TL;DR
· Citigroup maintains a Buy rating and a $300 price target for Nvidia, with still about 47% upside versus the July 8 closing price.
· The meeting minutes say AI demand remains strong; the Rubin roadmap has not been materially affected, and CPO has entered the early production stage.
· “$100 billion per GW” is not a direct revenue forecast; non-cloud demand, CPO adoption, and AI returns still need verification.

After communicating with Nvidia’s investor relations team, Citigroup maintained its Buy rating and $300 price target for Nvidia. Based on the July 8 closing price of $204.12, this target implies roughly 47% potential upside. For an AI bellwether whose market cap is already close to $5 trillion, the market’s focus is not only whether GPUs can sell—they also want to know whether AI capex can continue to translate into revenue, whether the next-generation platform will be delayed, how network and energy-efficiency bottlenecks will be addressed, and whether high gross margins and share buybacks can keep supporting the valuation. It’s worth noting that specific wording such as Rubin, CPO, and revenue intensity per GW mainly comes from Citigroup’s communication minutes, and not from a complete set of disclosures officially released by Nvidia.

Key data for Nvidia’s Buy rating. July 8 close: $204.12. Price target: $300. Expected return: about 47%. Market cap: about $4.94 trillion.

$300 price target first bets that demand hasn’t loosened

Citigroup’s $300 price target still fundamentally rests on its growth expectations for the data-center business. Its valuation framework shows the target price is based on FY2027 earnings expectations and valuation multiples close to Nvidia’s average over the past three years. In other words, this view does not rely mainly on higher valuation multiples; it bets that the scale of earnings still has room to be revised upward.

The demand-side language is on the bullish side. For the Meta cloud plan that the market is watching closely, Nvidia did not provide specific commentary at the customer level, but in the conversation it emphasized that overall demand remains strong, and that the company’s current priority is still to do its best to meet customer needs.

The demand mix is also changing. In the past, AI infrastructure deployments were primarily led by hyperscale cloud providers. In the past two years, demand from AI labs, sovereign nations, and enterprise on-prem deployments has increased. The assessment in the meeting minutes is that as “physical AI” applications develop, the market for non-hyperscale cloud providers could become larger in the future.

This remains an expectation, not results that have already been realized. For investors, the more direct question is whether AI labs, sovereign AI, and enterprise on-prem deployments can produce continuous orders—not just provide interim supplementation outside the peak of cloud providers’ capex cycles.

Rubin delays don’t change the roadmap; CPO moves to early production

The market’s concern about Nvidia’s roadmap is concentrated on whether the new-generation GPU platform and system interconnect方案 can be advanced as planned. The Citigroup meeting minutes say management stated that delays related to Kyber Rubin Ultra do not affect the overall roadmap, and that the NVLink domain configurations shown at Computex have not changed.

This is critical for Nvidia. AI server competition is no longer just about the performance of a single GPU; the cost of training and inference is jointly determined by a full rack, full cluster, network interconnect, and energy efficiency. If the roadmap suffers a clear delay, cloud providers’ procurement timing, customers’ migration plans, and the market’s view of Nvidia’s long-term gross margin would all be affected.

Another focus is CPO, i.e., co-packaged optics technology. By bringing optical interconnect closer to the chip package, it reduces power consumption and latency for high-speed data transmission, and is seen as one of the key technologies for the future expansion of AI clusters. Per Citigroup’s minutes, Nvidia said CPO has entered volume outside production with Spectrum-X, and customer willingness to adopt it is high.

But this still isn’t full commercial deployment. The minutes mention that clearer choice may begin with the FY2028 Feynman platform, at which point customers could choose NVLink combined with CPO, or continue using copper cabling. In other words, CPO is moving from concept validation toward early production, but the final adoption mix, cost advantages, and the pace of large-scale delivery by customers still need further disclosures.

This phrasing is more constrained than simply emphasizing a “CPO breakout.” It shows Nvidia is laying groundwork for next-generation AI clusters in advance, but the technical route can be chosen; it doesn’t mean customers will immediately switch at scale.

“$100 billion per GW” can’t be taken directly as a revenue forecast

The number most likely to be misread in the conversation is the long-term “$100 billion per GW” comment that Jensen Huang previously made. Citigroup’s minutes explain that this should be understood more as a long-term trend in efficiency and revenue intensity, rather than a revenue forecast that can be directly applied to any single year.

At the current level, it is about $30 billion to $40 billion per GW. If GPU energy efficiency continues to improve with each generation, then under the same power constraints, stronger compute capacity can be deployed, and the revenue density generated by AI infrastructure could rise as well. The minutes note that Blackwell GPUs improve power-and-efficiency by 25 times versus Hopper—one of the important foundations for higher revenue intensity per GW.

This logic has two prerequisites. First, chips, networks, cooling, and system design must continuously improve compute per watt. Second, customers must be able to earn sufficient returns from AI applications and be willing to turn higher performance into more capital expenditures. If AI investment returns improve, revenue intensity per GW could move upward. If monetization from applications is below expectations, the timeline for achieving this long-term goal would be extended.

Therefore, “$100 billion per GW” is more like Nvidia’s depiction of long-term infrastructure efficiency, and should not be interpreted as a guaranteed revenue target.

Gross margin and buybacks support the valuation, but high growth must still be delivered

On the financial side, Nvidia’s latest official guidance indicates that for Q2 FY2027, the gross margin is expected to be in the mid 70% range; GAAP gross margin is 74.9%, and non-GAAP gross margin is 75.0%, with variations of 50 basis points up or down. For a company with a market cap close to $5 trillion, whether high gross margins can be maintained is an important condition for the market to keep assigning a high valuation.

Capital return is also placed in a more prominent position. Nvidia’s Q1 FY2027 returned about $20 billion to shareholders, and authorized an additional $80 billion for buybacks; the quarterly dividend was raised to $0.25. The Citigroup minutes also mention management’s goal for this year to return 50% of operating cash flow to shareholders.

The recent issuance of $25 billion in bonds has also drawn attention. Public filings and media reports indicate this is Nvidia’s first bond issuance since 2021. Company and market interpretations lean toward improving financial flexibility, rather than signaling worsening cash flow.

Risks have not disappeared as a result. The downside factors Citigroup listed include share loss driven by competition in the gaming business, slower adoption of the new platform, volatility in auto and data-center sales, and changes in crypto mining demand. For Nvidia today, the biggest constraint is still whether AI infrastructure construction can continue to progress at high intensity—and whether it can be converted into real customer revenue.

Bold forward-looking vision helps, but the short term still depends on orders and delivery

Beyond the main storyline, Nvidia has also addressed several frontier and adjacent topics. In recent official releases, the company published open models such as Nemotron, Cosmos, and Alpamayo, aiming to help sovereign nations and enterprises accelerate AI adoption rather than directly competing with the most cutting-edge closed-source models.

Space computing is also part of the long-term narrative. Nvidia has officially released the NVIDIA Space-1 Vera Rubin module for orbital data centers and space computing. Similar directions still face significant engineering challenges, and in the short term it is not the main basis supporting the $300 price target.

On external technology integration, the Citigroup minutes mention that Nvidia has not officially announced a partnership with d-Matrix yet, but management said the company is always willing to evaluate and integrate external technologies. Such wording still belongs to the scope of the communication minutes and cannot be taken as proof that a specific collaboration has already been implemented.

With Citigroup offering nearly half of the upside, the core support still comes from AI demand, the platform roadmap, network upgrades, and high gross margins. What the market ultimately wants to see is the detailed adoption of CPO, the delivery pace before the Feynman platform, whether orders from non-hyperscale customers can continue to increase, and whether revenue intensity per GW can be confirmed by real AI application returns.

Click to learn more about律动BlockBeats when joining hiring positions

Welcome to join the official律动 BlockBeats community:

Telegram subscription group: https://t.me/theblockbeats

Telegram discussion group: https://t.me/BlockBeats_App

Twitter official account: https://twitter.com/BlockBeatsAsia