Nvidia has required many of its cloud partners outside China to keep their Chinese customers below a certain level, which has shifted between 30% and 49%, though the rule wasn’t written into any contracts, people familiar with the matter told The Wall Street Journal in a report published on 14 November. The WSJ reported that Nvidia began to limit China ties last year.
In Malaysia, which data-center industry leaders told the WSJ could become the second-largest data center hub in the world after Virginia, the state of Johor went from hosting 10 megawatts of data-centre capacity half a decade ago to 2.1 gigawatts today, according to a 2026 report by the ISEAS–Yusof Ishak Institute in Singapore. That is 210 times, by our arithmetic, and the same report puts Malaysia’s data-centre investment at more than $34 billion.
Business Pill · GPUs Explained
A one-minute explainer of what GPUs are and why AI training depends on them. It teaches the general idea only and does not discuss Nvidia’s partners, Johor or this report.
The key insight: As we read the two sources together, the cap people familiar described to the WSJ was a company-set limit on a channel that kept growing: Nvidia kept many partners’ Chinese customers below a level between 30% and 49%, while Johor’s data-centre capacity rose 210 times in about five years, by our arithmetic from the ISEAS report.
Beyond the customer cap, Nvidia also urged some cloud partners to weed out Chinese nationals from top management and push out Chinese investors, the people told the WSJ. Cloud providers are now looking at regions other than Southeast Asia, such as Japan, Brazil, Central Asia and Europe, to build capacity that can serve Chinese customers, the WSJ wrote.
The WSJ article quotes no Nvidia comment on the cap. It quotes an Nvidia representative as saying: “America cannot win the AI race unless America’s AI infrastructure wins worldwide”.
In Southeast Asia, “Chinese demand is responsible for the massive scale of data centers”, James Chai, the ISEAS researcher who wrote the report, told the WSJ. Data centres outside China can import Nvidia chips and rent the computing power to Chinese customers, the WSJ wrote.
Data-centre capacity hosted in Johor, Malaysia: about 10 megawatts in early 2021 and 2.1 gigawatts in 2026, from the ISEAS–Yusof Ishak Institute report Behind Malaysia’s Data Centre Boom. The 210 times figure is our arithmetic from the report’s rounded numbers.
Johor’s Build in the ISEAS Numbers
The ISEAS report, Behind Malaysia’s Data Centre Boom, says most of Malaysia’s data-centre investment takes place in Johor, the state just above Singapore. Its appendix dates Johor’s roughly 10 megawatts to early 2021. Malaysia’s share of capacity in colocation sites, where tenants bring their own servers, is 59%, against a Southeast Asian average of 56%, the report says.
The report cites GDS Holdings, whose international arm is now DayOne: on its second-quarter 2024 earnings call, the GDS chief executive said the Southeast Asia client mix was around 70% from China, from three clients. The report says once colocation is counted, “Chinese workloads may rival or exceed American ones”.
Power is the constraint the report flags. The 3.8 GW consumed by data centres completed so far equals 13% of Peninsular Malaysia’s system capacity, it says, and the national water regulator approved fewer than 18% of data-centre water applications.
The Operators the WSJ Named
Aolani Cloud, a Singapore cloud provider and Nvidia partner, funded some $2.5 billion in hardware in Malaysia by the beginning of this year, with ByteDance as its anchor customer, the WSJ reported. That is about 7% of the $34 billion investment figure, by our arithmetic. Aolani says it has systems in place to strictly adhere to US regulations, the WSJ wrote.
A single technology company accounts for nearly 70% of DayOne’s business, according to a recent DayOne filing cited by the WSJ, and people familiar with the matter said that customer was ByteDance. Two DayOne sites in Johor rank among the largest AI data centres in Epoch AI’s dataset, as our piece on Malaysia’s tracked AI compute shows. Alibaba plans to use Nvidia-powered servers in a Vena Group data centre under construction in Senawang, people familiar told the WSJ.
What Washington and Kuala Lumpur Are Doing
The Commerce Department is working on rules that would add Malaysia, Indonesia and Singapore to a group of countries subject to export restrictions, people familiar told the WSJ; exporters of chips such as Nvidia’s Blackwell series would need a licence that prohibits Chinese customers from remotely accessing the computing power. The WSJ wrote that the administration revoked the Biden-era chip import limits in May 2025 and promised alternative curbs, and that 18 months later it has yet to put them in place.
Malaysia said on 13 November that it “will not allow itself to be used to get around export controls” and was investigating Megaspeed and its Malaysian affiliate, the WSJ reported. Malaysia is also weighing a law to monitor data-centre chip use, as our piece on the proposed law sets out, and Singapore police are still investigating Megaspeed, as our piece on the Singapore probe shows.
Records seen by the WSJ show Megaspeed arranged nearly $12 billion of Nvidia chip orders for companies in China between March 2024 and June 2025, as our report on the WSJ records sets out; our comparison with the $4.8 billion in a Senate letter puts the two figures side by side. Megaspeed said on 15 November that recent media reports about its business are “not supported by verifiable evidence”, Reuters reported. Our Nvidia hub tracks each step.
The Structural Read
As we read the WSJ report, a cap that was not written into contracts depends on how Nvidia checked it, and the article does not say how compliance was measured or which partners it covered.
As we read the ISEAS report, Johor’s growth makes the host country part of the compliance chain: the report says US export controls increasingly impose compliance obligations on host countries and operators, and the WSJ reports Commerce is drafting licence rules for Malaysia, Indonesia and Singapore.
James Chai, ISEAS-Yusof Ishak Institute, quoted by the WSJ, 14 November 2026
“Chinese demand is responsible for the massive scale of data centers”
Three Implications
FOR NVIDIA SHAREHOLDERS People familiar told the WSJ that Nvidia required many cloud partners outside China to keep Chinese customers below a level between 30% and 49%; the article quotes no Nvidia comment on that cap.
FOR DATA-CENTRE OPERATORS IN MALAYSIA The ISEAS report says the 3.8 GW consumed by completed data centres equals 13% of Peninsular Malaysia’s system capacity, and the water regulator approved fewer than 18% of data-centre water applications.
FOR EXPORT-CONTROL WATCHERS The WSJ reports that Commerce is working on rules requiring licences for Blackwell-class exports to Malaysia, Indonesia and Singapore that would bar remote access by Chinese customers.
The framework says the AI supply chain is “creating geopolitical leverage that operates independently of who has the best models”.
As we read it, Johor is that leverage in physical form: where data-centre capacity sits now shapes who can rent the chips, and the framework does not discuss Nvidia’s partner cap, Johor or this report.
What Is Not Established
The cap: the 30%-49% range comes from people familiar with the matter, and the WSJ says it was not written into contracts. The article does not name the partners it applied to, say how compliance was measured, or give the level in force today.
Johor’s customers: the ISEAS report’s view that Chinese workloads may rival or exceed American ones is its estimate once colocation is counted, not a measured split, and the 70% client mix is a GDS figure for its own Southeast Asia business in 2024. The 210 times figure uses the report’s rounded capacity numbers.
Legality: renting computing power to Chinese customers from data centres outside China is legal under current rules, the WSJ wrote, and neither source says any operator named here broke US export law. FourWeekMBA did not contact Nvidia, the operators or the Commerce Department for this piece.
This Week In The AI Supercycle
AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source
This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.
Nvidia kept many cloud partners’ Chinese customers below a level that shifted between 30% and 49%, people familiar told the WSJ, while Johor went from 10 megawatts to 2.1 gigawatts of data-centre capacity in about five years, 210 times by our arithmetic from the ISEAS report. The Commerce Department is drafting licence rules for Malaysia, Indonesia and Singapore, people familiar told the WSJ.
A note on sourcing. The cap, the operator figures and the regulatory steps come from the Wall Street Journal’s report of 14 November 2026, which we read in full; several rest on people familiar with the matter, as marked. Johor’s capacity, Malaysia’s investment total, the colocation shares, the GDS client mix and the power and water figures come from the ISEAS–Yusof Ishak Institute report Behind Malaysia’s Data Centre Boom (TRS22/26), which we read in full.
Megaspeed’s statement is as reported by Reuters. The 210 times and 7% figures are our arithmetic. FourWeekMBA did not contact Nvidia, the operators or the Commerce Department for this piece.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
