Few points on $NVDA, lift of H20 ban, H20 and B30 upside, and policy implications 🧵
1) The surging AI inference and compute demand from China: Much of Chinese compute and inference has yet to be reflected in $NVDA ’s reported numbers.
I believe given the pace of Chinese AI development, I think the compute and inference demand in 2H25 and Q1 2026 will be even greater than late 2024 and Q1 2025 (we see how compute growth plays out in 1H25 globally)
2) H20 Upside (rough estimate): If we assume $NVDA ships 500K units (fair estimate I think) in the coming quarter at an average ASP of $10K–$12K, that translates to 5-6 billion in potential revenue.
3) B30 Upside (rough estimate): Assume $NVDA ships 1.5 million units (some estimate from the street is 2M) over the next 12 months (roughly 375K per quarter), and assuming an ASP of $7,000, that equates to $2.625 billion in additional quarterly revenue.
Together, that is about quarterly 7.625B - 8.625B upside revenue for $NVDA in coming quarters. Of course, the number could be different based on the estimate of shipment.
I am optimistic given the significant compute and inference demand coming out from China is quite likely, and if you have been following China's AI development, the demand should be larger than Q4 2024 and Q1 2025.
The shipment of B30 is likely start in Q4 (unlike H20, should be ready to resume shipment anytime). So it should reflect on NVIDIA's revenue likely in Q1 2026 instead of Q4 2025.
4) Policy alignment is VITAL: This move also signals a clearer and more predictable alignment between U.S. regulatory policy and NVIDIA’s stance. This should reduce a key source of uncertainty that has weighed on both investor sentiment and industry planning. Remember, this has been a key contentious point that has yet been addressed since the H20 ban in mid-April.
5) One key question remains in my mind: One question to ask is, since H20 ban is lifted, what does that mean for previous reported control threshold—1.4 Tb/s memory bandwidth and 300 GB/s networking speed? This is a very important question to ask since it will associate with B30 and future chip design flexibility.
6) U.S.-China momentum building: Given the new policy shift, it could also be an effort of momentum building for potential Trump-Xi summit and Trump's visit to China in Q4 2025. I have documented this earlier.
7) Downside risk for Chinese chipmakers: Lastly, Chinese AI chipmakers like Huawei, Cambricon and its associated supply chains are likely to take notable hit. The thesis is no longer there at least for both short term and medium term. end
@UnrollHelper
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Couple signals and development you should aware of regarding U.S.-China tension and negotiation.
1) Chinese state media CCTV reportedly that U.S. and China agree to boost challenges for talks at ALL LEVELS—likely the first time of such wording during Trump 2.0.
2) Secretary Rubio's meeting with Wang Yi in Kuala Lumpur, Malaysia is described as "very constructive and POSITIVE," and further suggested that the Trump-Xi summit is likely and stated "there's a strong desire on both sides to do it."
3) Just not long time ago, Bloomberg also reportedly that Commerce Department is exploring CEO's interest in accompanying Trump to visit China in September (this is, to my understanding, TRUE).
4) All of these development come right after: a) U.S. lift control on EDA software, ethane, jet engine and b) China reportedly lift rare earth control, as well as c) U.S. and China reach trade agreement without disclosing specific details few weeks ago.
5) Taken together, while structural and high level landscape between U.S and China remained unchanged, aka strategic competition, there are enough signals that a) U.S. and China are working closely to address trade tension and b) there is a real intention here to path way for potential Trump-Xi summit in 2H25.
Lastly, few key events to watch beside possibility of Trump's visit in China. APEC Summit (October 28-31 in South Korea) and G20 Summit (November 22-23 in South Africa).
That being said, we still need to monitor what's coming in the next few months, particularly on whether China has fulfill its pledge (reportedly) on rare earth control and whether there is new control measure (adjusted diffusion rule, chipmaking equipment control, etc) coming from U.S., which both could derail the "atmosphere/momentum building" so far.
Wow. According to WSJ, Mark Zuckerberg is spending his days firing off emails and WhatsApp messages to recruit to best AI talent, and reportedly reach out to HUNDREDS of researchers, scientist, and infrastructure engineers, etc to recruit them to new Superintelligence team. /1
"Some of the people who have received the messages were so surprised they didn’t believe it was really Zuckerberg. One person assumed it was a hoax and didn’t respond for several days." /2
And Meta’s chief executive isn’t just sending them cold emails. Zuckerberg is also offering hundreds of millions of dollars, sums of money that would make them some of the most expensive hires the tech industry has ever seen. In at least one case, he discussed buying a startup outright. /3
So before we dive in, let's start with a simple HBM 101 class—What is HBM?
HBM stands for High Bandwidth Memory. It’s a special kind of DRAM, stacked vertically and connected to processors through tiny wires inside the silicon called TSVs (through-silicon vias). TSV allows the direct connection of multiple HBM DRAM dies to increase overall memory bandwidth. /2
So what HBM provides and its advantage over traditional DRAM? Simple: Speed+Efficiency
1) High speed: HBM delivers terabytes per second which is more than 20x faster than regular DDR memory
2) Power efficiency: Because it’s right next to the logic die, energy is saved since data doesn’t have to travel far
3) Area efficiency: Delivers highest capacity per unit area /3
1) Nvidia’s H20 GPU with HBM3 had become the most sought-after accelerator in China amid rapidly increasing inference and computing demand. Prior to the new restrictions, shipments were projected to reach 1.4 million units in 2025.
2) CXMT is now only 3-4 years behind global leaders in high-bandwidth memory (HBM) development, aiming to produce HBM3 in 2026 and HBM3E in 2027 amid notable technological improvements in DRAM.
3) CXMT still faces major roadblocks — these include U.S. export controls on lithography and other equipment, a volatile geopolitical environment, limited access to global markets, and the uncertain pace of technological development against market leaders.
*Thanks for the edits by @jordanschnyc and Lily Ottinger
*The author would like to express sincere appreciation to those who provided valuable feedback on this piece, including @ohlennart (RAND), @SKundojjala (SemiAnalysis), @kyleichan (Princeton University), and @Huang_Sihao (Oxford University).
In December 2024, the U.S. released new export control packages targeting Chinese access to high-bandwidth memory, or HBM, and various types of semiconductor manufacturing equipment, including tools essential for HBM manufacturing and packing. The new rule also added over 140 Chinese chip manufacturers and chip toolmakers to the Commerce Department’s Entity List.
The new rule was designed to further constrain China’s AI development by leveraging the chokepoint on HBM, ultimately controlled by three companies around the world — SK Hynix, Samsung, and Micron. The restriction around SMEs, on the other hand, aims to limit China’s ability to develop its own HBM.
China AI accelerators (Huawei, Biren, Enflame, etc) still relies on foreign HBM.
The Chinese AI accelerators are no exception when it comes to reliance on HBM. For example, Huawei’s latest AI accelerators, the Ascend 910B and upcoming 910C GPUs, are mainly equipped with 4 and 8 HBM2E, respectively, mostly sourced from Samsung before the December 2024 restriction went into affect, with some sourced after the restriction. Similarly, other Chinese GPU makers such as Biren, Enflame, and VastaiTech are likely using HBM2 and HBM2E from either SK Hynix or Samsung. Biren’s BR100 GPUs that launched in 2022 incorporate 4 HBM2E, and Enflame’s DTU released in late 2021 uses 2 HBM2.
Morgan Stanley: "China's DeepSeek Moment" (March 2025). (1/3) — Quite Objective, Concise, and Insightful Observation.
"DeepSeek’s emergence is more than just an artificial intelligence (AI) milestone – it’s a timely symbol of China’s ambition to claim a leadership role in the tech revolution."
Morgan Stanley: "China's DeepSeek Moment" (March 2025). (2/3)
"Beyond financial markets, DeepSeek’s emergence comes at a moment of national pride. The box-office triumph of “Ne Zha 2” and video game “Black Myth: WuKong” have given rise to a grassroots confidence, rather than a top-down directive from Beijing, reinforcing newfound belief in national identity and tradition."
Morgan Stanley: "China's DeepSeek Moment" (March 2025). (2/3)
"As Beijing intensifies its tech ambitions, DeepSeek represents more than just an AI milestone—it’s a symbol of China’s broader resurgence in innovation, investment and global competitiveness."