Roko's Basilisk
Posts
Retreats In The Chip War

Retreats In The Chip War

Plus: crypto jitters, Super Bowl spots, Amazon and Google’s AI budgets.

Roko's Basilisk
February 09, 2026

Here’s what’s on our plate today:

🛰️ AI chips reshape US-China strategy and leverage.
🧩 Bite-sized brains: Bitcoin slide, Super Bowl ads, and AI spending.
🧠 Roko’s pro tip: read chip geopolitics, not benchmark charts.
📊 Poll: How strict should AI chip export rules be?

Let’s dive in. No floaties needed…

The AI Talent Bottleneck Ends Here

AI teams need PhD-level experts for post-training, evaluation, and reasoning data. But the U.S. pipeline can’t keep up.

Meet Athyna Intelligence: a vetted Latin American PhD & Masters network for post-training, evaluation, and red-teaming.

Access vetted PhD experts, deep STEM knowledge, 40–60% savings, and U.S.-aligned collaboration.

_{*This is sponsored content}

The Laboratory

The AI chip standoff: Why Washington and Beijing are both stepping back to stay ahead

In armed conflicts, not every engagement can be won by force, and decisive victories are rare. In these situations, a strategic retreat allows time to regroup, reassess, and re-engage under better conditions.

In modern geopolitics, the idea of a strategic retreat is often used beyond the battlefield and in the lobbies of global power centers to protect interests and ensure continued dominance in spheres of influence.

The U.S. is shifting from outright restrictions on AI chip exports to a more conditional, economically driven strategy to preserve its technological edge. Photo Credit: FT.

So naturally, when a technological advancement like artificial intelligence made its mark on how governments strategize for the future, everything related to the technology became part of broader game plans to maintain the status quo.

This competition is evident in the drive to control one of AI’s most critical components: chips.

Ever since OpenAI kicked off the AI race, the U.S. has been working to control access to advanced AI chips. Consecutive administrations have restricted the export of these chips, especially to China. However, after years of outright bans, the U.S. has pivoted to a more pragmatic and economically viable approach that allows it to control supply while reducing the risk of being surpassed by China.

In January 2026, after months of deliberation, Washington formally allowed the sale of NVIDIA’s second-most-powerful AI chips to China. The sale of these chips was intentionally conditional and can be seen as a strategic retreat by the U.S.

Washington’s conditional opening to China

The U.S. approval of NVIDIA chip sales to China included several conditions, including a 25% fee paid by NVIDIA and AMD directly to the U.S. Treasury.

Additionally, China cannot purchase more H200 chips than the U.S. does. NVIDIA has sold about 2M H200 units domestically so far, which puts China’s maximum at around 1M.

It must be noted that in this deal, the third party is NVIDIA, which stood to lose the most from a protracted ban on the sale of its chips to China. To address this, NVIDIA is relying on large orders from China, which could enable it to net around $8B from just two Chinese buyers, turning export controls into a shared financial upside. The figure assumes that NVIDIA will generate over $10B in revenue from chip sales in China, backed mainly by orders from Alibaba and ByteDance, which were reportedly ready to order more than 200K H200 chips each.

So far, the deal appears unfavorable for China, which may explain why Chinese customs authorities instructed agents not to permit NVIDIA's H200 chips into the country.

Chinese officials also directed domestic technology companies not to purchase the chips unless necessary.

However, regulators soon reversed their decision.

Why did Beijing accept conditional access?

Chinese regulators gave Alibaba, Tencent, and ByteDance the go-ahead to place orders for Nvidia’s H200 AI chips. But they, too, placed conditions on importing NVIDIA chips.

For starters, regulators in China are asking companies to buy domestic chips alongside NVIDIA chips, even though the exact ratios are unclear. The goal is not just technical but political. It allows Beijing to accept foreign technology while protecting local manufacturers like Huawei and forcing AI firms to run parallel systems.

And though there are signs of internal tension, as some reports suggest, customs officials have delayed shipments despite formal approval. China, too, has chosen the path of strategic retreat.

Beijing’s strategy of controlled dependence

China’s commerce authorities reportedly view the US tariffs as a de facto tax and are urging companies to lean more heavily on domestic alternatives. For firms like Alibaba and ByteDance, this creates uncertainty. They may have permission to order chips, but there is no guarantee those chips will arrive.

This ambiguity is deliberate. Washington wants to show that access to advanced technology is negotiable under strict conditions. Beijing wants to signal that reliance on foreign hardware is temporary, even if domestic options are not yet good enough.

That gap still matters. The H200 is far more capable than the downgraded chips China previously received, and for training large AI models, that difference determines whether systems are commercially viable at all.

For China, the retreat buys it time while ensuring access to advanced chips. And it may be some time before the country can rely on domestically produced AI chips.

The structural barriers to semiconductor self-sufficiency

Currently, Huawei’s Ascend chips lag significantly in efficiency and require more power and space to achieve results comparable to those of NVIDIA chips.

An even bigger constraint is memory. China’s domestic production of high-bandwidth memory limits the number of usable AI accelerators Huawei can realistically ship each year, regardless of how many logic chips it produces.

Software is another hurdle. NVIDIA’s CUDA ecosystem represents years of accumulated tooling that developers rely on. Switching to domestic chips means giving that up, and Chinese alternatives have not yet reached maturity. Buying NVIDIA hardware is not just about buying silicon. It is buying an entire development environment.

Then there is DeepSeek, which complicates the picture. The company showed that smarter algorithms can dramatically reduce compute requirements, triggering price wars and rapid competitive responses across China’s AI sector.

Supporters of weaker controls argue this proves hardware restrictions matter less over time. Skeptics counter that DeepSeek relied on older NVIDIA stockpiles and indirect access to banned hardware, and it is unclear whether such efficiency gains scale without continued access to top-tier infrastructure.

So, while the dynamics around advanced AI look complicated in China, allowing NVIDIA chips gives it time to figure things out while continuing to innovate.

While China works to close its technological gaps, the U.S. retains another advantage. Domestic demand for H200 chips is already slowing as customers transition to the newer Blackwell architecture.

This could create an awkward situation for NVIDIA as it must decide whether to push its newer, more profitable chips in the West or keep older H200 sales alive at home to preserve access to the Chinese market.

While the company has said it is not factoring China sales into its forecasts, it remains a commercial entity that prioritizes profit over ideology.

The situation is one of managed interdependence, in which both the U.S. and China seek access to NVIDIA chips without giving the other an advantage.

Managed interdependence

The current geopolitical situation reflects the changing nature of competition. The question is no longer whether China gets advanced chips, but what price it pays and what limits it accepts. Volume caps, tariffs, testing requirements, and domestic purchase mandates point to a new model where access is conditional and negotiated.

This sets a powerful precedent. Future chip generations may follow a similar pattern, becoming exportable only after newer architectures appear, keeping China a step behind without forcing full decoupling. But nothing is guaranteed. Political opposition, customs delays, manufacturing limits, or faster progress by Chinese firms could upend the plan.

For the time being, the age of unrestricted semiconductor markets is over, but total separation never happened. What is emerging instead is managed interdependence, where cutting-edge technology is simultaneously a security concern, a revenue stream, and a bargaining chip. That tension is now baked into the system.

Historical lessons in technology control

Current U.S. restrictions on semiconductor access resemble the trade tariffs of the British Empire, which maintained cohesion while limiting access for competing powers.

Although the U.S. has made a strategic retreat by permitting the sale of less powerful chips to China, it continues to exert influence over global supply chains.

By treating access to AI chips as a negotiable privilege rather than a universal market good, the United States aims to slow the loss of exclusivity while preserving the global system.

China is accepting these constraints for now while investing heavily to prepare for when they no longer apply.

What is emerging is not a clean break or decisive victory, but a prolonged struggle over access: who obtains advanced technology, in what quantity, and under what conditions. History shows that true power lies not in total destruction, but in knowing when to regroup. Currently, both technological powers are regrouping. The outcome of this AI race will depend on which side continues to innovate, invest in infrastructure, and manage the risks of AI dependency.

Roko Pro Tip

💡

Use ‘geopolitics-aware’ risk planning for AI: if any part of your stack depends on U.S.-China-sensitive hardware (NVIDIA, TSMC, etc.), map single points of failure and identify at least one alternative supplier or region for each.

The context to prepare for tomorrow, today.

Memorandum merges global headlines, expert commentary, and startup innovations into a single, time-saving digest built for forward-thinking professionals.

Rather than sifting through an endless feed, you get curated content that captures the pulse of the tech world—from Silicon Valley to emerging international hubs. Track upcoming trends, significant funding rounds, and high-level shifts across key sectors, all in one place.

Keep your finger on tomorrow’s possibilities with Memorandum’s concise, impactful coverage.