# The Stack China Built: Manufacturing Scale, Stolen Time, and the Compute Double Game

**Summarized by Grok 4.5**

**A long-form synthesis of U.S. Treasury and Justice Department actions, FBI wanted notices, New York Times reporting and opinion, Tom’s Hardware on China’s AI grid, Konstantin Pilz’s DeepSeek transcript summary, JD Power / ACEA EV data, and Garry’s List on San Francisco political money—read as one industrial strategy rather than a pile of scandals.**

*Long-form for blog publication · Sources as cited. Charges, wanted notices, and indictments are allegations unless a conviction is stated; mentions of companies, nonprofits, or individuals do not by themselves establish illegal conduct. Where sources conflict or rest on anonymous officials, that is noted.*

---

## At a glance

Twelve public sources—federal press releases, court-adjacent DOJ statements, a major-newspaper history of rare earths, an opinion on scientific visas, a hardware-industry report on a national AI data-center plan, an EU auto-registration snapshot, an X-thread summary of a DeepSeek CEO call, and a San Francisco political investigation—do not describe “China crime” as a personality trait. They describe **a state-enabled industrial system that treats American openness as an input**.

| Layer | What the sources show | Illustrative document |
|------|------------------------|------------------------|
| **Chokepoints** | Multi-decade rare-earth and materials strategy from the top of the party-state | *NYT*, rare-earth origin story (July 2025) |
| **Scale manufacturing** | Export-ready EVs growing fast even as EU new-car sales fall | JD Power / ACEA on BYD (2026) |
| **Talent & labs** | Visa pipelines, visiting scholars, dual-use research risk; also the cautionary Qian Xuesen history | DOJ Michigan; *NYT* opinion on visas |
| **Economic espionage** | MSS officer sentenced for aviation trade-secret campaign | DOJ (Yanjun Xu, 20 years) |
| **Cyber as industry** | Treasury “major” APT breach; data brokers sanctioned and charged | *NYT* Treasury hack; Treasury OFAC sb0042 |
| **Chemical supply** | China-based firms and nationals charged/wanted for fentanyl precursors and cutting agents | FBI CEI; USAO SDOH |
| **Compute + narrative** | ¥2 trillion AI grid with 80% domestic chips—while CEOs say NVIDIA is scarce | Tom’s Hardware; Pilz / DeepSeek |
| **Local politics (US)** | Closed loops of public money and power in one SF district—relevant as a *vulnerability type*, not as a China attribution | Garry’s List / YCD |

The through-line this piece proves:

**China’s manufacturing and corporate-espionage posture is not a string of rogue entrepreneurs. It is a stack: lock materials and factories, compress R&D timelines by taking what others invent, map and pressure U.S. institutions digitally, profit from chemical and financial gray markets where enforcement is asymmetric—and, on AI, talk down the U.S. compute moat in public while racing to build sovereign compute in private.**

That is why the same flag keeps appearing in U.S. charging documents. It is not because “Chinese people commit more crime.” It is because a large, capable party-state has made **technology acquisition, industrial self-sufficiency, and strategic competition with the United States** core missions—and because U.S. openness (universities, capital markets, software supply chains, chemical commerce) still offers high-yield points of entry.

![Six layers of industrial catch-up strategy](../images/china-strategy-layers.svg)

*Figure 1. Reading the sources as a stack, not a crime blotter.*

---

## 1. Why these sources belong on one page

Most readers encounter these stories in separate feeds: a spy trial in Ohio, a Treasury breach over the holidays, a fentanyl poster on the FBI site, a BYD sales chart, a Bloomberg-sourced data-center plan, a San Francisco ethics saga. The silo is the problem. **Industrial strategy is multi-domain.** Deng Xiaoping-era materials policy and a 2026 AI-grid blueprint are the same habit of mind applied to different bottlenecks. An MSS officer soliciting GE Aviation files and a contractor data broker selling access to U.S. networks are different tools for the same goal: **shorten the path from foreign know-how to domestic capability**.

The sources also force three distinctions that internet argument usually flattens:

1. **State direction versus individual crime.** Yanjun Xu was, by the U.S. government’s proven case, a Ministry of State Security officer. A Guangzhou chemical salesman charged with fentanyl conspiracies is not automatically an intelligence officer. Both can be real; the mechanisms differ.
2. **Ethnicity versus jurisdiction.** The Justice Department charges *nationals*, *companies*, and *state actors* under U.S. law. Treating every Chinese student as a spy is both unjust and strategically stupid—as the Qian Xuesen story argues—yet pretending university labs have no dual-use risk is also a choice with costs.
3. **Narrative versus balance sheet.** Claims that “frontier models will be commoditized” and that export controls “already failed” are political economy, not pure technical facts. The DeepSeek call transcript summarized by Konstantin Pilz and the $295 billion domestic AI buildout are useful correctives: **the people building Chinese AI still treat high-end compute as scarce and precious**.

Read together, the dossier is less “why is China evil?” and more **“how does a peer competitor industrialize under and around a rules-based order that assumes goodwill?”**

---

## 2. Shared baseline: what these documents jointly take as given

Across agencies and newsrooms, several premises recur without serious dispute in this corpus:

- **The PRC treats critical technology as a national project**, not only a market outcome. Rare earths, aerospace composites, semiconductors, and AI infrastructure appear as state priorities with multi-decade horizons.
- **U.S. institutions remain attractive targets** because they hold advanced IP, financial intelligence, and open research culture.
- **Cyber operations attributed to China-linked APTs** have hit U.S. government and critical-infrastructure-adjacent systems, including the Treasury Department in late 2024.
- **Synthetic-opioid supply chains** continue to involve China-based chemical companies and individuals, per FBI wanted notices and Southern District of Ohio indictments.
- **China is simultaneously constrained and ambitious in AI compute**: export controls bite (per Chinese executives’ own words in the Pilz summary), and Beijing answers with sovereign capital and domestic-chip mandates (Tom’s Hardware / Bloomberg reporting).
- **Manufacturing scale is a strategic weapon of its own.** BYD’s European growth while the wider EU market softens is not a press-release curiosity; it is capacity, price, and vertical integration meeting a tariff-war world.

None of that requires believing every indictment will produce a conviction, or that every Chinese firm is an MSS cutout. It requires taking seriously that **pattern + state capacity + economic size** produces a volume of U.S. cases that smaller peer competitors simply cannot match.

![Timeline of selected cases and milestones](../images/china-crime-timeline.svg)

*Figure 2. Selected milestones from the source set—from Qian Xuesen to the 2026 AI grid.*

---

## 3. Materials first: rare earths as the original long game

The *New York Times* rare-earth history (July 5, 2025) opens with a fact Western publics only recently internalized: when China temporarily suspended most rare-earth exports, leaders elsewhere scrambled—**but for almost half a century the metals had already been a top-of-government Chinese priority**.

The piece’s origin story is metallurgical, not mystical. Mao-era iron and steel emphasis produced volume without quality. British and American metallurgists had shown that a dash of **cerium**—a relatively separable rare earth—could improve ductile iron used in pipelines and car parts. When Deng Xiaoping consolidated power in 1978, he elevated technocrat **Fang Yi** over science and technology. Fang moved quickly to extract rare earths from ore at **Bayan Obo**, among other steps that turned a materials afterthought into a strategic industry.

What matters for this synthesis is not the full chemistry of the lanthanides. It is the **time horizon**. While Western firms treated rare earths as a messy, low-margin, environmentally ugly business, Chinese policy treated them as **leverage in waiting**—the kind of leverage that later shows up in magnets for EVs, wind turbines, defense systems, and electronics. Temporary export suspensions work as signals precisely because refining and separation capacity cannot be rebuilt in a budget cycle.

**Surmise:** Manufacturing strategy in the Chinese state tradition still starts with **physical bottlenecks**. Espionage and cyber are accelerants; materials and factories are the chassis.

---

## 4. Scale as strategy: BYD and the EV export wave

JD Power’s Autovista24 write-up (updated through mid-2026) reports that while EU new-car registrations softened in January 2026, **BYD deliveries surged 175.3% year on year to 13,982 units**, lifting EU market share to about **1.7%** (up 1.1 percentage points). The brand outsold Mini, Mazda, Honda, Lancia, and Alpine in that snapshot and aimed to roughly double European points of sale toward **2,000**, with ambitious German partner targets.

This is not “espionage.” It is **industrial policy plus vertical integration plus price**. Chinese EV makers spent a decade overbuilding capacity for a domestic market that rewards scale; exports then pressure foreign OEMs whose cost structures assume higher margins and slower learning curves. Europe’s political answer—tariffs, minimum import prices, local production bargains—acknowledges the point: **open markets without industrial strategy become dumping grounds**.

Pair this with rare earths and the picture sharpens. Control upstream materials, dominate midstream refining, flood downstream product markets, and you do not always need to steal the last algorithm. You can **price competitors out of the learning curve**. Corporate espionage, when it occurs, is how you skip steps inside that machine—not a substitute for the machine.

---

## 5. Economic espionage with a badge: Yanjun Xu

If rare earths are patient capital, the **Yanjun Xu** case is impatient intelligence work.

According to the Department of Justice (sentence announced November 16, 2022; later affirmed on appeal), Xu was the first Chinese government intelligence officer extradited to the United States to stand trial. A federal jury in Cincinnati convicted him of conspiracy to commit economic espionage, conspiracy to commit trade-secret theft, attempted economic espionage, and attempted trade-secret theft. He was sentenced to **twenty years**.

The government’s account, grounded in trial evidence, describes a career Ministry of State Security officer who, from at least 2013, targeted leading aviation companies. Techniques included:

- Aliases, front companies, and university invitations  
- Paying travel and stipends for “presentations”  
- Coordinating with others to access computers while guests were at dinner  
- Soliciting GE Aviation information related to an exclusive **composite aircraft engine fan module**  
- Running parallel efforts involving a French aerospace firm and malware  
- Handling a U.S.-based asset (Ji Chaoqun) tasked with biographical collection and recruitment

FBI Director Christopher Wray’s statement at sentencing framed the case bluntly: the Chinese government tasked an officer of its spy service to steal U.S. trade secrets so China could advance commercial and military aviation “at the expense of an American company.”

This is the pure form of **corporate espionage as state policy**. It is not a graduate student oversharing a paper. It is a professional intelligence service running a playbook against private industry. GE’s estimated value of the targeted IP, reported in secondary legal analysis of the appeal, ran on the order of tens to low hundreds of millions of dollars—significant, but the strategic value is larger: **years of composite-engine learning without years of composite-engine failure**.

---

## 6. Labs, visas, and the dual caution of Qian Xuesen

On November 5, 2025, the Justice Department charged three PRC research scholars—**Xu Bai, Fengfan Zhang, and Zhiyong Zhang**—associated with a University of Michigan laboratory, in a case involving conspiracy to smuggle biological materials into the United States and false statements to CBP. The complaint describes multiple shipments of concealed materials related to roundworms sent from the PRC; a fourth person, Chengxuan Han, had already pleaded no contest to smuggling and false-statement counts, been sentenced, and been removed. After Han’s case, the university investigated; the three defendants allegedly refused to participate, were terminated, rebooked flights, and were intercepted attempting departure via JFK. **A complaint is not a conviction**—the release says so explicitly—but the case sits in a DOJ narrative of repeated exploitation of research pathways.

That same spring, Secretary of State Marco Rubio announced the administration would work to **“aggressively revoke” visas** of Chinese students, including those with CCP ties or studying in “critical fields,” among roughly a quarter-million Chinese students in a million-strong foreign-student population.

Into that climate, the *New York Times* opinion essay of May 30, 2025, on **Qian Xuesen** functions as a deliberate counterweight. Qian, a Caltech/MIT rocket scientist who helped build early U.S. guided missiles and worked at the edge of the American aerospace establishment, was deported in 1955 amid McCarthy-era suspicion—despite, the essay notes, prosecutors clearing him of sedition and espionage charges—and was traded in a prisoner swap. He returned to China, persuaded Mao to fund a modern weapons program, and became the father of China’s missile and space efforts. Former Navy Secretary Dan Kimball’s line, quoted in the piece: deporting Qian was “the stupidest thing this country ever did.”

The essay’s policy moral is clear: **xenophobic dragnets can gift strategic capability to a rival**. The Michigan and Xu cases’ policy moral is equally clear: **openness without enforcement is also a gift**. A serious country holds both truths. The industrial strategy on the Chinese side does not need every student to be an agent; it needs **enough pathways**—scholarships, joint labs, talent programs, talent-plan recruitment—to keep the knowledge transfer flowing while denial remains incomplete.

---

## 7. Cyber: from Treasury workstations to a data-broker market

### 7.1 The Treasury breach

On December 30, 2024, U.S. officials disclosed that a China state-sponsored advanced persistent threat had compromised Treasury systems via a third-party remote-support provider, **BeyondTrust**. A stolen security key allowed remote access to certain workstations and unclassified documents. Treasury classified the intrusion as a **“major cybersecurity incident”** under its APT policy. *The New York Times* and other outlets reported that officials assessed it as espionage—aimed at Treasury’s unique view of financial systems, sanctions, and the Chinese economy—rather than the destructive pre-positioning seen in some utility-grid campaigns. Earlier the same year, Chinese intelligence had reportedly cracked email accounts used by Commerce Secretary Gina Raimondo amid semiconductor export-control decisions.

### 7.2 Sanctions and the contractor ecosystem

On March 5, 2025, Treasury’s OFAC designated **Zhou Shuai**, a Shanghai-based hacker and data broker, and his company **Shanghai Heiying Information Technology Company, Limited**. OFAC said Zhou, collaborating with previously sanctioned **Yin Kecheng**, illegally acquired, brokered, and sold data from sensitive U.S. networks—technology firms, a defense industrial base contractor, a communications provider, an academic health system, and a county government among the victims. Intelligence requirements in 2020 appeared to track CCP priorities: telecom data, border crossings, religious researchers, media personnel, public servants. DOJ unsealed related charging documents; State offered up to **$2 million** each for information leading to arrests of Zhou or Yin. Yin had already been tied to the Treasury compromise in earlier January 2025 sanctions.

The structure matters as much as the names. This is not only “PLA Unit X hacks Y.” It is a **gray-market cyber industry**—contractors, brokers, freelancers—nested against intelligence services, selling access and datasets. That model scales. It also blurs lines between crime and statecraft until a designation or indictment draws them.

**Surmise:** For a competitor that wants financial intelligence, sanctions foresight, and corporate data, cyber is cheaper than a carrier battle group and faster than a five-year plan. Manufacturing strategy without cyber is incomplete; cyber without manufacturing ambition is just theft.

---

## 8. Fentanyl precursors: industrial chemistry meets asymmetric enforcement

The FBI’s “China-Based Drug Conspiracy 1” wanted notice names **Zhanpeng Huang, Xiaojun Huang, Meixiang Yao, and Yuqing Feng**, affiliated with Guangzhou Tengyue Chemical Company, LTD., for alleged conspiracy to distribute controlled substances and international money laundering. The allegation: businesses posing as online pharmacies or chemical companies knowingly exported controlled substances and **cutting agents** (“cut”) to U.S. traffickers to boost fentanyl yield and potency, roughly January 2022 through August 2025, with Southern District of Ohio warrants issued September 2, 2025.

A related Southern District of Ohio grand-jury action (March 2026) charged additional Chinese nationals and pharmaceutical companies—including Shandong-linked entities in secondary reporting—with narcotics and money-laundering conspiracies involving fentanyl precursors and cutting agents such as **medetomidine**, and, in the later indictment, material-support theories after certain cartels were designated foreign terrorist organizations. Prosecutors describe open marketing to traffickers, crypto payment rails, and shipments toward Mexico and the United States, including southern Ohio.

Here the “why so many U.S. cases” question has a cold answer: **the United States is the demand sink for synthetic opioids, Chinese chemical industry is vast and globally networked, and extradition and regulatory symmetry are weak**. That combination produces American charging documents even when the principal defendants never set foot in Dayton. It is not evidence that the CCP runs every precursor mill as a ministry program. It *is* evidence that **industrial capacity + weak dual-use chemical controls + profit** can devastate a foreign society while the producing jurisdiction treats it as someone else’s enforcement problem—until diplomacy and secondary sanctions raise the cost.

For this article’s industrial-strategy frame, fentanyl is the dark mirror of rare earths and EVs: **chemistry at scale, export logistics, online storefronts, and payment rails**—the same globalization toolkit, optimized for harm.

---

## 9. Soft spots at home: San Francisco as a vulnerability type

Garry’s List’s investigation of **Young Community Developers (YCD)** and District 10 politics is not a China story. It is a story about a nonprofit that became a node in San Francisco’s “community benefits” pay-to-play ecosystem: former director Dwayne Jones facing extensive bribery-related charges in a separate kickback scheme; successor Shamann Walton moving from YCD leadership to supervisor while the district concentrated community-benefits leverage; current CEO DJ Brookter seeking the same seat with Walton’s endorsement while the organization draws the bulk of its roughly **$17 million** revenue from government sources and Brookter has sat on police and sheriff oversight bodies.

Why include it? Because great-power competition is not only about what Beijing does. It is about **whether open societies can police their own conflicts of interest, zoning fights, and public-money loops**. Prior Vibe Code Racing synthesis on data centers and influence campaigns (Garry’s List / Bitcoin Policy Institute material) already argued that authentic local environmental concern can run *parallel* to foreign-aligned messaging that slows U.S. compute. The YCD piece is a reminder of a different failure mode: **capture and rotation of local power** that has nothing to do with MSS yet everything to do with whether American governance can execute industrial policy at city scale.

An adversary does not need to invent American dysfunction. It only needs to outlast it.

---

## 10. The AI double game: “models commoditize” while the grid gets built

### 10.1 What DeepSeek’s CEO said when the mic was less careful

Konstantin Pilz’s July 2026 thread summarizes a leaked DeepSeek CEO **Liang Wenfeng** call (transcript-based; original audio not public at the time of the thread). The quotes, if accurate, are devastating to the “export controls failed / compute doesn’t matter” line:

- DeepSeek’s gap with the U.S. is “really… resources”—card count still “fairly small,” about **20,000 H-equivalent**, much of it newly arrived.  
- Leading U.S. labs, Pilz notes for scale, work with **orders of magnitude more** H100-class capacity.  
- Strategy: buy **NVIDIA** “in large batches,” “the more the better,” willing to pay a premium; “we can buy some **non-compliant** cards.”  
- Huawei 950s: useful for ecosystem and inference-scale work; roughly **four Huawei cards ≈ one NVIDIA**, with multi-year lag; not enough alone to train next-generation models at frontier scale.  
- Catch-up narrative: “two years behind… on one-twentieth of the compute”—and even ¥50 billion cannot magic a 50,000-GB300 training run into existence if the cards and power are not there.

This is the private balance-sheet view: **compute is the bottleneck**. Efficiency research is real; it does not abolish physics or foundry capacity.

### 10.2 What Beijing is building anyway

Tom’s Hardware, citing Bloomberg sources, reports China drafting a plan to spend roughly **2 trillion yuan (~$295 billion)** over five years on a nationwide AI data-center web, with **at least 80% domestic technology including AI chips**, carriers such as China Mobile and China Telecom operating most sites, and a single computing grid targeted around **2028**. Folding power-grid upgrades could push totals far higher. Domestic foundry reality bites: SMIC’s advanced nodes run hot utilization; HBM supply constrains Ascend-class builds; domestic suppliers may cover only around three-quarters of Chinese AI chip demand by 2030 even as the market swells. Beijing has already tightened rules: earlier floors of 50% local chips for data centers, then bars on foreign accelerators in state-funded projects.

So: **executives prefer NVIDIA; the state mandates Huawei-class silicon; the budget funds racks and power either way.**

### 10.3 Why push commoditization of frontier models?

Here is the surmise the user asked for, grounded in the above:

**If American elites believe frontier models are about to become cheap commodities, they underinvest in the scarce layer—energy, advanced packaging, fabs, and trained clusters—while over-indexing on open-source vibes and short-term inference cost.** That belief is *convenient* for a compute-constrained competitor. It encourages:

1. **Relaxation of export-control politics** (“they already have parity”).  
2. **Slower U.S. permitting and buildout** for power and data centers.  
3. **Technology transfer and open-weight romanticism** that transfers methods without transferring the West’s energy and chip stack.  
4. **Market panic cycles** that punish U.S. AI capex as “bubble” while China socializes AI infrastructure through state carriers and bonds.

Meanwhile China invests heavily in data centers because **leaders do not actually believe compute is irrelevant**. They believe the opposite: that whoever owns abundant, controllable, domestically supplied compute will own the next industrial stack—from model training to robotics to military decision systems. The public “commoditization” line and the private rack-buying panic are not a contradiction. They are **information operations and industrial policy aimed at different audiences**.

![Narrative versus hardware reality](../images/compute-narrative-gap.svg)

*Figure 3. Exportable story on the left; revealed constraints and state response on the right.*

---

## 11. Synthesis: what the manufacturing + espionage stack reveals

| Question | Answer from the corpus |
|----------|-------------------------|
| What is the manufacturing strategy? | Secure materials (rare earths), dominate scale industries (EVs/batteries), force foreign markets to absorb surplus capacity, and localize high-value inputs over decades. |
| What is the espionage strategy? | Professional services (MSS) steal time-compressing IP (aviation composites); cyber brokers and APTs steal data and access; academic pathways can be abused for materials and know-how. |
| Why so many U.S. crime links? | Size + state capacity + U.S. target richness + weak extradition + domains where U.S. law reaches overseas supply (drugs, cyber, export controls, trade secrets). Volume of enforcement is not a census of “national character.” |
| Why commoditize-models talk *and* data-center mega-spend? | Talk aims at U.S. political economy and open research culture; spend aims at the actual scarce resource Chinese builders admit they lack. |

The stack reveals a coherent theory of power under technology competition:

1. **Time is the scarce good.** Espionage buys years. Scale manufacturing buys learning curves. Cyber buys foresight.  
2. **Openness is an attack surface.** Universities, SaaS admin tools, chemical e-commerce, and capital markets are all load-bearing for the United States—and therefore doors.  
3. **Narrative is a production input.** If the U.S. tells itself that AI is “just software,” it will underbuild the industrial substrate China is financing with sovereign debt.  
4. **Asymmetric enforcement is a feature.** Precursor chemistry and cyber contractors can harm the United States while remaining hard to punish inside China.  
5. **Self-sufficiency is the end state.** Domestic chips, domestic rare-earth magnets, domestic EVs, domestic model training—not because autarky is efficient, but because **sanctions risk is priced as existential**.

---

## 12. Why “linked to so many crimes in the USA”—a careful answer

A responsible synthesis rejects two cartoons:

- **Cartoon A:** “The Chinese are a criminal people.” This is bigotry dressed as analysis. Hundreds of thousands of Chinese students, workers, and citizens are not agents; many are victims of CCP coercion themselves.  
- **Cartoon B:** “It’s all racism and McCarthyism.” This cannot survive contact with Xu’s conviction, the Treasury APT attribution, OFAC’s broker designations, or the fentanyl precursor dockets.

The non-cartoon answer has five parts:

1. **Base rates and size.** A country of 1.4 billion with the world’s largest manufacturing base and a globalized diaspora will appear often in *any* large country’s enforcement statistics.  
2. **State priority.** When economic espionage and technology transfer are official priorities, the residual criminal cases are not random noise—they are the visible edge of a policy.  
3. **Target selection.** The United States still leads in many frontier technologies, hosts the deepest capital markets, and runs the dollar system Treasury oversees. That is why Treasury and GE Aviation matter more than the equivalent offices in a mid-sized economy.  
4. **Enforcement locus.** U.S. attorneys charge what U.S. law can reach. Chinese precursors killing Ohioans show up in Dayton indictments; American demand is part of the causal chain, which does not erase supplier intent where proven.  
5. **Measurement bias.** Democratic openness means we *see* U.S. indictments. We do not see a symmetric Chinese docket against American nationals for comparable state-directed IP theft—because the systems are not symmetric.

So: **linked often because the competition is real, the U.S. is the prize market and prize lab, and American process generates public paper.** That is diagnosis, not demonology.

---

## 13. Takeaways

1. **Treat the stack as a stack.** Rare earths, EVs, lab access, MSS trade-secret ops, cyber brokers, and AI grids are one industrial theory expressed through different ministries and gray markets.  
2. **Defend openness by enforcing it.** Visa scrutiny for dual-use risk and protection for legitimate scholarship are complements, not opposites—the Qian mistake and the Xu playbook are both real.  
3. **Close the SaaS and contractor attack surface.** The Treasury–BeyondTrust pattern is a governance failure as much as a China success.  
4. **Do not outsource chemical dual-use control to hope.** Precursor and cutting-agent cases will continue until cost and extradition change supplier calculus.  
5. **Ignore AI narrative warfare that contradicts Chinese builders’ own words.** When CEOs say they will pay premiums for non-compliant NVIDIA cards while Beijing budgets hundreds of billions for domestic racks, believe the racks.  
6. **Build U.S. power, fabs, and permitting capacity.** The competitor is not waiting for your zoning board to find consensus.  
7. **Fix local capture for its own sake.** San Francisco-style pay-to-play is a self-own; strategic competition only raises the stakes of ordinary corruption.

---

## 14. Sources & further reading

1. U.S. Department of the Treasury, “Treasury Sanctions China-based Hacker Involved in the Compromise of Sensitive U.S. Victim Networks,” press release sb0042 (March 5, 2025).  
2. *The New York Times* Opinion, “The U.S. Deported This Chinese Scientist, in a Decision That Changed World History” / Qian Xuesen (May 30, 2025).  
3. Konstantin Pilz (@KonstantinPilz), X thread on DeepSeek CEO Liang Wenfeng leaked call (July 23, 2026), post ID 2080353024816411089.  
4. *The New York Times*, “China Hacked Treasury Dept. in ‘Major’ Breach, U.S. Says” (December 30, 2024); corroborating Reuters/BBC coverage of BeyondTrust key compromise.  
5. *The New York Times*, “China’s Rare Earth Origin Story, Explained” (July 5, 2025).  
6. FBI Wanted, “China-Based Drug Conspiracy 1” (CEI), subjects Huang, Huang, Yao, Feng / Guangzhou Tengyue Chemical.  
7. U.S. Attorney’s Office, Southern District of Ohio, grand-jury charges against additional Chinese nationals and pharmaceutical companies in drug trafficking and money-laundering conspiracies (March 2026); related September 2025 multi-defendant precursor case.  
8. U.S. Department of Justice, “Chinese Government Intelligence Officer Sentenced to 20 Years… Attempting to Steal Trade Secrets From Cincinnati Company” (Yanjun Xu, November 16, 2022).  
9. U.S. Department of Justice, “Three Chinese National Scholars from University of Michigan Laboratory Charged for Conspiring to Smuggle Biological Materials into the U.S.” (November 5, 2025).  
10. Garry’s List, “The Nonprofit Behind San Francisco’s Pay-to-Play Scandal Wants Another Supervisor Seat” (YCD / District 10).  
11. Tom’s Hardware, “China drafts $295 billion plan to build a national AI data center grid running on 80 percent domestic chips” (Bloomberg-sourced, 2026).  
12. JD Power / Autovista24, “BYD enjoys growth amid declining new-car sales in the EU” (January 2026 ACEA data; article dated/updated 2026).

Secondary context used sparingly for verification: Reuters and BBC on the Treasury breach; TRM Labs summary of the March 2026 Ohio FTO-related precursor indictment; NPR/WSJ secondary on Xu and BYD where primary pages were paywalled.

---

## Appendix: numbers cheat sheet

| Metric | Figure | Source path |
|--------|--------|-------------|
| AI grid plan | ~¥2T / ~$295B over ~5 years | Tom’s Hardware / Bloomberg |
| Domestic chip target for that grid | ≥80% | Same |
| DeepSeek H-equivalent (claimed) | ~20,000 | Pilz / Liang transcript summary |
| Implied U.S. lab scale (context) | Orders of magnitude more H100e | Pilz commentary |
| Huawei vs NVIDIA (Liang) | ~4 cards ≈ 1; multi-year lag | Pilz |
| Xu sentence | 20 years | DOJ 2022 |
| Treasury breach notice | Dec 2024; major APT incident | *NYT* / Treasury letter reporting |
| OFAC Zhou / Shanghai Heiying | March 5, 2025 | Treasury sb0042 |
| State rewards (Yin / Zhou) | Up to $2M each | Treasury / State |
| BYD EU Jan 2026 | +175.3% YoY; 13,982 units; ~1.7% share | JD Power / ACEA |
| YCD approx. annual revenue | ~$17.1M, mostly government | Garry’s List / ProPublica Nonprofit |
| Chinese students in U.S. (approx.) | >250,000 of ~1M foreign students | *NYT* opinion context |
| GE IP loss range (secondary) | ~$65–150M estimated | Legal commentary on Xu appeal |

---

*End of article body.*