Here’s what’s on our plate today:

• 🧠 China’s Nvidia ban and why the chip war just escalated.

• ⚡️ ChatGPT’s teen safeguards, AI agents in Teams, and Notion’s automation.

• 🐊 Roko’s tip: Chips to ecosystems—CUDA, stacks, and cloud workarounds.

• 🗳 Monday Poll asks if the U.S. should keep easing exports to China.

Let’s dive in. No floaties needed…

The Laboratory

Decoding China’s Nvidia ban

Before the launch of artificial intelligence models, Nvidia was known for being the company that powered realistic explosions and lifelike depictions in video games. Its CEO, Jensen Huang, was not the global figure he is today, and Nvidia chips were mostly used in gaming PCs. However, today, the company is at the center of the AI revolution, designing chips that underpin most AI applications.

Nvidia’s meteoric rise has made it a bargaining chip in geopolitical competition. That reality became clear when China’s internet regulator banned domestic firms from buying Nvidia’s AI chips. The ban ordered by the Cyberspace Administration of China (CAC) targets chips like the RTX 6000D, which were tailor-made for the country. The timing of the ban has also raised eyebrows, as it came right around the time Nvidia was getting ready to sell its chips in China after months of negotiations between the U.S. and China, and the company.

China has historically been an important market for the chipmaker. However, Nvidia’s overall sales in the country have been declining. Sales in China contributed 21.4% of the company’s overall sales in 2023. However, by 2025, China’s share has dropped to just 13%. The drop is not just due to the export restrictions imposed by the U.S., but also because China pushed for domestication of its advanced chip supply.

U.S. rules target China’s AI ambitions

Between 2018 and 2025, consecutive U.S. administrations rolled out sweeping rules restricting advanced AI chips, supercomputing items, and support by U.S. persons to China.

The initial rules were announced in October 2022, and further tightened in 2023 to close loopholes. In January 2025, the Bureau of Industry and Security (BIS) issued new interim rules making the export of advanced chips even more cumbersome by adding due diligence and a broader AI diffusion framework.

For the Americans, these measures were a way to block military AI applications (autonomous systems, hypersonic missile simulation, codebreaking) from getting into the hands of the Chinese. However, seeing the increasing pressure from the U.S., China, and Nvidia, they found workarounds.

To circumvent the restrictions, Nvidia developed ‘China-specific’ variants like A800/H800. The company’s CEO also met U.S. President Donald Trump to explain his position and secure continued sales in China.

The U.S. eased some restrictions and allowed the company to sell scaled-down versions of its chips. As part of the easing, the Trump administration proposed that chipmakers like Nvidia and AMD pay 15% of revenue from sales of some advanced chips in China.

Following an understanding between Nvidia and the Trump administration, the U.S. Commerce Department started issuing licenses to Nvidia to export its H20 chips to China.

However, the numerous hoops Nvidia had to jump through to be able to sell its chips and the subsequent easing seem to have ticked off the Chinese government. Beijing grew suspicious of Nvidia’s security measures, especially in light of U.S. lawmakers calling for advanced AI chips to have either software backdoors, killswitches, or location tracking features.

Security fears put Nvidia under scrutiny

In July 2025, Reuters reported that Nvidia placed orders for 300,000 H20 chipsets with contract manufacturer TSMC to meet expected demand from China. At the time, it was reported that the company asked Chinese companies interested in purchasing Nvidia H20 chips to submit new documentation, including order volume forecasts from clients.

The request for information did not sit well with Chinese regulators. The CAC explained it was concerned by a U.S. proposal for advanced chips sold abroad to be equipped with tracking and positioning functions. The regulator even summoned Nvidia to a meeting to explain if its H20 AI chips had any backdoor security risks, as it was worried that Chinese user data and privacy rights could be affected. A backdoor risk refers to a hidden method of bypassing normal authentication or security controls.

Nvidia, in response to concerns, published a blog post reiterating that its chips did not have backdoors or kill switches and appealed to U.S. policymakers to forgo such ideas, saying “it would be a gift to hackers and hostile actors”.

However, the damage was done, and Chinese regulators began to caution tech firms (like Alibaba, Tencent, ByteDance) to avoid or suspend purchases of Nvidia’s H20 chips, especially for government or national security-related work, due to these security doubts.

The recent ban has taken things one step further, threatening to completely shut out Nvidia from the Chinese market and derailing the U.S.’s efforts to block China’s AI ambitions.

China’s focus on local chips

On the surface, China cites security concerns as the main reason for banning Nvidia. However, another reason for banning Nvidia chips could also be that the company has been barred from selling its high-end AI chips in the country, and the performance of its older generation of chips is no longer unmatched.

According to a Reuters report, Huawei launched its latest AI chip, the Ascend 910C, in the first quarter of 2025. Its successor, the Ascend 950, will be launched next year and will come in two variants. That will then be followed by the 960 version in 2027 and the 970 in 2028.

By reducing reliance on Nvidia, China hopes to boost domestic production. Chinese authorities have reportedly pushed domestic tech companies to turn to domestic suppliers like Huawei, Alibaba’s T-Head, Biren, and Cambricon.

A large rollout in Xining, the western province of Qinghai, by China Unicom and T-Head shows it’s possible to build AI systems at scale without Nvidia. Meanwhile, Huawei has begun preparing much bigger Ascend-based systems, expected in 2025 and 2027.

The focus isn’t only on the chips themselves but also on the full ecosystem: software, interconnects, and ready-made system designs. Government policies and demonstration projects are meant to convince big cloud providers that domestic solutions can handle training and inference tasks, even with U.S. export limits. Reports suggest some local chips already match or beat the cut-down Nvidia versions allowed in China. Analysts note that U.S. controls increase costs and slow development, but they also push China to become more self-reliant.

All these measures appear to have been counterproductive when viewed from the Western perspective.

However, the question of breaking away from reliance on Western technology is not a binary one. China still has areas that need to be addressed before it can become self-reliant.

Why China’s goal of self-reliance remains distant

Though U.S. restrictions on physical chips still bite, China can still access advanced computing through cloud service rentals abroad. Chinese firms can rent compute from U.S. or allied data centers running Nvidia H100s, sidestepping export restrictions. Washington has started floating rules to regulate ‘compute as a service’, but this is early-stage and politically tricky.

Currently, Nvidia grabs headlines, but the real choke points are in EDA software (design tools) and the lithography technology needed to mass produce high-end chips, which are controlled by ASML in the Netherlands. Another area where China struggles is advanced packaging.

As of now, if China can keep scaling at mature nodes (7nm+), local chips will be good enough for many AI tasks, but catching up on 3nm/2nm without foreign tooling is still a challenge.

While China may manage to replicate AI chips, CUDA, Nvidia’s proprietary development environment, still poses a major challenge. Most tech companies currently using Nvidia chips rely on the platform, and shifting to a new platform would be expensive. There is also the concern that Huawei’s CANN or Alibaba’s Pingtouge toolchains may not be able to match CUDA-level stickiness.

All these factors, and how China plans to tackle them, will determine if this pivot works.

The limits of decoupling

Chinese authorities appear to have a clear long-term strategy in place to reduce their reliance on Western technology. At the same time, the U.S.’s strategy to use Nvidia as leverage against the Chinese appears to have hit a roadblock.

China’s purchase ban is a political and industrial policy decision, not a mere procurement tweak. It accelerates a demand shift away from Nvidia that, if sustained, could neutralize one of the U.S.’s strongest pressure levers. The U.S. can still curb China’s progress, but only by evolving from blunt trade restrictions to smarter, coalition-backed controls while doubling down on its own innovation.

If Washington’s strategy is to keep playing just the Nvidia card, Beijing’s play is straightforward: don’t buy the card.

Meanwhile, Nvidia, the company that once sat on the sidelines, is now caught between two of the most powerful nations on the globe. How Huang manages the company’s progress while walking the tightrope between geopolitical tensions will determine if Nvidia can retain its crown.

For Nvidia, in the short term, the damage may be limited to corrosion of its stock value; however, in the long run, the company may risk damaging its pricing strength due to slipping grasp of scarcity and irreplaceability.

Roko Pro Tip

Bite-Sized Brains

• ChatGPT moves on teen safety: OpenAI is developing age verification tools to limit underage access after a teen suicide linked to its chatbot.

• Microsoft debuts Copilot agents: New Teams update brings personalized AI agents that automate tasks like meeting follow-ups and info synthesis.

• Notion enters the AI agent race: Notion just launched agents to automate data analysis, task workflows, and even external tool integrations.

Monday Poll

Meme Of The Day

Rate This Edition