GB300 volume ramp and Vera Rubin introduction could drive another NVIDIA beat, with China H200 shipments providing an additional upside variable

Overview

This report previews NVIDIA's F2Q27 results and F3Q27 guidance. J.P. Morgan believes that the GB300 rack volume ramp, the on-schedule introduction of Vera Rubin, and potential China H200 shipments will support another "beat and raise," although the share-price reaction is more likely to depend on management's commentary regarding AI compute competition, infrastructure financing, the China business, and memory specification adjustments.

Core views

J.P. Morgan's supply-chain research indicates that NVIDIA's GB300 rack shipments continued to ramp in F2Q27, while residual GB200 shipments also continued, with total rack volumes expected to increase approximately 15% Q/Q to 17,000 to 18,000 units. Following unexpectedly strong growth in F1Q27, Networking growth in F2Q27 is expected to more closely track Compute, with both increasing by approximately the mid-teens percentage range Q/Q. Edge revenue is expected to increase by a mid-single-digit to high-single-digit percentage Q/Q and approximately 20% Y/Y, slowing from 29% Y/Y growth in F1Q27. Taken together, the report expects F2Q27 total revenue could reach $94 billion to $95 billion, up approximately 15% Q/Q and above the market consensus of $92.1 billion. Looking ahead to F3Q27, the report expects rack shipments to maintain a similar pace, increasing 13% to 14% Q/Q to 19,000 to 20,000 units, including initial Vera Rubin shipments of 1,000 to 2,000 units. Vera Rubin's entry into the product mix is expected to increase blended average selling prices by 5% to 10%, driving Compute revenue growth of approximately the mid-teens percentage range Q/Q; Networking is expected to grow at a similar pace. Factoring in some China H200 shipments, J.P. Morgan believes the company could provide revenue guidance of $107 billion to $108 billion, approximately 3%, or $2.5 billion to $3.5 billion, above the market consensus of roughly $104.5 billion. Regarding margins, the report believes there is limited risk to management reiterating a gross margin in the mid-70s percentage range. However, the sharp increase in memory costs raises questions about the long-term sustainability of this margin level. The report also cautions that near-term upside to the numbers may not automatically translate into a clear share-price catalyst: over the past four quarters, the company's revenue guidance has averaged 4% above consensus, but its share price declined by an average of 3% and 5% over the 7 and 30 days following the releases, respectively. Consequently, the market's focus this time is more likely to be on how management explains medium- to long-term competitiveness and key uncertainties rather than solely on the magnitude of the beat. The first key issue is NVIDIA's competitive position relative to alternative AI compute platforms such as ASICs, AMD, and Cerebras. The increasing share of alternative compute platforms within the AI compute total addressable market is expected to remain a persistent narrative overhang, and with multiple projects scheduled to ramp over the next several years, this concern is unlikely to be fully eliminated. J.P. Morgan expects management to emphasize platform flexibility, particularly its applicability to the rapidly growing enterprise AI and sovereign AI markets, as well as the significant generational reduction in per-token costs. The report estimates that Vera Rubin could reduce platform cost per token by approximately 90% compared with Blackwell Ultra, supporting customer adoption and market-share retention. J.P. Morgan continues to believe that NVIDIA will maintain its leadership in the AI compute market, while also expecting GPU and ASIC/XPU shares of the AI compute total addressable market to gradually approach equilibrium over the next several years. The second issue is the recently announced large-scale infrastructure financing agreements. The report expects management to characterize these arrangements as infrastructure enablers rather than demand stimulators, because underlying demand already exists. Through limited asset support and third-party capital, NVIDIA aims to remove bottlenecks in power, data center construction, and financing, thereby helping convert existing AI compute demand into actual deployments. The third issue is the stalled China business. Shipments of H200 GPUs to China, which have been approved but have yet to commence at scale, represent a significant upside variable for F3Q27 guidance and subsequent forecasts. There have recently been reports of initial customer deliveries; J.P. Morgan estimates that every 100,000 H200 units shipped could generate approximately $3 billion in incremental revenue. Therefore, the actual scale and pace of China shipments could materially alter the guidance outcome. The fourth issue is the adjustment of memory specifications for Rubin, Rubin Ultra, and the Vera CPU. Research by J.P. Morgan's global team indicates that Rubin may use two SKUs with either 288GB or 192GB of HBM4; Rubin Ultra has been adjusted from 16-layer HBM4E to 8-layer or 12-layer HBM4E; and the Vera CPU's SOCAMM capacity has been reduced from the previous 1.5TB per CPU to 768GB, a reduction of approximately 50%. The report believes these adjustments are driven almost entirely by tight memory supply: to fulfill its compute product shipment commitments to customers, NVIDIA needs to reduce per-system capacity when available memory is insufficient. The ultimate sales mix of the different SKUs remains unclear, and the effects of the specification changes on revenue growth and margins are also undetermined, making this an expected key topic on the earnings call. Regarding valuation, the report believes NVIDIA's valuation has declined to levels clearly below normal and historical levels, currently trading at approximately 17x and 13x market CY27 and CY28 EPS estimates, respectively. Meanwhile, steadily rising EPS estimates have continued to support the share price; J.P. Morgan expects this relationship to persist as earnings continue to expand. Its $280 price target applies a 28x multiple to NVIDIA's approximately $10 in annualized earnings power upon exiting CY26, with the multiple comparable to its long-term EPS growth rate. Based on these earnings and valuation assessments, J.P. Morgan reiterates its Overweight rating and $280 price target. The longer-term fundamental thesis also includes execution across the company's business segments. The report believes that although the first half is typically weaker than the second half, strong PC gaming demand can remain a revenue driver and offset the structurally declining PC OEM business; the data center business benefits from hyperscale customers adopting GPU-accelerated deep learning to process large datasets. The performance of the automotive and enterprise businesses is also viewed favorably, although whether autonomous driving can achieve strong market adoption remains to be seen.

Analysis framework

J.P. Morgan first uses supply-chain research to estimate GB300, GB200, and Vera Rubin rack shipments, then aggregates shipment volumes, product-mix-driven changes in average selling prices, and the growth rates of Compute, Networking, and Edge into quarterly revenue forecasts. The report then compares its forecasts with market consensus and with guidance beats and share-price performance over the past four quarters, before further assessing AI compute competition, infrastructure financing, China H200 shipments, and memory supply constraints. Finally, the report determines its price target using forward P/E multiples and annualized earnings power upon exiting CY26.

Methodology notes

• Industry/Sector Analysis Framework

  Supply-chain research and rack shipment modeling

  The report uses supply-chain research to estimate quarterly shipment volumes for GB300, GB200, and Vera Rubin racks and derives Compute, Networking, and overall revenue performance accordingly.

• Industry/Sector Analysis FrameworkPrice-Volume Decomposition

  Rack sales volume and blended average selling price decomposition

  The report separately estimates rack shipment growth and the 5% to 10% increase in blended average selling prices after Vera Rubin enters the product mix, then combines the two to derive revenue growth.

• Event-Driven Strategy and Behavioral FinanceExpectation Gap/Expectation Management

  Comparison of company guidance with market consensus

  The report compares quarterly revenue and potential management guidance with market consensus and references share-price performance over the 7 and 30 days following guidance beats in the past four quarters to assess the catalytic effect of an upside surprise in the numbers.

• Competition and Strategy FrameworkMoat / competitive advantage

  Platform flexibility and per-token cost advantage

  The report cites the platform's applicability to enterprise AI and sovereign AI, as well as Vera Rubin's approximately 90% reduction in per-token costs relative to Blackwell Ultra, as the basis for NVIDIA's ability to sustain customer adoption and market leadership.

• Valuation MethodPE/PEG valuation

  Alignment of forward P/E with long-term earnings growth

  The report compares the stock's market-implied CY27/CY28 P/E multiples of approximately 17x/13x and applies a 28x valuation multiple to approximately $10 in annualized earnings power upon exiting CY26 to derive a $280 price target.

Asset mapping & comparison

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

• NVIDIA Corporation (NVDA.US)

  The GB300 shipment ramp, Vera Rubin introduction, and potential China H200 shipments are expected to drive near-term revenue above market expectations, while earnings expansion and a lower forward valuation support the report's positive stance.

  Strengths

  Leadership in the AI compute market, platform flexibility, applicability to enterprise AI and sovereign AI, and Vera Rubin's approximately 90% reduction in per-token costs relative to Blackwell Ultra.

  Weaknesses

  Tight HBM and SOCAMM supply has led to reductions in product specifications, raising questions about the long-term sustainability of gross margins; the PC OEM business is in structural decline, and widespread autonomous-driving adoption remains unproven.

  Comparison

  J.P. Morgan expects NVIDIA to maintain its leadership in AI compute, but GPU and ASIC/XPU shares of the AI compute total addressable market may gradually approach equilibrium over the next several years.

  Risks

  Intensifying AI compute competition, lower-than-expected hyperscale customer adoption of deep learning, weakening PC gaming demand, rising memory costs, and the effects of specification adjustments on revenue and margins.

Key data

• Estimated F2Q27 revenue$94-95BUp approximately 15% Q/Q and above the market consensus of $92.1B
• F2Q27 rack shipments17-18kUp approximately 15% Q/Q, including the GB300 volume ramp and residual GB200 shipments
• F2Q27 Networking growthmid-teens% Q/QExpected to more closely track Compute growth
• F2Q27 Edge growthMSD%-HSD% Q/Q; approximately 20% Y/YY/Y growth below F1Q27's 29%
• Estimated F3Q27 rack shipments19-20kUp 13%-14% Q/Q
• Initial Vera Rubin shipments1-2k racksExpected to be included in F3Q27 rack shipments
• Vera Rubin blended average selling price increase5%-10%Estimated increase in blended ASP after the product enters the mix
• Estimated F3Q27 revenue guidance$107-108BApproximately 3%, or $2.5-3.5B, above the consensus estimate of roughly $104.5B
• Expected gross marginmid-70s%The report expects management to reiterate this level
• Historical revenue guidance beat4% averageRelative to market consensus over the past four quarters
• Historical share-price performance after guidance7-day average -3%; 30-day average -5%Indicates that near-term upside to the numbers may not create a significant positive catalyst
• Reduction in Vera Rubin platform cost per tokenapproximately 90%Estimated reduction relative to Blackwell Ultra
• China H200 revenue sensitivityapproximately $3B/100k unitsEstimated incremental revenue for every 100,000 H200 units shipped
• Rubin HBM4 specifications288GB or 192GBTwo SKU configurations estimated by J.P. Morgan
• Rubin Ultra HBM4E specificationsadjusted from 16-Hi to 8-Hi/12-HiThe report believes this is primarily attributable to tight memory supply
• Vera CPU SOCAMM specifications768GB/CPUA 50% reduction from the previous 1.5TB specification
• Market-implied CY27/CY28 P/Eapproximately 17x/approximately 13xThe report believes this is clearly below normal and historical valuation levels
• Price-target valuation basis28x × approximately $10.0028x applied to approximately $10 in annualized earnings power upon exiting CY26
• Price target$280.00Dec-26 price target, maintained unchanged
• Reference share price$214.72Price as of August 21, 2026
• PC gaming business exposureapproximately 53%Used by the report to illustrate the downside risk from weakening consumer PC gaming demand

Impact & implications

The report believes that the continued GB300 volume ramp, initial Vera Rubin shipments, and potential China H200 deliveries collectively support further increases in revenue and earnings expectations, but the market may not respond significantly merely because the company beats expectations again. More important implications stem from whether NVIDIA can maintain its leadership as alternative AI compute platforms expand, whether it can remove power and data center financing bottlenecks, and whether memory supply constraints will impair product revenue and gross margins. The lower forward valuation and steadily rising EPS expectations constitute the report's primary valuation basis for maintaining its Overweight rating.

Risks

• Significant increases in memory costs could weaken the long-term sustainability of gross margins in the mid-70s percentage range.
• Specification reductions caused by tight HBM and SOCAMM supply could affect revenue growth and margins, with the actual impact depending on the ultimate SKU mix.
• Expansion by alternative AI compute platforms such as ASICs, XPUs, AMD, and Cerebras could result in NVIDIA securing fewer deployments and less market share than expected.
• If hyperscale customers reduce their adoption of deep learning, data center revenue and earnings forecasts would face downside risk.
• Macroeconomic uncertainty could suppress PC gaming demand; the report states that NVIDIA's approximately 53% exposure to the PC gaming business makes consumer weakness a valuation and earnings risk.
• Uncertainty remains over whether autonomous driving can achieve strong market adoption.

What to watch

• Watch whether F2Q27 revenue reaches $94 billion to $95 billion and whether F3Q27 guidance reaches $107 billion to $108 billion.
• Watch whether the GB300 volume ramp, residual GB200 shipments, and initial shipments of 1,000 to 2,000 Vera Rubin racks materialize as planned.
• Watch the actual scale and launch pace of China H200 shipments, with each 100,000 units corresponding to an estimated approximately $3 billion in incremental revenue.
• Watch how management responds to AI compute competition from platforms such as ASICs, XPUs, AMD, and Cerebras.
• Watch whether the large-scale infrastructure financing agreements can remove power, data center, and financing bottlenecks.
• Watch the SKU mix following the memory specification adjustments for Rubin, Rubin Ultra, and the Vera CPU, as well as their effects on revenue and margins.
• Watch management's commentary on gross margins in the mid-70s percentage range and rising memory costs.