Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Glenn Profile picture
Glenn
@GlennLuk
Co-founder/Director @HealthCareInc | Previously @Catalyte_io | VC/PE @Investcorp Technology Partners — Tech | Economic Development | Investing | Greater China
13 subscribers
Jan 29 • 22 tweets • 5 min read
One focus area for DeepSeek was going "lower level" with PTX.

Here's a nice explainer on CUDA/PTX 👇.

While this is an admittedly arcane/technical subject, it actually provides critical insights into:

1⃣ the development cost question
2⃣ export controls
3⃣ future trajectories First some quick background: PTX is a low-level coding language analogous to Assembly.

This is "deep" software engineering: you are now talking in the native language of the machines, with the thinnest of abstraction layers.

I discussed more here, too:

Jan 28 • 23 tweets • 5 min read
If people freaked out about DeepSeek wait until they hear about how it is scaling inference. “Third of the cost” is

1) Likely conservative, and

2) Didn’t require EUV
Jan 25 • 9 tweets • 3 min read
Three reasons why this is the wrong take:

1⃣ Continent-sized economies follow different economic laws (specialization vs. self-sufficiency)

2⃣ Optimal sector allocation

3⃣ China's STEM workforce is young 1⃣ Continent-sized economies follow different economic laws (specialization vs. self-sufficiency)

Larger economies can be more self-sufficient (pursue higher levels of autarky) without losing the benefits of specialization.

Jan 25 • 14 tweets • 5 min read
This is a key point about EVs. Many think it is about the electrification.

It is really more about the tech (the "I"): software-defined UX + tech component integration.

Keep in mind this comes from CATL's Co-Chairman, whose batteries are tied to the electrification aspect. A prime example of this is how PHEVs/EREVs are the fastest-growing segment in China right now.

They are built on the same modern platforms as BEVs, where the ICE is simply another integrated component to extend range beyond battery-only.

Jan 25 • 13 tweets • 4 min read
This debate with Pettis on Germany is interesting because you see many of the same flawed arguments when Pettis talks about China in the same light. Yalcin points out that Pettis focuses just on the goods trade surplus while ignoring the intangibles trade in services and tech/IP, where the U.S. runs huge surpluses.

This makes surpluses in “surplus countries” look much bigger than they truly are.
Jan 24 • 15 tweets • 3 min read
Despite all the hoopla over Deepseek, this whole story speaks more to China's broader AI ecosystem and accumulated human capital than a single company. While this has brought well-deserved accolades to High-Flyer / Deepseek + will make it a magnet for elite AI talent, I suspect what will be more impactful in the LT are the new creative use cases that can take advantage of usage costs reduced by more than an OoM.
Jan 23 • 20 tweets • 4 min read
I like to try to understand all sides so let's do "devil's advocate" analysis on this "Moonshot" or "New Manhattan Project".

This means really trying to understand the rationale of its sponsors. The above-framing/marketing is instructive on the overall vision and mission statement of the project.

This is about achieving AGI first on the premise that whomever achieves it wins the second Cold War (just as the atomic bomb / nuclear weapons framed the first one).

And achieving AGI first is itself premised on the importance of scaling total compute resources, which is both a chips and energy infrastructure question.
Jan 22 • 30 tweets • 6 min read
Keep in mind chips & servers are short-lived assets that depreciate quickly; there’s a reason accountants calculate useful lives in the 3-5 year range.

(This is fundamentally different from long-lived assets like rail & power infrastructure w/ 50+ year useful lives) Because of this, the unit economics of such investment demand a much faster return on capital.

The weighted average cost of capital will be much higher — you can’t use a ton of LT debt to finance “compute”. It will have to be mostly equity.
Jan 20 • 5 tweets • 2 min read
This is so dumb because I think it misses the underlying point here about TikTok. It is about control, not economic ownership.

China wants control. So does the U.S.

Not clear what Trump actually wants here with a JV, as Bytedance is already >60% owned by foreign investors. Image If Beijing wants to be a smart ass, it will push for a new bill that limits foreign ownership in domestic media properties to 20% and force the sale of ~40% foreign ownership in Bytedance.

Remember Alipay?

Perhaps it will even do this by offering to exchange that Bytedance stake for the 50% “joint venture” ownership stake in TikTok. That would be rich with irony.
Jan 18 • 14 tweets • 4 min read
You can see some major Dunning-Kruger effect with some TikTok refugees who have gone onto Chinese social media and made "first contact" with a large mass of Chinese netizens in a native Chinese environment. They don't seem to appreciate that what they are seeing is a self-selected and filtered slice of China.

XHS started as a social media app that catered to relatively well-to-do younger urban women.

It has expanded more widely since but its monetization model (e-commerce) evolved out of this core group, which selects a certain type of "lifestyle" content and user base.
Jan 17 • 26 tweets • 7 min read
One shouldn't think of China as a normal-sized country, but a continent.

18th/19th century concepts of the benefits of economic specialization were formulated in an era where most economic activity took place locally (butchers, smiths, merchants etc). It was also before computers, robots, software - which can increasingly substitute human labor, the key production factor.

And it was also before the era of renewables, where in the 21st century any country can be energy self-sufficient with widely available solar.
Jan 16 • 14 tweets • 4 min read
Xiaohongshu is the first time where a mass scale of Americans are getting exposed to the real** China, in a medium where Chinese voices dominate content creation.

** ofc even here it isn't 100%; filtered via moderation + userbase selection bias (well-to-do younger women)Image TikTok was explicitly non-Chinese with Chinese voices walled off in Douyin + high friction costs for foreigners joining the walled garden (need Chinese phone number / loca App store access).

WeChat is not used much outside the diaspora community and more P2P, less "broadcast" focused.

Clubhouse was dominated by English speakers.
Jan 13 • 22 tweets • 5 min read
Key economic feature of renewables is zero marginal cost “free” energy.

Batteries can soak inter-day/week excess renewable power.

But what if we generate excess beyond that? How do we make it useful?

Then we have to look beyond the grid. That’s where green hydrogen comes in. The energy from excess renewables power that the grid cannot immediately (within a week or so) make economically useful can be stored in other useful forms.

Powering electrolysis of hydrogen, a base reagent for other useful forms like methanol, is one of the solutions.
Jan 11 • 13 tweets • 5 min read
Data-oriented thread on size & scope of multinational operations in China:

In 2020, total assets and sales revenues in China each surpassed $3T, equivalent to ~20% of GDP.

They directly employ 12M people paying salaries ~40% higher than average.

All sources in the links.https://www.mckinsey.com/mgi/our-research/the-china-imperative-for-multinational-companies Fortune Global 500 data says the world's 500 largest companies generating ~$41T in total sales in 2024.

Chinese cos make up a quarter (133 to be exact) of this list ➡️~$30T of global revenue by non-Chinese MNCs.https://www.prnewswire.com/news-releases/fortune-announces-2024-fortune-global-500-302213681.html
Jan 10 • 38 tweets • 9 min read
Noah has hit this "Japan can build infrastructure; America should finance it" theme a couple times now.

It is a good jumping-off point to show how it's based on a highly abstracted worldview where the concepts of "money + finance" get fully untethered from real world context. I write from the perspective of someone who started my career dealing with project finance and a wide variety of power + other infrastructure deals in China and throughout the APAC region.

Here is a young me at a gas-fired power plant in Taoyuan in July 2003. Image
Jan 9 • 10 tweets • 4 min read
Vietnam is attracting similar levels of manufacturing FDI than India despite being 14x smaller in population.

India attracts ~$70B of FDI but less than a third is manufacturing. Singapore is the largest FDI investor, followed by Korea, China, HK and Japan.

(Singapore is less Singapore itself but the “ASEAN financial gateway” through which global investment is funneled … a bit like HK for China) https://www.vietnam-briefing.com/news/vietnams-top-investors-and-fdi-beneficiaries-january-to-november-2024.html/
Jan 6 • 55 tweets • 17 min read
I found most of this piece echoes many themes I have been discussing and specifically Barry's read on Xi (1H of interview).

But instead of going through them, I think it's more useful to dive into a handful of areas where I did not just nod along.

thewirechina.com/2025/01/05/bar… 1⃣ Total factor productivity as a "top down" proxy for underlying productivity/efficiency

I am very skeptical about productivity claims based on TFP because I think TFP is one of the most misunderstood (and often, mismeasured) economic statistics, particularly as it relates to China.

Note: I am not saying China's productivity growth / trend is fantastic. I am saying I don't know if one can confidently claim underlying productivity is not improving on the basis of potentially questionable TFP figures.Image
Jan 5 • 30 tweets • 10 min read
It's Sunday morning and once again I have to deal with this recurring nonsense about China's HSR system.

This is going to be a speed round because I have to do many biweekly pilgramage to Costco.

It's 8:50 AM ET right now. Let's see how fast I can get this done. 1⃣ The unit economics of the most successful rail models (e.g. HK, Japan) is not just ticket sales but capturing value from generated foot traffic.

Anytime you see numbers like this highlighted below, they are excluding non-transport revenue 👇

Image
Jan 4 • 9 tweets • 2 min read
I swear Chinese firms see every failed capital-intensive business idea in the West as a challenge/opportunity. LFP batteries

reuters.com/article/techno…
Dec 26, 2024 • 25 tweets • 7 min read
If these numbers are true** the economics could work.

Project estimated to generate close to ~300 TWh (~3% of current total power demand) + cost ¥1T ($137B) to build.

** & notably incl. long-distance transmission to get power to population centers

scmp.com/news/china/sci… At current residential electricity tariffs (~8 cents/kWh), the project would generate ~$22B of revenue per year on $137B upfront project cost, for a ~16% (revenue) return on assets.

For comparison, China Railway (px + freight) runs at ~13% on this metric.
Dec 24, 2024 • 26 tweets • 6 min read
Re: “failure to build a consumption engine”

Periodic reminder that “consumption” defined for GDP purposes is not consumption that regular people are familiar with.

This is the problem with conflating “consumption” used for GDP accounting purposes with real world consumption. First, macro folks should know better that all GDP is eventually “consumed”.

The only diff. b/n “consumption” (aka “household expenditures”) and “investment” (aka “gross fixed capital formation”) is that “consumption” is consumed in the present and “investment” is consumed over time.