AIBRIEF

AI Reaches National-Security Status: Cheng Shih-chia Urges Taiwan to Pursue Software-Hardware Integration

林紀旭 James LinEditor-in-Chief
Published · Updated
According to a report by money.udn.com, Cheng Shih-chia (程世嘉), a Stanford University computer science graduate with 20 years of AI research and practical experience, said advanced AI now shows tool-calling ability and goal-directed behavior. He cited the US government's 18-day takedown order on an advanced AI model this year as evidence that AI has risen to national-security and military status.

Who is Cheng Shih-chia, and what is his background for commenting on AI's national-security status?

According to a report by money.udn.com, Cheng Shih-chia (程世嘉) holds a computer science master's degree from Stanford University and has 20 years of AI research and practical experience. CNA (cna.com.tw) reported the identical credential, describing him as having "20 years of AI research and practical experience" with a Stanford CS master's degree. Both outlets frame this background as the basis for his remarks on AI's shifting status.

Why has AI been elevated to national-security status, and how have governments responded?

Cheng told reporters that "advanced AI already possesses the ability to call tools and exhibits 'goal-directed' traits," according to money.udn.com. As evidence that AI has moved beyond a purely commercial technology, he pointed to a concrete government action: the United States government this year required AI vendors to take an advanced AI model offline for 18 days, a move both money.udn.com and CNA reported as showing "AI has risen to the level of national security and military affairs."

CNA's report adds a timeframe and a further detail: the 18-day takedown occurred "in June and July of this year," and it notes that "software itself has also begun facing export controls." Following the US action, CNA reported that Chinese authorities "convened founders of open-source large models for a meeting to discuss countermeasures" — indicating that the US takedown order prompted a parallel policy response from Chinese authorities.

In light of both governments' moves to control access to AI models, Cheng advised, according to money.udn.com, that companies "should download and back up open-source models as soon as they see them," explaining: "because you never know which day China or the US will block it." This advice directly follows from the 18-day US takedown and the Chinese government's countermeasure meeting cited above.

What does Taiwan's semiconductor employment picture reveal about the case for software-hardware integration?

Both money.udn.com and CNA reported the same figures on Taiwan's semiconductor sector: Taiwan's stock market capitalization has climbed into the world's top 5, yet direct semiconductor employment stands at only about 300,000 people — just 3% of Taiwan's total workforce of more than 10 million. The two outlets presented this gap between market weight and workforce size as the backdrop for calls to expand into software-hardware integration.

How will AI reshape corporate productivity and talent structures?

According to money.udn.com, Cheng introduced the concept of "intelligence inflation," arguing that when top-tier AI reaches an IQ of 130 — "professor and genius level" — the value of knowledge and basic output will depreciate rapidly. He further stated that "super users" who know how to organize teams of AI agents can see individual productivity rise by 20 times or more.

What common stance do Chien Li-feng and Cheng Shih-chia share on Taiwan's independent innovation?

CNA reported that Chien Li-feng (簡立峰) said the most irreplaceable human qualities are "thinking power" and "slow thinking" — a view that stands opposite to Cheng's account of AI reaching professor-level IQ. Chien also said, per CNA, "don't just follow the world's agenda — follow the agenda you set for yourself," adding that maintaining subjectivity is what allows an irreplaceable position to be established amid the AI wave.

Key figures at a glance

MetricValueReported by
US advanced AI model takedown duration (2026)18 daysmoney.udn.com; CNA
Cheng Shih-chia's AI research/practical experience20 years (Stanford CS master's)money.udn.com; CNA
Taiwan semiconductor direct employment~300,000 workersmoney.udn.com; CNA
Semiconductor share of Taiwan's total workforce (10M+)3%money.udn.com; CNA
Taiwan stock market capitalization rankingTop 5 globallymoney.udn.com; CNA
Top AI model's estimated "IQ," per Cheng130 (professor/genius level)money.udn.com
Productivity gain for AI-agent "super users," per Cheng20x or moremoney.udn.com

What this means

The facts reported by money.udn.com and CNA outline a chain of cause and response: the US government's 18-day takedown of an advanced AI model this year, followed by China convening open-source model founders to discuss countermeasures, is what Cheng points to as proof AI has reached national-security status — and it is this same sequence that underlies his advice for companies to back up open-source models before they are blocked. Separately, the employment data show a structural tension in Taiwan's economy: a top-5 global market capitalization resting on a semiconductor workforce that is only 3% of total employment. And on the question of what AI displaces, Cheng's claim of a 130-IQ, professor-level AI sits directly against Chien's claim that thinking power and slow thinking remain irreplaceable — two views reported side by side in the same CNA account without either outlet reconciling them.

📊 Evidence

📎 Sources

  1. money.udn.com
  2. cna.com.tw
林紀旭 James LinEditor-in-Chief

Related

BRIEF

Indonesia Launches First University AI Center: UGM, Indosat and NVIDIA Open NVAITC in Yogyakarta

According to NVIDIA's blog, Indonesia's Ministry of Communication and Digital Affairs, Indosat Ooredoo Hutchison, NVIDIA and Universitas Gadjah Mada (UGM) launched the UGM Indosat NVIDIA AI Technology Center (NVAITC) in Yogyakarta this week — the country's first university-based AI technology center. Powered by NVIDIA's full-stack AI platform and Indosat's GPU Merdeka platform, NVAITC already backs projects tied to Indonesia's more than 1 million annual tuberculosis cases and a farm sector employing nearly 30% of the workforce.

EffectStory 編輯部 ·
BRIEF

Not TSMC: NVIDIA's Second-Largest Equity Stake Is Musk's SpaceX

According to a Liberty Times (LTN) report citing NVIDIA's SEC filing, NVIDIA held about 128.8 million Class A shares of SpaceX as of the end of the second quarter, making it SpaceX's sixth-largest investor. CNYES reported the stake was valued at roughly $21 billion at quarter-end — a position trailing only NVIDIA's Intel holding among its disclosed equity stakes.

EffectStory 編輯部 ·
BRIEF

UBS: Micron's Standard DRAM Margins Set to Overtake HBM, Nearing 90%

According to a UBS report cited by TechNews and Cnyes, Micron's (美光) standard DRAM gross margin is projected to climb from roughly 44%-50% in 2025 to 89%-91% by mid-2026 and 92% by year-end, overtaking HBM margins of 56%-70% over the same period. UBS attributes the reversal to HBM's heavier wafer and packaging demands squeezing standard-memory supply, plus simultaneous AI-server demand for both HBM and DDR5.

EffectStory 編輯部 ·