13 quotes from 1 episode on Odd Lots, each with a timestamped link to the source.
13 quotes1 episode
The short version
China is rebuilding its semiconductor stack around domestic hardware to achieve high-tech self-reliance. Kyle Chan pointed out that Huawei connects clusters of lower-performing chips with networking technology while DeepSeek stations staff at Huawei to address software gaps.
Most interesting insights
Huawei tells neural processing unit buyers to source their own HBM from elsewhere.
“At this point, there has even been reporting that Huawei is saying to customers like, 'We can make the NPUs, the neural processing units, but it's sort of BYOH, bring your own HBM. You're going to have to source that from elsewhere.'”
Huawei links clusters of chips to offset single-die gaps
Huawei bundles many chips together using networking technology to make up for single-chip limits. DeepSeek stations staff directly inside Huawei offices to support the hardware integration.
“And in particular, you have a really tight partnership between DeepSeek and Huawei, where they literally have personnel situated at each other's companies.”
“Huawei have been trying to improve not just their capabilities at the single chip level, but try to leverage their expertise as a networking technology company to bundle more of these chips together, to tightly connect them together, and to kind of make up for what they lacked in quality with quantity.”
Equipment bans cap manufacturing capabilities at domestic foundries
SMIC serves as the premier Chinese chip foundry but faces equipment limits from international supply chains. ASML export restrictions on EUV lithography machines set a ceiling on the process nodes domestic chipmakers can reach.
“Huawei works a lot with SMIC. SMIC is sort of like the premier foundry in China for chipmaking. They are kind of like the analog of TSMC, and they are limited by their equipment, by the lithography machines that they have, and by their own supply chains.”
“When that switch happened, when, for example, during the first Trump administration, you got the effective ban on EUV lithography machines from ASML to China, that was a huge wake-up call because that basically capped the ability of Chinese chipmakers to make chips of a certain process node.”
Chinese AI labs want Nvidia silicon while Beijing targets domestic independence
Chinese AI companies seek Nvidia H200 chips to keep up with the US frontier. Beijing accepts the immediate friction, prioritizing high-tech semiconductor self-reliance over short-term economic gains.
“…you see this tension where you have the Chinese AI companies clamoring to get the H200s that Trump allowed to be exported to China.”
“And in this specific case, you have this trade-off between today, can you close the gap with the US AI frontier? Can you narrow that gap? Or are you focused on tomorrow, being able to build a fully self-reliant semiconductor supply chain.”
“If you start with the premise that the goal is not GDP growth in and of itself, the goal is high-tech self-reliance of sort of like winning the industries of the future. If you start with that premise, everything starts to make a lot more sense.”
“Up until now, the main problem was getting Huawei's AI chips, the Ascend series, to be reliable enough and functional enough to be used by Chinese AI companies who had grown up building on Nvidia and the Nvidia CUDA eco- CUDA ecosystem.”
Chinese AI developers like DeepSeek and Zhipu AI are co-designing hardware directly with Huawei, embedding staff inside each other's offices to fix software bugs.
Huawei Ascend chips already run production inference workloads, but historically failed on training runs for 2- to 3-trillion parameter frontier models.
The first Trump administration ban on ASML EUV lithography machines shifted Chinese industrial policy from writing passive checks to forcing a full-stack supply chain rebuild.
SMIC serves as the domestic substitute for TSMC, but equipment bans cap the process nodes it can reach.
How we attribute quotes. Every quote was matched against the episode transcript, so the words and the timestamp are real (we trim filler words like "um", nothing else). The name comes from our written summary of the episode, and we use it only when a separate check of the captions finds that person on the episode. YouTube gives us no voice-by-voice transcript, so open the timestamp to hear who is talking. See a wrong name? Tell us and we fix or remove it.