1. 首页
  2. ›Citi Revisits NVIDIA: What It Takes to Deliver 75% Growth

Citi Revisits NVIDIA: What It Takes to Deliver 75% Growth

N
nashnova research
·今天发布·阅读约 18 分钟

Citi revisited Nvidia management after earnings, pressing on the conditions behind 75% revenue growth — tighter financing thresholds, rising supply commitments, and Rubin's ramp pace will determine whether FY28 targets hold.

01

Why is Nvidia pulling back credit support for NeoCloud customers?

•
Nvidia previously guaranteed NeoCloud partners — firms that lease Nvidia compute and re-rent it — that if capacity went unleased, Nvidia would take it on. If rents exceeded a baseline, both sides shared the upside.
•
By the October quarter-end, guarantee balances had dropped roughly $4 billion sequentially. This means → mature NeoClouds can now self-fund using investment-grade customer contracts and prepayments, reducing reliance on Nvidia's credit.
•
The bar for new support has risen materially — partners with weaker downstream offtake will struggle to qualify. A "five-hundred-billion-dollar financing discussion" involving six partners remains ongoing; management declined to comment on its size or structure.
•
In plain terms = a smaller guarantee balance does not mean cancelled orders, but Nvidia is no longer underwriting new customers as freely — a clear risk-tightening signal.
02

Why did supply commitments rise by $100 billion at the same time?

•
Even as the financing side tightened, supply commitments grew roughly $100 billion sequentially, driven mainly by memory, plus substrates and optical components. Longer commitment horizons partly explain the larger total.
•
Management expects memory prices in calendar 2027 to exceed 2026 levels, adding explicitly: had memory not risen this much, platform configurations would differ. This means → memory inflation is directly reshaping system bills of materials.
•
The FY28 gross-margin outlook of 72%–73% already reflects higher memory costs. Key bottlenecks remain wafers, advanced packaging, and substrates — advanced-node wafer lead times run six to seven months; advanced packaging takes longer.
•
On the supply chain: Samsung Electro-Mechanics has been onboarded as a new substrate supplier, and a new ASE/SPIL agreement extends their partnership into CoWoS advanced packaging — a process that stacks multiple chips on a single substrate. Management stressed the TSMC relationship remains solid.
03

What does that 10% China revenue actually include?

•
Measured by customer-headquarters location, China accounted for roughly 10% of October-quarter revenue — spanning gaming, workstation, automotive, networking, and H20/H200 products. Not all of it is restricted AI-chip revenue.
•
Management reiterated it does not plan to continue shipping H20/H200. The resulting $2 billion-plus sequential headwind is baked into the January-quarter $130 billion revenue guide.
•
Singapore revenue is in the single digits; about 99% serves a single U.S. hyperscaler. In plain terms = headquarters location and ship-to location measure different things and cannot be conflated.
•
Remote-access restrictions land at the individual data-center level — customers must prove a given non-U.S. data center has no Chinese offtake, or Nvidia cannot ship to it. The January-quarter guidance band widened to ±3%, roughly one percentage point wider than usual.
04

Whose "orders" are that $2.5 trillion?

•
The $2.5 trillion cited on the earnings call is a sum of remaining performance obligations or backlog disclosed by hyperscale cloud customers — not Nvidia's own order book.
•
This means → between cloud-service revenue and Nvidia hardware orders sit construction, procurement, financing, and deployment steps. Mapping the entire figure directly to Nvidia's future revenue overstates visibility.
•
Management said it would not update FY28 guidance without sufficiently strong visibility, yet provided no specific calendar-2028 guidance — only an expectation that a substantial share of compute demand will carry into 2028, with no commitment that growth will continue.
05

Rubin is ramping well — so why no crossover date?

•
The January quarter is Rubin's second ramp quarter — Nvidia's next-generation GPU platform. Management called progress smooth, with revenue dollars and share both set to rise, and confirmed shipments to major hyperscalers and large NeoClouds.
•
The company has yet to specify the quarter when Rubin revenue will cross above Blackwell. Confronted with estimates that Blackwell revenue declined sequentially in the October quarter, management did not deny it, calling product transitions routine and cautioning against reading a broad demand decline into the numbers.
•
NVLink, InfiniBand, and Spectrum-X all posted strong year-on-year growth, though the company did not break out individual figures. Expectations for FY28 Vera CPU revenue to more than double and enterprise revenue to more than double remain unchanged.
•
Automotive remains the largest enterprise vertical, followed by financial services, healthcare, and spatial computing. Management emphasized FY28 growth is broad-based, not dependent on any single customer.
06

What do the Australia delay and the 10-Q hold-up signal?

•
The Australian 2GW project was originally targeted for completion by end of calendar 2027. Management said recent announcements require a timeline reassessment; no new completion date was provided.
•
On the delayed 10-Q filing — the standard quarterly financial report for U.S.-listed companies — Nvidia expressed full confidence in its financial data but said the delay involves matters related to an ongoing investigation it wants to understand more fully before filing.
•
This reflects a gap between verbal assurance on financials and the formal compliance confirmation that a filed document provides — the two should be evaluated separately.
•
In plain terms = this follow-up did not change Nvidia's overall view on compute demand, but it made the delivery conditions far more specific: Rubin's ramp cadence, customer financing capacity, and formal disclosure progress are the three key variables that will determine whether FY28 growth actually materializes.

市场有风险,内容仅供研究参考,不构成投资建议。

相关标的

4
  • NVIDIANVDA.US-1.6%

    Direct subject of the event. Citi's follow-up focuses on FY28 revenue delivery conditions, with tighter financing thresholds and Rubin ramp timing directly impacting revenue expectations.

  • Taiwan SemiconductorTSM.US+0.8%

    NVIDIA's primary supplier for advanced packaging and wafers. Supply commitments increased by approximately $100 billion, with capacity bottlenecks directly translating into TSMC orders.

  • ASE TechnologyASX.US+2.3%

    New agreement expands collaboration into CoWoS advanced packaging. NVIDIA's supply chain expansion directly increases its packaging business volume.

  • Micron TechnologyMU.US-0.3%

    NVIDIA's incremental supply commitments are primarily memory-driven, and memory prices are expected to rise in 2027, directly benefiting Micron as a major HBM supplier.