Analyst timothy.arcuri@ubs.com +1-415-352 5676 Natalia Winkler, CFA Analyst natalia.winkler@ubs.com +1-415-352 4626 ab 15 December 2025 Powered by YES UBS Evidence Lab Global Research Cloud Accelerator Study Blackwell Broadens Out, Pricing Holds Up Blackwell broadens out Given ongoing investor concerns about durability of AI demand, we revisit GPU cloud pricing and availability in collaboration with UBS Evidence Lab (>Access Dataset). We think this data illustrates the AI demand environment in general and t ...

Key Takeaways SMCI's AI GPU platforms generated over 75% of first-quarter fiscal 2026 revenues.Super Micro Computer launched new liquid-cooled 4U and 2-OU systems, now ready for volume shipments.SMCI expects revenues to rebound in the second quarter of fiscal 2026 after order delaysSuper Micro Computer’s (SMCI) AI servers are primarily optimized for NVIDIA’s HGX B300, B200, GB300 NVL72 and RTX PRO 6000 Blackwell server systems containing NVIDIA Blackwell GPUs for AI factories, large-scale AI labs, hyperscal ...

President Trump has lifted a US ban on exports to China of Nvidia's H200 chips. The H200 is more powerful than the H20, designed for export to China, but less powerful than the cutting-edge B200. Here's what to know about Nvidia's AI chips https://t.co/JuMPUUmJS8 ...

President Trump has lifted a US ban on exports to China of Nvidia's H200 chips. The H200 is more powerful than the H20, designed for export to China, but less powerful than the cutting-edge B200. Here's what to know about Nvidia's AI chips. https://t.co/Soz6ISRuOX ...

Core Viewpoint - The article discusses the potential release of NVIDIA's H200 in China, analyzing the implications from both the U.S. and Chinese perspectives, focusing on inventory clearance and market dynamics. Group 1: Reasons for U.S. Release - NVIDIA's CEO is advocating for the release of H200 to clear inventory, as the current market is dominated by the B series products, making it difficult to sell H200 in the U.S. [2] - The U.S. data centers are facing power supply issues, and the newer Blackwell architecture is more energy-efficient, leading to a gradual phase-out of older models like H100/H200. [2] - The ideal solution for NVIDIA is to legally sell H200 to China if it cannot be absorbed in the U.S. market. [2] Group 2: China's Attitude - There is a divided opinion in China regarding the release of H200; some believe that domestic AI chips are not yet competitive, while others fear that agreeing to the release could hinder local chip development and give the U.S. leverage. [3][11] - Economically, there seems to be no strong reason for China to ban the import of H200. [4] Group 3: Performance and Market Impact - The performance of H200, particularly in terms of computing power and memory bandwidth, currently exceeds that of domestic AI chips. [5] - Many existing codes are based on the Hopper architecture, making H200 easy to integrate for large companies. [8] - The domestic production capacity for high-end GPUs is not expected to significantly increase until 2027, indicating a continued reliance on foreign technology. [8] Group 4: Implications for Domestic Market - H200 has practical applications for Chinese customers, primarily in training scenarios, while domestic chips are more suited for inference tasks. [12] - The economic benefits of H200 may be limited due to rising memory prices, which could offset any price reductions. [13] - The overall impact of H200 on domestic GPU cards is expected to be minimal, as it does not directly compete with them. [13] Group 5: Market Reactions - The news about H200's potential release has caused market fluctuations, but the actual impact is likely to be limited, with key factors being policy direction, market demand, and funding conditions rather than just technical availability. [14]

Regulatory Landscape - Chinese regulators are preventing ByteDance from using Nvidia chips in new data centers [1] - China has issued guidelines requiring new, state-funded data center projects to use domestic AI chips exclusively [1] - Data centers less than 30% complete must remove foreign chips or cancel purchase plans; more complete projects will be assessed individually [1] - The new guidelines cover Nvidia's H20 chip, as well as more powerful processors like the B200 and H200 [1]

Core Insights - Google is launching an aggressive TPU@Premises initiative to sell its computing power directly to major companies like Meta, aiming to capture 10% of Nvidia's revenue [1][14] - The TPU v7 has achieved performance parity with Nvidia's flagship B200, indicating a significant advancement in Google's hardware capabilities [1][6] Summary by Sections Google's Strategy - Google is shifting from being a "cloud landlord" to a "arms dealer" by allowing customers to deploy TPU chips in their own data centers, breaking Nvidia's monopoly in the high-end AI chip market [2][3] Meta's Involvement - Meta is reportedly in talks with Google to invest billions of dollars to integrate Google's TPU chips into its data centers by 2027, which could reshape the industry landscape [3][5] Technological Advancements - The latest Google model, Gemini 3, trained entirely on TPU clusters, is closing the gap with OpenAI, challenging the long-held belief that only Nvidia's GPUs can handle cutting-edge model training [5][10] - The Ironwood TPU v7 and Nvidia's B200 are nearly equal in key performance metrics, with TPU v7 slightly leading in FP8 computing power at approximately 4.6 PFLOPS compared to B200's 4.5 PFLOPS [7][10] Competitive Landscape - Google's TPU v7 features a high inter-chip connectivity bandwidth of 9.6 Tb/s, enhancing scalability for large model training, which is a critical advantage for clients like Meta [8][10] - Google is leveraging the PyTorch framework to lower the barrier for developers transitioning from Nvidia's CUDA ecosystem, aiming to capture market share from Nvidia [11][13] Nvidia's Response - Nvidia is aware of the competitive threat posed by Google's TPU v7 and has been making significant investments in startups like OpenAI and Anthropic to secure long-term commitments to its GPUs [14][16] - Nvidia's CEO has acknowledged Google's advancements, indicating a recognition of the competitive landscape shifting [14]

Core Viewpoint - The article discusses the potential release of the H200 chip in China, highlighting its specifications and performance compared to domestic AI chips, while also considering the geopolitical context surrounding the decision [2][3][20]. Specifications and Performance - The H200 chip features significant improvements over the H100, including 141 GB of HBM3e memory and a memory bandwidth of 4.8 TB/s, compared to the H100's 80 GB and 3.35 TB/s [9]. - The H200's FP64 Tensor Core performance is 34 teraFLOPS, which is competitive with other high-end chips like the B200 and H100 [18]. Market Context - The article notes that the H200 is currently priced higher than the B200 in certain cloud service providers due to its suitability for high-precision computing and its scarcity [17]. - The usage of H200 in overseas cloud servers remains high, driven by legacy workloads that are difficult to migrate from older cards [19]. Geopolitical Considerations - The potential release of the H200 in China is contingent on the U.S. government's stance, particularly the influence of hawkish advisors [3][20]. - The article suggests that if the U.S. does allow the release of the H200, it is likely that China would follow suit [20].

Core Viewpoint - The article discusses the potential changes in U.S. export policies regarding advanced AI chips, particularly focusing on NVIDIA's H200 GPU and its implications for the Chinese market [4][5]. Group 1: U.S. Export Policy Changes - The U.S. is considering whether to allow the export of NVIDIA's H200 GPU to China, which is more powerful than the currently permitted H20 chip [2][4]. - The H200, released in 2023, is based on the Hopper architecture and is the highest-performing AI chip in its class, although it lags behind the newer Blackwell architecture B200 [4]. - There is no final decision yet on the export approval, and discussions may not lead to actual approvals [4]. Group 2: Historical Context and Current Stance - Under the Biden administration, export controls were implemented in 2022 to restrict advanced AI chip exports to China [5]. - NVIDIA's CEO, Jensen Huang, indicated that the company's sales prospects in the Chinese market are currently "zero" and that efforts are being made to persuade both U.S. and Chinese governments to allow re-entry into the market [5]. Group 3: International Reactions - Former President Trump previously stated that while selling semiconductors to China is possible, the most advanced products should not be sold [4]. - The UK’s Chancellor of the Exchequer, Scott Benton, mentioned that Blackwell chips could only be exported to China after they are no longer considered cutting-edge technology, which could take one to two years [4]. - Recently, the U.S. has shifted its stance by allowing the export of NVIDIA's latest chips to Middle Eastern countries like Saudi Arabia and the UAE [4].

Core Viewpoint - Nvidia is expected to report strong earnings for Q3 FY2026, with adjusted earnings per share projected at $1.26 and revenue estimated at $55.28 billion, reflecting over 55% year-over-year growth [1] Group 1: Data Center Business - The data center business is anticipated to be a key growth driver, benefiting from the increasing adoption of cloud solutions and strong demand for Nvidia's chips in the generative AI and large language model markets [2] - Estimated revenue for the data center segment in Q3 is $48.04 billion, indicating a robust year-over-year growth of 56.1% [2] Group 2: Gaming and Professional Visualization - The gaming segment is showing signs of recovery, with estimated revenue for Q3 at $4.71 billion, representing a 43.7% increase compared to the previous year [2] - The professional visualization segment is also expected to continue its growth trend, with estimated revenue of $678.9 million, reflecting a 39.7% year-over-year increase [3] Group 3: Automotive Sector - The automotive segment is projected to see continued improvement, with estimated revenue for Q3 at $624.8 million, which would mark a 39.1% year-over-year growth [3] Group 4: Generative AI Market - Nvidia is positioned as a leader in the generative AI chip market, with increasing demand across various industries, including marketing, healthcare, and gaming [4] - The global generative AI market is expected to reach $967.65 billion by 2032, with a compound annual growth rate of 39.6% from 2024 to 2032 [4] Group 5: Analyst Sentiment - Analysts from Jefferies and Wedbush expect Nvidia to exceed earnings expectations and raise future guidance, citing strong capital expenditure trends from large-scale enterprises [6] - Bank of America maintains a target price of $275, anticipating assurances from Nvidia executives regarding their capacity to meet demand [7] - Oppenheimer analysts have raised Nvidia's target price, identifying it as the most likely winner in the AI sector [9]