Beijing did not invent any of these mechanisms. Texas Instruments had used joint ventures
in Taiwan in 1968. Samsung had bought a DRAM design from Micron in 1985. What was new in
the China case was the scale, the speed, and the political reading. The 2018 USTR
Section 301 report called the package forced technology transfer. The
day-to-day reality was that the cost was paid voluntarily, transaction by transaction,
by American firms that had concluded the China market was worth more than the marginal
leakage of any single deal.
The taxonomy below catalogues the five principal channels through which Chinese
semiconductor capability was built up between roughly 2010 and 2020. None of the
cases recorded here was a midnight visit by a man with a thumb drive. They were
conference rooms, term sheets, and offer letters — almost none of which moved
illegally. By the time the channels began to close, an enormous amount had
moved through them.
Channel intensity
- Capital-led
- Knowledge-led
- Hybrid / structured
CHANNEL I · HYBRID
I
Joint ventures
— design IP licensed to a Chinese-domiciled entity
Vehicle: 51/49 holding coCleared: case-by-case CFIUS
American firm faces antitrust pressure or market-access friction in China; agrees
to share design IP in exchange for partial ownership of a Chinese-domiciled venture
that sells into the protected mainland market. Each was structured to comply with
the letter of export controls in 2014–2016. Each became a target of those same
controls within three to five years.
AMD × THATIC
Tianjin holding entity. AMD licenses original Zen architecture for an x86 server chip. Hygon Dhyana ships, indistinguishable from Epyc but for crypto block.
$293M
+ royalties
ENTITY-LISTED 2019
IBM × Inspur
POWER8 cores + design assistance. China POWER Technology Alliance launches under OpenPOWER umbrella.
Aug 2014
OpenPOWER
DISCOURAGED
Qualcomm × Guizhou
HXT (Huaxintong). ARM-based Centriq server core; Guizhou supplies funds and political cover. StarDragon 4800 tapes out late 2018.
$570M
combined
UNWOUND 2019
CHANNEL II · HYBRID
II
Architecture licensing
— giving away an instruction set to keep it alive
Vehicle: standards consortiumCleared: BIS general license
IBM’s OpenPOWER (2013) and ARM’s Cortex licensing program created legitimate
channels by which Chinese firms could obtain the design language of a modern CPU.
From IBM’s vantage, the trade was rational: keep POWER alive in China by
licensing it. From Beijing’s vantage, it was a bargain — the planners had
never managed to build a competitive server CPU; OpenPOWER handed them the design.
OpenPOWER Foundation → Inspur
By 2014, a dozen Chinese firms cluster into the China POWER Technology Alliance.
2014
Consortium
UNDER REVIEW
ARM China × Hopu Investment
51/49 JV established 2018; Allen Wu refuses to leave when fired in 2020. Two-year governance crisis.
$775M
valuation
CONTESTED
CHANNEL III · CAPITAL
III
Outbound M&A
— equity acquisitions of foreign capability
Vehicle: state-backed funds via private wrappersCleared: 14 of 30+
Public, dramatic, politically combustible. Closed deals worth roughly $14 billion
between 2015 and 2018, against tens of billions more that were withdrawn, blocked,
or never formally announced. The Lattice block (Sept 2017) reframed the calculus
for every subsequent deal; FIRRMA (Aug 2018) closed the window for good.
Tsinghua Unigroup → Micron
$23B bid for Idaho DRAM maker. Largest Chinese acquisition of a US firm ever attempted.
$23.0B
Jul 2015
ABANDONED
Fujian Grand Chip → Aixtron
German MOCVD tool maker. Obama’s presidential block — first prohibition before close.
€670M
Dec 2016
PRES. BLOCK
Canyon Bridge → Lattice
Sole backer: China Reform. Trump’s first CFIUS prohibition. Founder Benjamin Chow later convicted on insider trading.
$1.3B
Sep 2017
PRES. BLOCK
Hua Capital + CITIC → OmniVision
Santa Clara CMOS image-sensor maker, taken private. One of the few that closed.
$1.9B
Apr 2016
CLOSED
CHANNEL IV · KNOWLEDGE
IV
Talent recruitment
— offer letters with 3–5× compensation
Vehicle: Thousand Talents Plan, est. 2008Cleared: not regulated
Quieter than M&A and, in the long run, more consequential. Chips are designed by
people. The most accessible reservoir, from Beijing’s point of view, was Taiwan:
same language, easy travel, deep family networks across the strait, and four
decades of accumulated process expertise. By the late 2010s, Taiwanese authorities
estimated Chinese firms had drawn nearly 60,000 professionals from
across the diaspora since 2008.
Charles Kau→ Tsinghua Unigroup
Godfather of Taiwan’s DRAM industry. Inotera EVP at 61, joined Beijing-backed Unigroup in late 2015.
2015
DRAM
RETURNED 2018
Liang Mong-song→ SMIC
Ex-TSMC, ex-Samsung. Joined SMIC as co-CEO Oct 2017. Brought several dozen TSMC engineers. SMIC at 14 nm by 2019, 7 nm by 2022.
2017
Foundry
RETAINED
Chiang Shang-yi→ Wuhan Hongxin / SMIC
Ex-TSMC CTO. Wuhan Hongxin collapsed mid-2020. Rejoined SMIC three days before Entity-listing.
2019
Process
DEPARTED 2021
CHANNEL V · KNOWLEDGE
V
Remote poaching
— shadow firms inside Taiwan itself
Vehicle: shell Taiwanese design housesCleared: not until 2021 raids
The most aggressive recruiters did not bother with relocation. They built shell
companies inside Taiwan that on paper looked independent. In practice they had
been set up by mainland firms — Bitmain prominent among them — to recruit engineers
away from TSMC, MediaTek and the local AI-chip houses without ever asking those
engineers to leave the island. Researchers at Taiwan’s DSET gave the practice
a name: the remote-poaching model.
Bitmain → WiseCore Technology
New Taipei shell firm, raided March 2021. Designed edge-AI silicon shipped to Beijing affiliates.
Mar 2021
Raid
19 DETAINED
Bitmain → IC Link
Sister shell. Investigation later extended to Sophon AI training chip and Beijing entity Jingshi.
Mar 2021
Raid
UNDER PROBE
~100 shell firms (est.)
DSET 2022 audit. By 2022, Taipei had banned mainland firms from using shells to recruit on the island; enforcement remained spotty.
2018-23
Pattern
REGULATED
CHANNEL VI · CAPITAL
VI
State subsidies
— the frame inside which the other four operated
Vehicle: Big Fund + ~60 provincial fundsCleared: WTO disputes pending
Around all of the above ran a grant economy of tax breaks, accelerated depreciation,
subsidised land, cheap utilities, and below-market loans from the Big Four state
banks. By the OECD’s later count, Chinese chipmakers received more
government support per dollar of revenue than firms in any other major
economy. None of the four channels above would have produced their results in the
absence of this fiscal envelope.
Big Fund Phase I+II+III
¥687B aggregate state equity (~$97B) across three rounds, 2014–2024.
$97B
State equity
DEPLOYED
Provincial guidance funds
~60 separate vehicles in Beijing, Shanghai, Hubei, Anhui, Jiangsu, Sichuan, Fujian, Guangdong.
~$150B
est. catalysed
ONGOING
The Apple parallel
28million workers trained
Patrick McGee’s 2025 history of Apple in China captured a parallel mechanism in
consumer electronics. Apple, McGee documented, spent something like
$275 billion upgrading its Chinese suppliers between 2016 and 2021 —
peaking at perhaps $55B a year. By Apple’s own count, the company had trained
more than 28 million Chinese workers since 2008, more than the entire labour force
of California.
McGee’s framing was that this had not been theft, or coercion, or even forced
technology transfer in the customary sense. It had been a transfer of process
knowledge in exchange for the lowest unit costs and fastest ramp times the
world had ever seen — conducted enthusiastically by an American company that did
not realise, until much later, what it was teaching the host country to do without
it. The chip industry’s version of that trade had a heavier overlay of
national security from the start. The underlying physics were similar.
The window opened in 2014. By 2020, it was shut.
Each of the channels above was structured to comply with the letter of the
relevant U.S. export-control regime in the year it was set up. Each
became, within three to five years, a target of the same regime once the
political climate shifted. The whole architecture of U.S. export controls had
been built for a Cold War in which the adversary’s economy was sealed off from
the world’s, not for one in which American firms could be paid to walk
capability across the border in plain daylight.
Inside Beijing, the planners drew the lesson that the channels could be closed
one by one. They began preparing for a different kind of catch-up — one that
would have to be done without American partners willing to sit in conference
rooms in Tianjin and sign their names. That preparation, and the
indigenisation push it triggered, was the next phase of the strategy.