nvidia's revenue in fiscal year 2027 to be US$396.1 billion and EPS of US$9.09; fiscal year 2028 revenue to be US$564.6 billion and EPS to be US$13.27; fiscal year 2029 revenue to be US$739 billion and EPS to be US$18.04.

In the long-term model, by calendar year 2030, Nvidia is expected to achieve revenue of $1.04 trillion, net profit of approximately $589.1 billion, and EPS Power (potential earnings per share) of approximately $25.26.

The AI data center market is still in a period of rapid expansion The fundamental reason why Bank of America is optimistic about Nvidia is still the long-term expansion of the AI data center market. The report estimates that current global IT spending is approximately US$6.3 trillion, of which the data center system market is approximately US$788 billion.

By 2030, data center system TAM is expected to increase to approximately US$2.1 trillion, with a compound growth rate of 33% from 2025 to 2030, which is significantly higher than the 9% growth rate of overall IT spending.

Among them, AI data center system TAM is expected to increase from US$273 billion in 2025 to US$1.7 trillion in 2030, with a compound growth rate of 44%.

Broken down, AI servers are expected to reach US$1.3 trillion by 2030, accounting for approximately 77% of AI data center system TAM; AI network equipment is expected to reach US$305 billion, accounting for 18%; AI storage is expected to reach US$81.4 billion, accounting for 5%. Within AI servers, AI accelerators are still the largest market.

Bank of America predicts that the AI accelerator market will grow from US$196.5 billion in 2025 to US$1.13 trillion in 2030. The HBM market will expand accordingly and is expected to reach $246 billion by 2030, accounting for approximately 20% to 23% of accelerator spending. In addition to GPUs, AI CPUs and network connections will also grow rapidly.

The overall server CPU market is expected to increase from approximately US$35 billion in 2025 to US$170 billion in 2030; of which AI CPU may increase from US$18 billion to US$140 billion.

The AI connection market is expected to increase from US$17 billion to US$110 billion, with optical connections and copper connections increasing to US$88 billion and US$22 billion respectively. (

#Stocks #Nvidia #Apple #Microsoft #Meta