Zijing Wu
@zijing_wu
Technology and all things interesting @FT in HK. Returning to journalism (ex BBG) after a decade as investor/entrepreneur. Still an idealist. Opinions mine.
580 Following 11.9K Followers
This is the first in a three-part series examining China’s breakneck build-out of AI infrastructure as it seeks to become the pre-eminent tech superpower. With @EleanorOlcott and @WillLangley96 from the hinterlands of Ulanqab to Shaoguan
Limited free link
Show more
Both countries fear uncontrollable AI, but rivalry keeps driving them apart. Can shared risk force cooperation on guardrails before it's too late? Our latest from speaking to insiders on both sides:
Can the US and China work together on AI safety?
Show more
People don’t seem to realise: distillation can’t be banned. One can only make it harder while total prohibition is technically impossible, like chip smuggling. Grasp that, and you’ll understand the AI politics on both sides of the pond far better.
Show more
DeepSeek’s fundraise might be the most bizarre in tech history:
* No CFO, no roadshow, and 1 email to submit commitments
* No voting rights, no board seats, and 5Y lockup
* Cleaning up shady SPVs charging 15–40 fees ahead of IPO
Absolute founder mode.
Show more
Latest column: China’s real robot revolution is not about humanoids
Limited free link:
The Ox unveiled. GLM 5.3 flash is multimodal and priced lower than DeepSeek’s V4 flash. The kill zone needs to be updated.
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog:
Available now across all official platforms:
Weights:
API:
Coding Plan:
ZCode:
Chat:
AutoClaw:
Show more
Exclusive: China’s YMTC aims to overtake Samsung & SK Hynix in flash memory race
* Internal meetings set target to take global #1# Nand spot by as early as end of 2027
* IPO will most likely price low at Rmb200-300B while >Rmb1T post valuation is expected
Show more
Great one from Will. The Chinese version of circular financing:
Humanoid robot makers are generating much of their revenue from selling machines to government-backed training centres — which then collect and sell training data back to the robot makers.
Show more
Scoop: China eases limits on Nvidia H200 chips
* ByteDance & Tencent received ~10K H200s; more to expect shipments of similar size
* Beijing tells cos to keep majority in HK (total H200 stock ~500k) but city lacks capacity to support use
Show more
Scoop: China poised to lift travel ban on Manus founders
* Red told team he will soon return to Singapore where the co will continue to operate from independently
* Manus plans major update after deal completion and ARR remains $300m+ post Meta
Show more
”No matter what lies ahead, a minefield or an abyss, I will have no hesitations and will blaze my trail, devoting myself entirely to my people and my country until the last day of my life.”
Zhu Rongji, reforming Chinese premier, 1928-2026
Show more
Dangerous to be on the lower end of the value chain, no matter where you are, esp standardised processes such as IT and customer services. And coding starts fall under that category.
@zijing_wu Interesting to compare and contrast India and the Philippines 👇.
I suspect it has to do with IT services (where India’s services sector leans) vs. customer-facing BPO (where Philippines leans)
Show more
Great reporting on the ground from Chris and Krishn. “In India, a traditional IT job is totally dead.”
The AI threat to India’s IT jobs machine
The Greater Bay Area is becoming real, no longer just a political buzzword. Recently a tech founder told me he’s based in the Bay Area, and it took me a while to figure out he actually meant the GBA… 湾区 vs 大湾区
Show more
Was told it takes only about 30K GPUs to pre train a mega model. Inference compute is the real monster. But then most likely they’ll just serve distilled smaller models.
Chinese AI labs seem confident they have the compute needed to pre-train 5-10 trillion parameter models.
Someway, somehow compute does not seem to be such a major bottleneck, at least when it comes to reaching these levels of model size.
Show more
The next hot IPO may not happen so soon as Moonshot is still required to dismantle its VIE structure
Moonshot shake-up seeks to win Beijing nod for stock market debut
ByteDance targets mega AI model nearing Anthropic’s Mythos
* Close to 10T at early stage of pre-training (Mythos est 8T)
* Multiple Chinese labs training models ~5T (Fable est 5T)
* Seed’s no distillation approach for 1+ years
Show more
Starting a zoo at my office. Could I have a whale next?
Why our civilisation may be at dusk: we turn greatness into privilege, and demand apologies for it.
A rare interview where I found the questions more captivating than the answers. Well done @ZhugeYY
Here is my interview with Christopher Nolan: are you a bard for a civilization at dusk?
