Patel explains China can only buy A100-level GPUs from NVIDIA due to export controls, not H100s.
“Well, so NVIDIA made a special version of their a 100 GPU that can ship to China, and so that's that's pretty much it.”
Patel argues the unrestricted 1980i tool can reach five nanometer with only modest cost increases on total chip economics.
“And many folks including, not just myself, but many folks in the industry as well think that the 1980i could be used for five nanometer, right?”
Patel notes the unrestricted 1980i tool continues receiving upgrades through 2023, creating a regulatory loophole.
“But remember the 1980i has continued to be upgraded and there's an upgraded version in 2023, right? And that tool, sort of the line is in between those tool, right?”
Patel argues export restrictions clearly failed because China manufactured advanced Huawei chip demonstrating broader capabilities.
“The export restrictions have clearly failed because China is still able to manufacture. And what's surprising about the chip that Huawei did is, you know, it's a phone chip.”
Patel asserts chip export controls to China are working effectively despite smuggling through third countries.
“The chip side of things is actually being controlled quite effectively, I think. Right? Like, yes, there is, like, shipping GPUs through Singapore and Malaysia and and other countries in Asia to China.”
Patel argues US cannot effectively sanction advanced packaging because the largest company relocated to Singapore from Hong Kong.
“where The US can control China. Right? So advanced packaging capacity is kind of a shot because the vast the the largest advanced packaging company in the world was Hong Kong headquartered.”
Patel says China still receives over a million NVIDIA H20 GPUs annually despite October sanctions.
“Right? Because you have you have sanctions on how many NVIDIA GPUs you can get in. Now, they're still north of a million a year.”
Patel argues US export controls are counterproductive because China can now build better chips domestically than what's allowed for export.
“They can build better chips in China, then we restrict them in terms of chips that NVIDIA or AMD or an Intel can sell to China.”
Patel argues current export controls are flawed because China can build better chips domestically than what West can sell them.
“the equipment that is shipped can be used to build chips that are better than what the Western companies can actually ship them. John, Dylan seems to think the”
Patel identifies the 50,000 GPU cap per country as trivial compared to NVIDIA's six million unit annual production.
“Like, each country can only buy 50,000 GPUs for the next four years. And it's like, that's kinda nothing when NVIDIA's making, you know, 6,000,000 plus this year.”
Patel says Chinese data center operators are relocating to Singapore to circumvent US export controls while building in Malaysia.
“Right? Like, the largest operator in China of data centers in China, GDS, moved to Singapore and says they're Singaporean now.”
Patel explains the 7% rule limits non-US data centers, benefiting only Microsoft, Meta, Amazon, and Google.
“So, like, it's, like, Microsoft, Meta, Amazon, Google. Right? These four companies have, you know you know, 70 plus percent of their data center AI data center capacity in The US.”
Patel says Malaysia will add three gigawatts of data center capacity by 2027, mostly Chinese companies claiming to be Singaporean.
“Malaysia from 2024 to 2027, not the country itself, but companies operating there, mostly Chinese companies, many of them claiming they're now Singaporean companies. Right?”
Patel says Chinese data center companies are building three gigawatts in Malaysia by 2027, claiming Singaporean status.
“Like, the largest operator in China of data centers in China, GDS, moved to Singapore and says they're Singaporean now. It's like but now they're these companies are building three gigawatts of data center capacity.”
Patel reveals each country is capped at 50,000 GPUs for four years while NVIDIA makes 6 million annually.
“There is there is one obvious loophole, which is well, there's, like, strict caps. Right? Like, each country can only buy 50,000 GPUs for the next four years.”
Patel reveals Oracle planned 20 percent of capacity in Malaysia, now blocked by seven percent rule.
“Oracle was planning to have, like, 20% of their capacity in Malaysia. So now it's like, what do I do with these data centers that I was gonna build?”
Patel argues export controls aim to limit AI usage scale in China, not prevent AGI development.
“And I think that is a much easier goal to achieve than trying to debate on what AGI is.”
Patel argues export controls primarily limit AI inference deployment in China, not frontier model training capabilities.
“A large part of export controls, if they work, is just that the amount of AI that can be run-in China is going to be much lower.”
Patel reports ASML had 45% of equipment and more of revenue from China despite export controls.
“On equipment though, last year ASML, many quarters they've had 45% of their equipment and even more of their revenue come from China.”
Patel reports ASML had 45% of equipment and more of revenue from China despite export controls.
“On equipment though, last year ASML, many quarters they've had 45% of their equipment and even more of their revenue come from China. And likewise for ASML, Lam Research and all these companies, right?”
Patel says China imported one million H20 GPUs in recent quarters, enough to build largest cluster.
“This is more than enough if they even took 30 of it, concentrated it, to be a bigger cluster than any American company has.”
“But, like, you know, can you give them no GPUs? No. They're gonna retaliate. Like, there is a middle ground, and, like, Huawei is eventually going to have a lot of production capacity,”
Patel estimates NVIDIA had north of $20 billion in China H20 revenue that had to be written off after ban.
“Our our revenue estimate for NVIDIA in China for just h 20 was north of 20,000,000,000 because that's what they were booking in capacity slash had to write off.”
Patel says China increased lithography imports to 40% of equipment spending to stockpile before bans took effect.
“They were importing lithography at a much higher rate than that. Right? Like, 40% of their equipment imports were lithography, and they were stockpiling lithography equipment.”
Patel says Huawei trained significant models on Ascend chips before 2020 Trump ban took full effect.
“The the full ban. And so they were only able to make a small volume of these chips, but they had trained significant models on these chips that they made then.”
Patel reveals Huawei acquired nearly 3 million chips from TSMC through shell companies resulting in billion-dollar TSMC fine.
“But they were able to acquire 3,000,000 chips, 2,900,000 chips from TSMC through these other entities. Right?”
Patel reports Huawei acquired 2.9 million chips worth $500 million from TSMC through shell companies before getting caught.
“they were able to acquire 3,000,000 chips, 2,900,000 chips from TSMC through these other entities. Right?”
Patel explains TSMC must follow US export restrictions regardless of Taiwan's government due to banking and equipment dependencies.
“The way American export restrictions are upheld is that Taiwan utilizes American banking systems, American equipment industries. And so they'll still have to uphold any US export restrictions.”
Patel says NVIDIA has locked up over 60% of supply chain capacity this year in long-term contracts.
“The NVIDIA has locked up over 60% of the capacity this year in long term contracts alone, and they're buying more on top of that.”
Patel reports both US and Chinese governments unhappy about Chinese open-source models benefiting American companies more than Chinese ones.
“There have been reports in China about the Chinese government not so happy that American companies are profiting more from Chinese open source models than Chinese companies are.”