The Yield Tax

Beijing's Procurement Rule

If you manage a multi-billion dollar semiconductor fab, your professional life is governed by an unforgiving metric: yield. Yield is the percentage of chips on a silicon wafer that actually work. In the world of advanced processors, the difference between a 90% yield and an 80% yield isn't just an engineering footnote; it is the difference between a high-margin miracle and a very expensive pile of decorative glass.

To keep yields high, fab managers are traditionally very boring people. They want the most proven, sophisticated tools on the planet—usually from the U.S. (Lam Research, Applied Materials), Japan (Tokyo Electron), or the Netherlands (ASML)—because using an unproven machine to carve nanometer-scale transistors is a great way to get fired.

But in China, the definition of a "good tool" has recently changed. It no longer just means the one that works the best; it means the one that was made down the street.

Beijing has reportedly begun enforcing an unwritten mandate requiring chipmakers to use at least 50% domestically made equipment when adding new capacity. The rule isn't in any official policy book, but the procurement process has become a game of "prove you are buying Chinese". Here's Reuters:

China is requiring chipmakers to use at least 50% domestically made equipment for adding new capacity...

The rule is not publicly documented, but chipmakers seeking state approval to build or expand their plants have been told by authorities in recent months that they must prove through procurement tenders that at least half their equipment will be Chinese-made... Applications failing the threshold are typically rejected.

"Authorities prefer if it is much higher than 50%," one source told Reuters. "Eventually they are aiming for the plants to use 100% domestic equipment."

There is a fascinating bit of economic extraction happening here. For decades, domestic Chinese giants like SMIC preferred U.S. or Japanese equipment because, quite simply, it was better. They wouldn't normally give local alternatives like Naura or AMEC a look because the risk to their yields was too high.

By imposing this 50% rule, Beijing has effectively solved the "no one wants to be the first customer" problem. It is a forced marriage between the country's chipmakers and its emerging equipment providers.

But as any engineer at SMIC will tell you, the honeymoon is expensive. The gap between strategic success and commercial reality is measured in the "yield gap." At the advanced 7nm and 5nm nodes, SMIC's yields have been estimated to be as low as 30%. For comparison, industry leader TSMC routinely achieves yields above 90%.

From a certain perspective, China is mandating a massive R&D budget for the Chinese equipment industry, funded entirely by domestic customers who are being forced to use them. If Naura needs to test a new etching tool on a 7nm production line, they don't need to beg for a trial; the government has already told the fab manager they have to buy it.

The results are showing up in the balance sheets. Naura's revenue jumped 30% in the first half of 2025, and they filed a record 779 patents. They are improving because they finally have access to the "wet lab" of a functioning world-class fab.

The cost of this progress, however, is being borne by the chipmakers. They are essentially paying a sovereignty tax. They are forced to accept the risk of lower yields and less efficient production lines today in order to build a supply chain that Washington can't turn off tomorrow.

The U.S. government often talk about export bans as a way to starve Chinese industry. But the 50% rule suggests that Beijing is using those same bans to force an evolution. They are betting that if you trap enough smart engineers in a room with Version 1.0 tools and tell them they aren't allowed to leave until it works, eventually, you get a Version 10.0 that rivals the world's best.

It is a gamble on the idea that national sovereignty is more valuable than manufacturing efficiency. For the fab managers at SMIC, the immediate future involves a lot of trial-and-error with local machines they used to ignore. For the rest of the world, it is a reminder that when you deny someone access to your ecosystem, you eventually force them to build their own—and they might just start by making their most successful companies pay for the construction.

M0re on US-China Tech Rivalry:

• China’s AI Chip Deficit: Why Huawei Can’t Catch Nvidia and U.S. Export Controls Should Remain (CFR)
• If U.S.-China AI Rivalry Were Football, the Score Would Be 24-18 (WSJ)

On Our Radar

Our Intelligence Desk connects the dots across functions—from GTM to Operations—and delivers intelligence tailored for specific roles. Learn more about our bespoke streams.

Nvidia's Talent Grab

• The Headline: Nvidia is in advanced talks to acquire Israeli AI model-maker AI21 Labs for up to $3 billion, a deal primarily driven by the need to secure its elite research talent. (Reuters)
• ARPU's Take: Nvidia is paying a $1.5 billion premium over AI21's last valuation to bypass the global war for AI talent and instantly staff its new Israeli R&D center with a cohesive, world-class team.
• The Product Implication: The deal signals Nvidia's intent to move up the value chain from chip seller to model provider. By owning AI21's Jurrasic models, Nvidia can offer enterprise-grade, customizable LLMs as a service, directly competing with its own customers (OpenAI, Anthropic) in the B2B inference market. This creates channel conflict but also diversifies revenue, ensuring Nvidia captures value from the software layer even if hardware margins eventually compress.

The Singapore Shuffle

• The Headline: Meta is acquiring profitable AI agent startup Manus for $2 billion, while simultaneously executing a public de-coupling from the company's Chinese founders and investors to preempt US regulatory scrutiny. (TechCrunch)
• ARPU's Take: This is an "ARR-cquisition" designed to calm Wall Street. Zuckerberg is buying Manus $100M+ in recurring revenue—a tangible metric he can use to justify Meta's colossal $60B AI infrastructure spend. The noisy, public decoupling from China is the political price of admission for buying the one thing Meta's own AI division hasn't produced: a P&L.
• The Regulatory Implication: This deal establishes the "Singapore Airlift" as the official playbook for acquiring Chinese-founded AI startups. The strategy requires three non-negotiable steps to satisfy Washington: (1) Re-domicile the entity to a neutral third country like Singapore, (2) Purge all Chinese capital from the cap table, and (3) Publicly shutter all China-based operations. Meta is paying a premium not just for the asset, but for the geopolitical cleansing required to bring it onshore, creating a M&A blueprint for any company with a similar background.

P.S. Tracking these kinds of complex, cross-functional signals is what we do. If you have a specific intelligence challenge that goes beyond the headlines, get in touch to design your custom intelligence.

You received this message because you are subscribed to ARPU newsletter. If a friend forwarded you this message, sign up here to get it in your inbox.