NVIDIA will announce its quarterly earnings next week, and market focus has shifted from Blackwell to Rubin GPUs.

Author and source: Zhui Feng Trading Desk

Blackwell remains strong, but Rubin is the next valuation narrative.

In the short term, demand for Blackwell remains strong. Microsoft and Amazon’s cloud businesses showed more noticeable improvements during the quarter, while long-term commitments to compute power from companies like OpenAI and SpaceX continue to support market confidence in NVIDIA’s data center revenue.

Morgan Stanley noted that, even before Rubin has made a substantial contribution, NVIDIA's current growth is already very strong; an even greater acceleration may occur after the January quarter.

Rubin is one of the most important keywords in this earnings call.

Jefferies believes that Rubin’s ramp-up speed could be significantly faster than that of Blackwell’s initial phase, as Rubin continues the 72-GPU Oberon NVL72 rack form factor, enabling it to leverage the liquid-cooled data center deployment and manufacturing foundations established with GB200 and GB300.

At the same time, the new design significantly reduces the need for manual wiring and pipeline installation. Management previously mentioned that the assembly time for the computing tray is expected to be reduced from approximately 2 hours to just 5 minutes.

The significance behind this is that NVIDIA is transforming its AI chips into a replicable, scalable "AI factory system."

Jefferies expects VR/R200 to account for approximately 12% of NVIDIA's GPU revenue in Q3 of fiscal year 2027, rise to over 40% in Q4, and become the dominant revenue source in Q1 of fiscal year 2028.

If this pace materializes, the market’s valuation logic for NVIDIA will shift from "How long can the Blackwell cycle last?" to "Can Rubin unlock the next round of growth at a higher base?"

Vera CPU: Not the main course, but completes the "full-stack control"

In addition to the GPU, the Vera CPU is also a notable incremental factor in the financial report.

Jefferies expects NVIDIA's CPU revenue in fiscal year 2027 to reach $20 billion, including Grace and Vera; its estimates show that CPU revenue tied to GPU sales will be approximately $8.65 billion, while standalone Vera CPU revenue will be around $11.35 billion, with annualized revenue potentially nearing $32 billion by the fourth quarter.

In terms of absolute scale, the CPU still cannot be compared with NVIDIA’s AI GPU business and is more of a "nice-to-have." However, its strategic significance is substantial: the Vera CPU enhances NVIDIA’s control over AI server and AI factory architectures through low-power LPDDR5X and NVLink coherence connections with the Rubin domain.

This also puts new pressure on AMD and Intel, especially as large customers increasingly prefer to purchase entire server racks and complete systems—NVIDIA offers a comprehensive computing platform that includes GPUs, CPUs, networking, and software.

OpenAI large order: Secures demand but raises questions about "circular financing"

Of greater market interest is NVIDIA’s long-term computing power arrangement with OpenAI.

According to Jefferies' analysis, NVIDIA has arranged with SB Energy for SB Energy to build, own, and operate the PORTS-Pike Technology Campus in Ohio, leasing it to OpenAI under a 20-year agreement, with NVIDIA serving as the exclusive computing provider.

NVIDIA will provide credit support for the first 4.25 GW of IT capacity and hold an option for the remaining 3.75 GW, totaling up to 8 GW, while investing $1.5 billion in SB Energy.

NVIDIA stated that each generation of system deployment at the facility could correspond to approximately 1.5 million GPUs, generating $150 billion to $200 billion in revenue; OpenAI’s current and planned commitments total approximately 12 GW, scalable to 16 GW, representing about $600 billion in NVIDIA computing purchases by 2030.

This undoubtedly provides a clearer anchor for NVIDIA’s forward orders. However, the market will also ask: how much of this demand stems from customers’ own capital expenditures, and how much requires NVIDIA to drive it through credit support, investment, or revenue sharing?

Morgan Stanley believes that financing, market share, and gross margin are the three key issues currently of greatest concern to investors. NVIDIA’s management will likely emphasize how these structures help accelerate AI infrastructure development, but investors need to see clearer accounting disclosures, including details on contingent obligations, lease and power payment support, and residual value commitments.

This determines whether the market views it as "locking in future revenue" or worries that it becomes "exchanging balance sheet for growth."

Gross margin and cost pressures: Stable in the short term, still to be validated in the long term

NVIDIA's high gross margin has long been a key foundation for its valuation premium. Morgan Stanley expects the company to continue describing its gross margin for fiscal year 2027 as being in the "mid-70s."

However, the pressure is not absent. Costs for HBM, DRAM, advanced packaging, substrates, and front-end wafers continue to rise, and the ramp-up phase of new products is typically accompanied by challenges in yield, testing, and supply chain integration.

Morgan Stanley cautioned that the market's consensus expectation of a significant gross margin recovery in the second half of fiscal 2028 may be slightly overly optimistic.

This means that if NVIDIA provides a more cautious outlook on medium- to long-term gross margins during the earnings call, it could trigger short-term volatility; however, if management can articulate cost locking, pricing power, and system-level sales structure, it could instead serve as an "expectations cleanse."

The biggest bottleneck may not be in the chip, but in the electricity.

As AI capital expenditures progress further, electricity is likely to become the most realistic constraint.

Jefferies noted that the gap between chip demand expectations and grid supply capacity is widening.

Large cloud providers increasingly collaborate with utility companies to build dedicated gas-fired power plants and transmission infrastructure, while cutting-edge AI labs and new cloud providers are increasingly adopting "behind-the-meter" power solutions to bridge the timing gap between grid connection timelines and compute deployment.

This does not mean that AI infrastructure expansion will stop, but it will make the conversion of orders into revenue less linear.

Solutions such as gas turbines, backup power systems, and fuel cells are all being advanced, but equipment delivery timelines, grid connection approvals, local regulations, and community acceptance will impact the pace of data center construction. Nuclear power and small modular reactors are more likely to be variables after 2030.

Therefore, NVIDIA's core challenge over the coming years is whether its customers' data centers, power supply, and financing can keep pace in tandem.

Clarity is the catalyst for stock price movement.

Morgan Stanley maintains its "Overweight" and Preferred Stock rating on NVIDIA, with a price target of $288, noting that NVIDIA's current forward P/E ratio of approximately 17x for FY28 is not expensive relative to its historical levels, semiconductor peers, or the broader market.

However, a low valuation alone does not necessarily lead to an immediate price increase. Over the past four quarters, even after delivering strong earnings, NVIDIA's stock continued to react negatively the following day, indicating that the market has come to view "beating expectations" as the default script. What will truly drive valuation recovery going forward is whether management can clearly answer four key questions:

1. Beyond the October quarter guidance, will the January quarter see a significant acceleration due to Rubin;
2. Is Rubin's ascent smooth enough to maintain product leadership?
3. How significant are the financing obligations behind long-term agreements such as OpenAI;
4. Will power, regulatory, and cost pressures affect the delivery schedule after 2027?

Overall, NVIDIA’s earnings report is likely to remain strong. But the market has moved to the next phase: it’s no longer enough to see demand—it’s about how that demand is fulfilled by supply, power, financing, and margins.

NVIDIA's story remains grand, and its short-term performance is supported; however, as AI infrastructure investment reaches the trillion-dollar level, investors must evaluate whether an entire industrialized AI system can sustainably operate.