Wall Street backs Nvidia: $3 million price hike per rack far exceeds HBM costs, 18x PE is 30% lower than peers

```

Facing the market's dual concerns over NVIDIA's next-generation product roadmap and memory costs eroding profits, two major Wall Street investment banks have spoken out simultaneously. Using quantitative data and supply chain research, they have “defended” its fundamentals, directly pointing out that the current valuation has fully, if not excessively, priced in pessimistic expectations.

In its latest report, Bank of America quantitatively argues that the market has severely underestimated NVIDIA's pricing power. From the Blackwell to the Rubin architecture, the high-bandwidth memory (HBM) cost per rack only increases by $200,000 to $300,000, while NVIDIA’s per-rack pricing has been drastically raised by $2 million to $3 million. This strong cost pass-through ability is enough to ensure its gross margin remains solidly in the mid-70% range.

Citi’s supply chain research has completely dispelled concerns about product delays, confirming that the Vera Rubin Ultra roadmap and NVLink domain have not changed, and that Co-Packaged Optics (CPO) technology has already entered production. With NVIDIA’s forward P/E ratio dropping to 18x, a seven-year low, the endorsement from these two investment banks provides strong logical support for a valuation recovery for this AI leader.

Following post-earnings share price volatility and pullback, multiple trading desks have begun advising investors to buy on dips. Ahead of TSMC’s upcoming results as a potential catalyst, low-volatility, high-quality tech giants like NVIDIA are becoming a core direction for renewed capital inflows.

Pricing Power Crushes Cost Anxiety, Gross Margin Moat Remains Solid

Recently, the market has expressed significant concerns over whether rising HBM memory costs will erode NVIDIA’s gross margin. Bank of America’s quantitative analysis directly refutes this view. Supply chain data shows that the HBM cost per rack was about $1.9 billion, and from the Blackwell to the Rubin architecture, HBM cost per rack only increased by $200,000 to $300,000.

In contrast, NVIDIA, through comprehensive upgrades in computing power, interconnect architecture, and software, raised the per-rack pricing from $3–4 million with Blackwell, to $6–7 million with Rubin. The price increase of $2–3 million per rack not only fully covers the rise in memory costs, but also leads Bank of America to predict that the company’s gross margin is very likely to remain firmly in the mid-70% range.

On valuation, Bank of America believes the market has unjustifiably downgraded NVIDIA’s 2027–2028 earnings by about 30% to 35%. Currently, its forward P/E ratio is only 18x, hitting a seven-year low and sitting 30% to 35% lower than peers like Amazon, Meta, and Google, who are also investing in AI and face memory cost headwinds. Despite tech giants launching their own chips, NVIDIA’s revenue from hyperscale cloud service providers still soared 115% year-over-year, and Bank of America projects it can maintain over 65% to 70% AI chip market share in the long run.

Roadmap on Track, CPO and Next-Gen Architecture Dispel Delay Concerns

Beyond costs, market worries about adjustments to NVIDIA’s next-gen chip roadmap have also been eased by Citi’s research feedback. After communication with NVIDIA’s investor relations team, Citi confirmed that the Vera Rubin Ultra roadmap remains completely unchanged, and the NVLink domain (scale-up CPO switch) showcased at this year’s Computex is also consistent.

On the technology front, management reiterated that CPO used for scale-out networks is already in production via Spectrum-X, with high customer adoption. Starting in 2028, with the Feynman architecture’s release, customers will be able to choose flexibly between CPO-integrated NVLink or copper cables.

Strong demand also provides underlying support. Citi points out that in addition to hyperscale cloud vendors driving most deployments, demand from AI labs, Neocloud, sovereign countries, and enterprise on-premises deployments is accelerating. Management believes that as physical AI rises, the market share among non-hyperscale cloud vendors will eventually have to become larger.

Valuation Bottom Forming? Trading Desks Urge Buying High-Quality Leaders on Dips

Although NVIDIA delivered better-than-expected revenue and profit in its latest earnings report, revealing up to $95.2 billion in purchase commitments to lock in capacity while also approving an $80 billion stock buyback authorization, its stock still fell in after-hours trading post-earnings, continuing a recent pullback trend.

However, according to institutions, this technical and sentiment pressure is turning into an excellent buying window. Multiple trading desks are recommending clients buy on dips at this juncture, especially targeting low-volatility, high-quality tech giants.

With TSMC about to release new results, the prosperity of the semiconductor supply chain is expected to be further validated. Against a macro backdrop of restricted industry supply and ongoing increases in AI capital expenditure, NVIDIA—with its irreplaceable pricing power and clear product roadmap—is once again becoming the core asset for major capital to anchor amid valuation troughs.

Risk Warning & DisclaimerThe market has risks and investment requires caution. This article does not constitute personal investment advice and does not consider the special investment objectives, financial situation, or needs of any particular user. Users should consider whether any opinions, views, or conclusions in this article fit their specific circumstances. Investments made accordingly are at your own risk. ```