NVIDIA exchanges investment guarantees for Neocloud revenue sharing, with long-term annuity-like returns potentially becoming a new growth engine

Overview

The report focuses on the new Neocloud partnership model promoted by NVIDIA: the company provides guarantees for emerging AI clouds’ data center investments or GPU utilization, helping partners finance and expand capacity, while receiving a share of cloud services revenue in addition to normal product sales revenue. Morgan Stanley believes this model can introduce new compute spending entities beyond hyperscale cloud providers, strengthen NVIDIA’s full-stack ecosystem, and form predictable long-term revenue streams. Although the market is concerned about circular demand, credit risk, and leverage, the report still judges that persistently tight compute, land, power, and data center resources skew the risk-reward to the upside.

Core views

First, token demand growth still exceeds compute supply growth, training resources are continuously being redirected to inference, and existing hyperscale cloud providers are constrained by free cash flow, land, power, and data center conditions, so incremental compute supply has a genuine demand basis. Second, revenue sharing can extend NVIDIA’s profit model from one-time hardware sales to long-term cloud revenue, and when guarantees are triggered, it may obtain low-cost GPU capacity for R&D. Third, new spending entities adopt NVIDIA’s most mature and broadly supported hardware and software system, helping the company continue to maintain its de facto standard status through the Vera Rubin cycle. Fourth, transaction details remain incomplete, and investors need to focus on revenue recognition, guarantee obligations, customer leverage, and related-party transaction transparency.

Analysis framework

The report combines management roadshow information, existing Neocloud partnership cases, global cloud capex, and compute capacity forecasts to build a unit economics model centered on single-GW data center cost, GPU hourly rental rates, guaranteed price, utilization rate, and revenue-sharing ratio. It then estimates FY28 and FY29 revenue, gross margin, EBIT, and EPS accretion under different capacity and rental-rate scenarios; meanwhile, it evaluates stock risk-reward through bull, base, and bear valuation scenarios.

Methodology notes

• Earnings forecastRevenue-sharing unit economics model

  Uses the spread between GPU rental rates and the cost-recovery guaranteed price as the basis for revenue sharing

  The model first derives the cost-recovery price per GPU hour based on data center depreciation and operating costs, then assumes NVIDIA receives 35% of the spread between the rental rate and the guaranteed price, and uses this to estimate annual revenue per GW and EPS contribution.

• Scenario analysisDual-variable sensitivity analysis of capacity and rental rates

  Simultaneously changes annual GW scale and market rental rate per GPU hour

  The report uses 2 to 8GW capacity and multiple GPU rental-rate scenarios to conduct sensitivity analysis on incremental EPS for FY28 and FY29, showing the impact of Neocloud scaling speed and leasing prices on profitability.

• Valuation analysisBull, base, and bear scenario valuation

  Forms a target price range based on different earnings and P/E assumptions

  The base-case target price is US$288, roughly corresponding to 22x Morgan Stanley’s CY27 EPS forecast; the bull case is US$330, and the bear case is US$160.

• Strategic analysisEcosystem flywheel analysis

  Investment guarantees, infrastructure, cloud markets, and chip sales mutually reinforce each other

  Through equity investments, capacity guarantees, fiber and data center resources, the DSX reference architecture, and DGX Cloud Lepton, NVIDIA connects compute supply and demand, expands the number of partner clouds, and increases the standardization of its hardware and software platform.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• NVIDIA (US.NVDA)

  Core beneficiary and covered company in the report

  Strengths

  Has leading GPUs, networking, CPUs, software, and full-data-center architecture; can receive both product sales and cloud revenue sharing, and reinforce the Vera Rubin product cycle through ecosystem scale.

  Weaknesses

  High market share and high gross margin limit short-term valuation expansion, while new financing guarantees and physical infrastructure investments will increase business complexity.

  Comparison

  Compared with AMD and custom ASICs, NVIDIA has more mature infrastructure support, a developer ecosystem, and flexibility across workloads, making it more suitable for emerging cloud providers that lack large-scale operating experience.

  Risks

  Compute oversupply, partner defaults, revenue circularity, insufficient transaction transparency, export controls, tariffs, and customers’ in-house chips.

• CoreWeave, Nebius, IREN, and other Neoclouds

  NVIDIA’s partners for investment, capacity guarantees, and revenue sharing

  Strengths

  Can obtain NVIDIA supply chain access, financing support, reference architecture, and customer demand routing, shortening construction cycles and improving the ability to realize initial GPU rental rates.

  Weaknesses

  Capital intensive and relatively highly leveraged, with high sensitivity to on-time data center delivery, operating reliability, and GPU utilization.

  Comparison

  Mature operators are more likely to receive incremental investment through their track records of on-time delivery and high availability, while new entrants need to prove construction and commercialization capabilities.

  Risks

  Intensifying industry competition, declining rental rates, construction delays, shortages of power and data center resources, and refinancing pressure after guarantees expire.

• AMD and custom ASIC suppliers

  NVIDIA’s competitors in the AI accelerated computing market

  Strengths

  May compete for share through pricing, optimization for specific workloads, or customers’ in-house solutions.

  Weaknesses

  Relatively weaker in the general support, full-stack infrastructure, and broad ecosystem required by emerging cloud providers.

  Comparison

  The report believes the most natural choice for smaller cloud service providers remains the most widely adopted NVIDIA platform with the most complete infrastructure support.

  Risks

  If alternative platforms improve significantly in performance, cost, or software compatibility, NVIDIA’s ecosystem advantage could weaken.

• AI data center infrastructure

  Key supporting assets for Neocloud expansion

  Strengths

  Tight supply of fiber, land, power, facilities, storage, memory, and optical components can support demand for related assets and pricing power.

  Weaknesses

  Long construction cycles, with significant impacts from regulation, politics, supply chains, and capital costs.

  Comparison

  Current bottlenecks have expanded from simple GPU shortages to land, power, facilities, DRAM, NAND, CPUs, HDDs, and optical components.

  Risks

  Project delays, cost overruns, blocked resource permitting, and compute demand below expectations.

Key data

• Rating and target priceOverweight; US$288.00Based on the 2026-08-13 closing price of US$225.30, implied upside is approximately 27.8%.
• Risk-reward price rangeBull case US$330; base case US$288; bear case US$160The bull case roughly corresponds to 23x CY27 EPS of US$14, the base case roughly corresponds to 22x CY27 EPS of US$13.08, and the bear case roughly corresponds to 16x CY27 EPS of US$10.
• NVIDIA processor revenue shareApproximately 85%The report believes recent discussion of ASIC competition has intensified, but NVIDIA’s share remains broadly stable.
• Public Neocloud capex shareApproximately 6% of global cloud capexThis metric does not yet include sovereign clouds and private Neoclouds.
• Incremental compute from major cloud providers in 2027Approximately 25GWMorgan Stanley’s technology team forecast for the four major hyperscale cloud providers, excluding TPU.
• Revenue-sharing potential per GWEPS accretion of approximately 4% to 12%The specific result depends on GPU rental rates, guaranteed price, utilization rate, and NVIDIA’s sharing ratio.
• Model base costUpfront cost of approximately US$39 billion for a 1GW GB300 data centerAssumes IT equipment useful life of 5 years, facilities and other investments of 15 years, and a blended useful life of approximately 7 years.
• Revenue-sharing model assumptionsGuaranteed price of approximately US$2.12/GPU hour; utilization rate 75%; sharing ratio 35%The report’s example also uses a US$3 guaranteed price and US$10 market rental rate to illustrate the spread-sharing mechanism; actual contracts may vary case by case.
• Large-scale upside scenarioA 25GW ecosystem at US$20 million/MW could correspond to US$500 billion in annual revenueIf NVIDIA captures one-quarter of this at nearly full gross margin, the report estimates FY28 EBIT could be revised up by 60% and FY29 by 25%; this scenario is highly dependent on scale and sharing assumptions.
• Base earnings forecastCY27 EPS of US$13.08The base target price uses approximately a 22x P/E multiple.
• Revenue growth forecast82.0% growth in 2026; 52.4% growth in 2027In the report’s base case, data center and generative AI demand continue to support high-speed growth.
• Existing Neocloud partnership exposureCoreWeave ownership slightly above 10%; Nebius ownership approximately 10%; IREN five-year cloud contract of US$3.4 billionNVIDIA has also made private investments in Nscale, Crusoe, Lambda, and Firmus, and owns related rights to purchase IREN shares.

Impact & implications

For NVIDIA, the Neocloud model could connect hardware sales, cloud capacity usage, software platforms, and infrastructure investments into a closed loop, making the revenue structure more recurring and improving long-term earnings predictability, while consolidating its standard status in emerging AI cloud and sovereign computing markets. For Neoclouds, NVIDIA’s supply chain, financing guarantees, DSX full-data-center reference architecture, and demand routing capabilities can shorten time to deployment, but increased competition will also raise requirements for construction progress, reliability, and utilization. For the stock, this model provides upside optionality that is not yet fully incorporated into earnings forecasts, but it may also increase valuation discounts and volatility due to circular revenue, off-balance-sheet-like guarantees, and transaction complexity.

Risks

• Neocloud transactions may create circularity among chip sales, investment guarantees, and cloud revenue, reducing revenue quality and comparability.
• Partner cloud providers with weaker capital strength or higher leverage may expose NVIDIA to direct or indirect credit risk.
• If compute supply exceeds demand, GPU rental rates and utilization may decline and trigger capacity guarantee obligations.
• Transaction sharing ratios, guarantee trigger mechanisms, asset disposal rights, and revenue recognition methods remain unclear, and insufficient transparency may weigh on valuation.
• Data center construction faces constraints in land, power, facilities, fiber, and supply chains, and delayed production ramp-up would weaken project returns.
• AMD, custom ASICs, and cloud providers’ in-house chips may erode NVIDIA’s market share.
• If AI end demand does not materialize as expected, customers may significantly cut GPU purchases and capex.
• Tariffs, export controls, and China market restrictions may have a greater-than-expected impact on revenue and supply chains.
• Intensified Neocloud competition may lower rental rates and compress partner operators’ long-term returns.

What to watch

• NVIDIA’s disclosure of Neocloud contracts’ guaranteed prices, terms, sharing ratios, trigger conditions, and accounting treatment.
• Actual contracted GW scale, construction progress, GPU utilization, and leasing prices for revenue-sharing projects.
• Vera Rubin mass-production ramp, customer quota competition, and performance relative to ASIC and AMD solutions.
• Whether token demand growth continues to exceed the growth of training and inference compute supply.
• Supply bottlenecks in land, power, facilities, fiber, DRAM, NAND, CPUs, HDDs, and optical components.
• Financing ability, delivery timeliness, reliability, and cash flow of partners such as CoreWeave, Nebius, and IREN.
• Whether DGX Cloud Lepton and the DSX platform can effectively connect demand and improve utilization of partner cloud capacity.
• Whether NVIDIA’s capital commitments to fiber, data center leasing, and other physical infrastructure continue to expand.
• Whether sovereign AI projects and data center construction outside the United States can become new scaled spending entities.
• Export controls, tariff policies, and the possibility of NVIDIA recovering China-related revenue.