PDF 原檔：報告_JPM_NVIDIA_20260902_original.pdf
原始內容
J.P. Morgan
NVIDIA Corporation
Takeaways From Recent Meetings with Investors
Yesterday, we held a virtual NDR for investors, during which we spoke with Toshiya Hari, Vice President of Investor Relations & Strategic Finance at NVIDIA Corporation (NVDA). We highlight key takeaways from our conversation below:
- FY28 growth outlook is supported by broad-based improvements in demand and visibility; NVDA appears comfortable with the 70% Y/Y growth framework… Mr. Hari indicated that the 70% Y/Y growth framework for FY28 was not driven by any single factor, but rather by continued strength across the market, including hyperscalers, neoclouds, AI labs, sovereign AI, and enterprise/on-prem demand. In addition to having much stronger visibility into next year, the team noted that its decision to provide an out-year outlook was also motivated by the fact that there was a meaningful gap between Street estimates and what NVDA was seeing internally, which the company believed could have created supply-chain planning challenges for partners. That said, Mr. Hari noted that 70% Y/Y growth is a figure that NVDA is comfortable with and that, absent any supply constraints, the business could more than double Y/Y, reiterating the point that the current outlook remains supply-constrained rather than demand-constrained.
- Inference is becoming a larger part of NVDA's DC business, but platform fungibility makes it difficult to quantify the mix… One of the recurring points of discussion throughout our meeting centered on the mix of NVDA's training/inferencing revenues, though the team indicated that it was difficult to precisely size the training versus inference split because NVDA's platforms are highly fungible (e.g., customers are able to use Grace Blackwell products for training and then later shift those same assets toward inference workloads). Nevertheless, Mr. Hari did indicate that the mix was roughly 50/50 around ~18 months ago and expressed confidence in inference now being larger and continuing to grow as a percentage of the business over time.
- Advanced wafers and memory are the two most important supply-side bottlenecks… When discussing NVDA's ability to meet demand next year and ongoing supply constraints, Mr. Hari noted that memory and advanced wafers particularly stood out given their importance within NVDA's bill of materials (BOM). Accordingly, the team highlighted that it remains deeply engaged with TSMC and all three memory vendors (MU, SK Hynix, and Samsung), with NVDA holding ongoing discussions with each of these suppliers centered around improving supply availability.
- Customer mix continues to broaden beyond hyperscalers… Mr. Hari indicated that the two largest frontier model builders (OpenAI & Anthropic) should represent ~20% of NVDA's business today on an end-consumption basis, with that figure potentially moving toward ~25% in FY28. However, remember that this isn't necessarily reflected in NVDA's direct customer mix, since it sells compute to hyperscalers/neoclouds, who then sell capacity to model builders. Beyond the AI labs, Mr. Hari also noted that neoclouds are now a meaningful portion of NVDA's business (more than ~50% of ACIE), illustrating that growth is not solely dependent on the top hyperscalers, but that
See page 4 for analyst certification and important disclosures.
Overweight
NVDA, NVDA US
Price (01 Sep 26):$217.44
Price Target (Dec-27):$320.00
Semiconductors & Semiconductor Capital Equipment / IT Hardware
Harlan Sur AC
(1-415) 315-6700 harlan.sur@jpmorgan.com
Mayur Ramdhani (1-212) 622-1664 mayur.ramdhani@jpmorgan.com
Apoorva Kumar
(1-212) 270-0668 apoorva.kumar@jpmorgan.com J.P. Morgan Securities LLC
it is increasingly being driven by a wider ecosystem of customers building and/or leasing compute capacity.
- Further advancements in computing will require both open-source and closedsource models, not either/or; frontier model builders are also driving increasing gross profit per token as NVIDIA's compute platform drives strong gen-on-gen costper-token declines... Another recurring topic in our meeting was the debate between open-source and closed-source LLMs and whether one would become more prominent going forward. However, consistent with Jensen's prior comments, Mr. Hari emphasized that advancements in AI will likely require both open-source and closed-source models (not either/or). Internally, NVDA uses a broad range of closed models such as OpenAI and Claude, while for mission-critical workloads like chip design, the company relies on a blend of closed and open-source approaches. In any case, as long as model builders are able to monetize their products and continue to improve their economics, demand should flow through to chip providers such as NVDA. The team also notably mentioned that model-builder GMs appear to be improving, likely supported by NVDA's ability to drive gen-on-gen cost-per-token improvements.
- NVDA's financing vehicles are meant to support stronger forward demand for AI infrastructure… Mr. Hari also walked through several of NVDA's financing vehicles, including the company's revenue-share agreements with select neoclouds, the PORTSPike DC campus initiative, and the $500B private capital financing platform with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, KKR, and others. On the revenueshare agreements, NVDA explained that it helps backstop a base-price and shares in the upside if compute is leased at higher market rates (which creates recurring revenue optionality on top of core hardware sales). More broadly, the NVDA team continues to push back on circular financing concerns, as it believes its financing agreements are measured, capped, and supported by strong underlying demand, ecosystem returns, and the credit profile of the ultimate compute purchasers.
Investment Thesis, Valuation and Risks
NVIDIA Corporation (Overweight; Price Target: $320.00)
Investment Thesis
NVIDIA's vertically integrated platform (spanning multiple chips, rack systems, and the software stack that ties them together) is difficult to replicate and represents a competitive advantage relative to AI compute peers. Multi-year demand visibility remains exceptionally strong and is explicitly framed as a floor, excluding several incremental revenue streams while being layered on top of a structural, underappreciated shift of traditional enterprise workloads to accelerated compute that is independent of the AI training/inference cycle. We see constructive demand fundamentals coupled with an inflecting capital return supporting material upside to the stock.
Valuation
Our PT assumes NVDA trades at ~20x (reflecting long-term growth rate expectations) applied to CY26 EPS of $15.87.
Risks to Rating and Price Target
- TAM share capture by alternative compute platforms (AI ASICs/XPUs primarily, but also merchant competitors (AMD) and specialized accelerators) has been an ongoing narrative headwind, and one that is challenging to dispel entirely given the significant program ramps planned over the next several years; we expect GPU and ASIC/XPU share of the AI compute TAM to trend toward parity in coming years.
- Memory cost inflation and supply constraints have pressured platform specifications and raise questions over the longer-term durability of gross margins, introducing uncertainty around both top-line mix and profitability as the company scales back memory capacity to meet compute volume commitments.
Analyst Certification: The Research Analyst(s) denoted by an 'AC' on the cover of this report certifies (or, where multiple Research Analysts are primarily responsible for this report, the Research Analyst denoted by an 'AC' on the cover or within the document individually certifies, with respect to each security or issuer that the Research Analyst covers in this research) that: (1) all of the views expressed in this report accurately reflect the Research Analyst's personal views about any and all of the subject securities or issuers; and (2) no part of any of the Research Analyst's compensation was, is, or will be directly or indirectly related to the specific recommendations or views expressed by the Research Analyst(s) in this report. For all Korea-based Research Analysts listed on the front cover, if applicable, they also certify, as per KOFIA requirements, that the Research Analyst's analysis was made in good faith and that the views reflect the Research Analyst's own opinion, without undue influence or intervention.
All authors named within this report are Research Analysts who produce independent research unless otherwise specified. In Europe, Sector Specialists (Sales and Trading) may be shown on this report as contacts but are not authors of the report or part of the Research Department.