13 min readJust now

–

Stanford EE professor, TSMC vet, and Hanover venture partner Dr. Ali Keshavarzi and Hanover founding partner Joe Malchow created EE 292P and are teaching it at Stanford winter quarter 2026. Called “Atoms, Bits, and the National Interest,” or ABNI, we are attempting to trace the build-up of national power resulting from the growth of the semiconductor industry — from transistor physics through to great power competition.

We’re bringing in leaders from key companies in tooling, foundry, design, power electronics, AI applications, and other fields including economics and law.

Why do this? One, I (Joe) want to address industrial-economic knowledge gaps I’ve seen in the both the mature semiconductor and startup worlds. I want to move American EEs and CS away from “escape competition” and over to “win the competition.” And two, I want to synthesize this multi-decade history to make better investing decisions and policy recommendations. I got my start in Silicon Valley in semis (mixed-signal microcontrollers), but after Stanford Law spent virtually all of my time investing in new-build companies. This is my attempt to grow the body of knowledge of how semis work to create and project national power.

Press enter or click to view image in full size

Week 1 — Technology and Environment: Overview, Problems, Challenges, and Opportunities

Lecturer: Ali Keshavarzi (Adjunct Professor, Stanford EE) Guest Speaker: Steve Blank (Stanford MS&E)

Tuesday, January 6, 2026 Lane 200–205 Stanford, Calif.

Resources:

EE 292P Week 1 — Deck [PDF] Steve Blank — China/US Tech Competition Deck [PDF]

Course Overview: Atoms, Bits, and National Interest

This course examines the intersection of semiconductor technology, computing, and national security through three lenses:

1. Atoms — Semiconductor technology and the transistor industry
2. Bits — Computing systems, accelerators, GPUs, and CPUs
3. National Interest — Impact on economy, society, and national security

Press enter or click to view image in full size

Course Philosophy

Technology doesn’t exist in a vacuum. As technologists working “in the basement,” we need to understand how policymakers, economists, legal frameworks, and geopolitics influence and are influenced by our work. The course bridges two worlds:

• First half: Technology fundamentals (semiconductors, computing, AI)
• Second half: Innovation ecosystems, economic impact, policy, geopolitics, and leadership

Press enter or click to view image in full size

Press enter or click to view image in full size

Course Structure

The 13-week course follows a deliberate arc:

1. High-level introduction to technology and environment (Week 1)
2. Deep dive into semiconductors, materials, and devices
3. Build up through transistors, circuits, and chip design
4. Computing systems and architecture
5. AI applications and edge intelligence
6. Economic impact and interaction with society/law
7. Globalization, competition, and great power dynamics
8. Leadership and strategy (featuring Prof. Robert Burgelman on the Intel/Andy Grove case study)

Format: Mix of lectures, invited talks, and panel discussions. Some sessions run the full 110 minutes to accommodate panels.

Steve Blank: Historical Context: The Transience of National Dominance

How Nations Lose Their Dominant Position

Nations lose dominance through seven primary mechanisms:

1. Losing a war
2. Missing a technology transition ← Focus of this class
3. Missing new operational concepts ← Focus of this class
4. Losing allies
5. Declining economic power
6. Declining interest in global affairs
7. Internal/civil conflicts

Recent Geopolitical Shifts

• 1945–1991: Bipolar world (U.S. vs. Soviet Union)
• 1992–2018: U.S. as dominant global power (25+ years)
• 2018: National Defense Strategy identifies “2+3” threat environment with technology disruption
• 2022: Ukraine conflict reveals China’s asymmetric weapons capabilities
• 2024: Adversaries are:
• Creating/funding multiple regional crises via proxies
• Creating and amplifying internal U.S. division
• Building weapons systems at Silicon Valley speed
• Coordinating military/supply chains

Press enter or click to view image in full size

The U.S. reached peak dominance around 1950 and has been declining relative to other powers since 2000, while China has been rising rapidly.

Press enter or click to view image in full size

Press enter or click to view image in full size

The Perfect Storm: Semiconductor Technology Challenges

Three Converging Crises

1. Moore’s Law Slowdown

• Law of diminishing returns
• Requires massive R&D investment for incremental gains
• Only a handful of companies can afford to innovate at leading edge
• Node consolidation:
• 2001 (130nm): 17 companies
• 2007 (45nm): 12 companies
• 2015 (14nm): 4 companies
• 2020 (7nm): 2 companies
• 2022 (5nm): 1 company — TSMC

1. Dennard Scaling Ended

• Voltage scaling finished long ago
• Can’t simply shrink and reduce power proportionally
• Power density has become the limiting factor

1. Von Neumann Architecture Bottleneck

• Memory and processor are separate
• Data movement costs energy
• Communication cost (even on-chip) is expensive
• Most chips today are power-limited, not transistor-limited
• Could add more transistors if we could cool them

The AI Compute Crisis

Exponential Growth of Compute Requirements:

• Moore’s Law: Doubling every 2 years
• AI compute: Was doubling every 3–4 months (some sources say 6 months)
• This exponential growth is unsustainable

Training Costs:

• GPT-4: ~$100 million to train
• Gemini Ultra: Similar massive costs
• Compute measured in billions of petaflops

Energy Implications:

• GPT-4 training: ~50,000 megawatt-hours
• Every Google search: Wastes a AAA battery equivalent
• Chat GPT queries: Significant energy per simple question

Economic Unsustainability:

• Georgetown/DARPA/Nvidia/OpenAI study: If AI compute continues exponential growth, it could surpass U.S. GDP
• Linear GDP growth vs. exponential compute growth = unsustainable

Biden Executive Order: Requires reporting for models exceeding certain compute thresholds (100 billion petaflops for LLMs, lower for biology models)

The Efficiency Imperative

Intelligence Per Watt (IPW) Metric

Stanford CS researchers developed a new metric: Intelligence Per Watt

• Measures accuracy per unit of power
• Goes beyond chip-level efficiency to application-level efficiency

Key Findings:

• Hybrid approach (cloud + edge) is more sustainable than pure cloud
• Smaller, efficient models can achieve 70–90% accuracy for most tasks
• Economic impact potential:
• Best quality models: 90% efficiency → touches $20 trillion in economic activity
• Reasoning models: Lower efficiency → touches $6 trillion
• 70% accuracy = 90% of economy touchpoints for simple task

Three Elements for Extracting AI Intelligence:

1. Efficient chips and accelerators
2. Efficient model architecture
3. Post-training optimization

DARPA Software-Defined Hardware (SDH) Program

Goal: Break the power-performance tradeoff

• Traditional: High performance = high power, or low power = low performance
• DARPA program achieved 10x efficiency improvement
• Nvidia participated and improved GPU efficiency for sparse workloads
• Unknown how many features made it into modern GPUs

Paths to Efficiency

• Move away from von Neumann architecture
• Bring memory closer (high-bandwidth memory, advanced packaging)
• In-memory computing (compute where data resides)
• Novel computing paradigms (oscillator-based phase computing, etc.)

Semiconductor Ecosystem: A Complex Global Supply Chain

The Six Layers of the Semiconductor Industry

1. Chip IP Cores (Intellectual Property)

Licensed chip designs as building blocks:

• ARM: ~40% of chip IP cores globally
• Synopsys: 20%
• Cadence: 6%
• Others

2. Electronic Design Automation (EDA) Tools

Software for chip design — Three U.S. vendors dominate:

• Synopsys
• Cadence
• Mentor Graphics (now Siemens)

3. Materials and Subsystems

Specialized materials for manufacturing:

• Silicon wafers and crystal growing furnaces
• Gases & fluids: Neon, Fluorine, Argon, Helium, Arsine, Phosphine
• Photomasks, resists, top coats
• CMP pads/slurries
• Wafer handling equipment
• RF power equipment

Key suppliers: TOK, Sumitomo, Merck, DuPont, Entegris, JSR, Shin-Etsu, Fujifilm

4. Wafer Fab Equipment (WFE)

Machines that manufacture chips — Five companies dominate:

• Applied Materials (U.S.)
• KLA (U.S.)
• LAM Research (U.S.)
• Tokyo Electron (Japan)
• ASML (Netherlands) — supplies the most advanced lithography (EUV) ← Critical choke point

R&D Race Never Stops: These companies drive innovation on 12–18 month cycles, not TSMC. TSMC doesn’t invent the process — the wafer fab equipment companies do.

5. Chip Companies

Three Business Models:

A. Fabless Companies (Design only, no manufacturing):

• Apple, Nvidia, AMD, Qualcomm, Broadcom
• Send designs to foundries

B. Integrated Device Manufacturers (IDMs) (Design + Manufacture + Sell):

• Logic: Intel, Renesas, STMicroelectronics
• Memory: SK Hynix, Samsung, Micron, Toshiba
• Analog: Texas Instruments, ON Semiconductor

C. Foundries (Manufacture for others):

• TSMC (Taiwan) — dominates advanced logic manufacturing
• Samsung, UMC, GlobalFoundries, SMIC (China)

6. OSAT (Outsourced Semiconductor Assembly and Test)

Package and test finished chips:

• ASE, Amkor, JCET

Manufacturing Complexity

Creating a modern chip requires over 2,500 steps:

1. Growing pure crystalline silicon ingots
2. Sawing wafers from ingots
3. Polishing wafers
4. Material deposition/modification
5. Applying photoresist
6. Lithography — burning patterns
7. Etching and heating
8. Ion implantation for doping
9. Removing resist
10. Repeating 40–100 times for multiple layers
11. Cutting chips from wafer
12. Testing
13. Packaging and assembly

Yields: Mature processes achieve 30–90%

The TSMC Vulnerability

Critical Concentration of Risk

TSMC manufactures 90% of the world’s advanced logic chips

This creates an existential national security risk:

• If Taiwan falls to PRC, TSMC goes with it
• U.S. loses access to advanced chips
• All major tech companies depend on TSMC (Apple, Nvidia, AMD, etc.)

U.S. Response: CHIPS Act

Goals:

• Onshore chip logic, memory fabs, and packaging to the U.S.
• Reduce dependence on Taiwan

Investment:

• $53 billion in government incentives
• Over $200 billion in announced private sector investments

Historical Context: This represents a return to industrial policy after decades of free-market approach (Milton Friedman-style economics)

China’s Semiconductor Strategy: The Techno-Security State

Comprehensive National Integration

China has integrated four strategic frameworks:

1. Innovation-driven development strategy
2. National security strategy
3. Military-civil fusion development strategy
4. Military strengthening thoughts of Xi Jinping

Strategic Documents

• 15-Year Medium and Long-Term Science and Technology Development Plan
• Science, Technology, and Innovation 2030 Major Projects
• Military-Civil Fusion Development Strategy
• 15-Year Weapons and Equipment Development Strategy
• Defense Science and Technology Development Strategy
• 14th Five-Year Plan (2021–2025): Technology independence

China’s Semiconductor Ecosystem

Current State: China has competitors across the entire value chain, though most lag market leaders by 5–10 years

Chip IP Cores

• ~Emerging market
• Verisilicon as potential leader
• Mostly relies on international cores

EDA Tools

• ~10% market share
• Heavily dependent on U.S. vendors (Synopsys, Cadence, Mentor)
• Chinese incumbent: Hyperform
• New startups: Primarius, X-Epic, Hejian, Xpeedic, Semitronix, Amedac
• Note: Synopsys invested in Amedac (owns 20%)

Materials

• China has indigenous sources of specialized materials

Wafer Fab Equipment

• 5–10 years behind but catching up rapidly
• National effort with massive funding
• Companies: SMEE, Naura, Kingsemi, Piotech, AMEC
• Dependent on foreign equipment for leading-edge nodes

Fabless Companies

• Vibrant, large, well-funded segment
• CPUs/Data Center: Zhaoxin, Sunway, Phytium
• AI/ML: Bitmain, Cambricon
• Memory: YMTC
• Government encouraging them to build their own fabs

Foundries

• Heavily reliant on foreign fab equipment
• SMIC, HLMC, XMC, CanSemi, Huahong
• China subsidizes chip fabs by as much as 40% of revenues
• Hired thousands of engineers from Taiwan’s chip industry
• Massive IP theft from Taiwan and U.S. semiconductor industry

China’s Investment and Tactics

The “Big Fund” Investment Vehicle

China IC Investment Fund has invested in 60+ Chinese chip companies

Public investments:

• Logic: Tsinghua Unigroup, SMIC, Hua Hong Semiconductor
• Memory: YMTC, CXMT
• Stakes in VCs: SummitView Capital, Oriza Holdings, dozens of other funds

Funding:

• Phase 1 (2014): $20 billion
• Phase 2 (2019): $29 billion
• Ministry of Finance owns 37%

July 2022 Crisis: 13 current/former executives “under investigation” for corruption:

• Ding Wenwu (Big Fund President)
• Lu Jun (former Chief of Sino IC Capital)
• Multiple portfolio managers
• Recipients including Zhao Weiguo (former chairman of Tsinghua Unigroup)

Current Investment Frenzy

• U.S. Entity List driving massive internal investment
• Huawei sanctions were “the starting gun”
• 10,000+ companies with “semi” in their name seeking funding
• Nanjing Semiconductor University established (TSMC and Synopsys participating)
• Massive talent gap — hard to attract graduates
• Money flowing into capacity expansion, purchasing foreign equipment

Cautionary Tale — HSMC:

• Company with zero semiconductor experience
• Bought ASML machine
• Raised significant money
• Built fab incorrectly
• Collapsed — essentially a Ponzi scheme

Regulatory Weapons

China uses IP, standards, procurement, and competition authorities to advance goals:

Blocked Deals:

• Applied Materials’ acquisition of Hitachi Kokusai Electric

Forced Transactions:

• NXP: Sell RF power transistor business to Chinese state-controlled company
• Qualcomm: $975M fine, license patents to Chinese companies, enter joint venture
• Nvidia: Killed bid for ARM

Intellectual Property Warfare

Strategy:

• Encourage companies to register exclusive rights in China
• Develop IP services supporting protections
• Strictly enforce IP protections with increased penalties
• Explore requiring “legal” software pre-installed on all computers sold in China
• Government procurement promoting software legalization

Weaponizing IP Through Litigation:

• Chinese semiconductor companies (Fujian Jinhua, AMEC) use aggressive patent litigation
• Challenge exclusive use of proprietary technologies
• File copycat versions of U.S. cases in Chinese courts for counter-pressure
• Rely on utility patents (easier to create, harder to challenge)

What China Still Wants from Foreign Sources

Despite push for independence, China seeks:

• U.S. and foreign technology collaboration that remains unrestricted
• Countermeasures to work around U.S. restrictions
• Open-source technology platforms
• Talent programs to attract foreign experts to work in China

U.S. Export Controls: “Building a Dam Across Half the River”

October 2022: Comprehensive Semiconductor Export Controls

Restrictions on:

• Advanced chip manufacturing equipment
• Chip design software
• Outbound investment in semiconductors, quantum technologies, and AI

The Problem: Allied Countries Still Supplying

Despite U.S. controls:

• Dutch and Japanese chip equipment shipments to China have surged
• Record levels reached in 2024
• China imported over $60 billion in chipmaking machines in 2024
• Significant increases from both Japan and Netherlands

Steve Blank’s Assessment: “We’ve built a dam across half the river”

Export Control Debate

Hawks argue: Slow China’s progress, protect national security Critics argue:

• Further incentivizes China’s homegrown industry
• May be preferable to China invading Taiwan for TSMC

Policy Complexity: Not a simple answer. Trade-offs between:

• Economic impact
• National security
• Risk of Taiwan invasion
• Allied coordination
• Long-term strategic positioning

Economic and Innovation Context

Contribution to GDP Growth

Jason Furman (Harvard Kennedy School, former administration official):

• 92% of GDP growth came from information processing and software
• Without IT: First half of 2025 GDP growth would be 0.1% (essentially zero)

TSMC Forecast: Contribution of foundry and IP to GDP forecasted to grow significantly

The Great Stagnation Debate

Joe Malchow:

Pessimists (Robert Gordon, Northwestern):

• Age of economic growth is over
• “Low-hanging fruit” has been picked
• Remaining innovation requires too much effort
• Example: Boeing 707 revolutionized air travel, but Concorde wasn’t viable

Optimists (Mokyr, Nobel Prize winner, Northwestern):

• Nature of economy requires better metrics
• IT helps in different ways (globalization, quality products)
• Current measurement of economic growth is inadequate
• Techno-optimists see continued potential

Innovation Landscape: Startups vs. Large Companies

Large Company Challenges

Charles O’Reilly (Stanford GSB) — “Lead and Disrupt”:

• Study of S&P 500 over decades
• Late 1970s: Companies stayed in S&P 500 for 35+ years
• Today: Average life is 10–12 years
• Technology disruptions create steeper competition
• Less time at the top

Venture Capital’s Role

Professor Ilya Strebulaev (Stanford GSB) study:

• After ERISA Act (allowed pension funds to invest in VC)
• Over 40% of companies have been venture-funded (originally 43%, updated to 41%)
• Birth of venture capital fundamentally changed innovation landscape

Hardware vs. Software Investment Shift

Office of Strategic Capital / CFR study:

• Early days: Hardware and software investment were equal
• Hardware sometimes won over software
• By 2017: Ratio was 1:10 (hardware:software)
• This ratio has been problematic since 2012
• Recent trend: Starting to reverse as hardware becomes critical again

The Transistor: Born from a Perfect Storm

Historical parallel:

• AT&T/Bell Labs faced crisis: Couldn’t hire enough people to route calls
• Even hiring entire U.S. population wouldn’t meet demand
• Mechanical switches and vacuum tubes inadequate
• Solution: Study of solid-state switches
• Result: The transistor (75 years old, IEDM is 70 years old)

Key Takeaways

1. Do Not Underestimate China’s Ability to Innovate

• Chinese competitors exist in every stage of the semiconductor value chain
• While some/most lag market leaders, there is a tidal wave of:
• Government funding
• Venture capital
• Engineering talent

2. China Has a Comprehensive Strategy

• Integration of innovation, national security, military-civil fusion
• Economic tools deployed systematically
• The U.S. does not have an equivalent integrated approach

3. Technology Transitions Determine Dominance

• Missing critical technology transitions is a primary way nations lose dominant position
• Current transition: AI, advanced semiconductors, quantum computing

4. TSMC is a Critical Vulnerability

• 90% of advanced logic chip manufacturing concentrated in Taiwan
• Geopolitical risk is existential
• CHIPS Act attempts to address but takes years to build fabs

5. Export Controls are Incomplete

• Allies continue supplying critical equipment
• China actively developing domestic alternatives
• “Dam across half the river” metaphor

6. We Can Slow Them Down But Not Stop Them

• China’s determination, resources, and systematic approach
• Will continue advancing, even if at slower pace
• 5–10 year lag but catching up rapidly

7. Industrial Policy is Back

• After decades of free-market economics (Milton Friedman era)
• CHIPS Act represents major shift
• Federal funding historically drove innovation (NSF chart shows shift to industry)
• Question: Is government intervention good or bad in competition?

8. Efficiency is the New Frontier

• Can’t just add more transistors (power/cooling limits)
• Intelligence Per Watt matters
• Hybrid cloud-edge approaches
• Novel architectures beyond von Neumann

Questions for Consideration

1. How should the U.S. and allies coordinate export control policies more effectively?
2. What role should industrial policy play in maintaining technological leadership?
3. How can democracies compete with China’s integrated civil-military approach?
4. What are the implications of semiconductor supply chain concentration in Taiwan?
5. How should we balance technology collaboration with security concerns?
6. Is the current measurement of economic growth adequate for the IT age?
7. What is the right balance between startup innovation and large company R&D?
8. How do we make AI compute sustainable from energy and economic perspectives?
9. Should the U.S. adopt a more comprehensive techno-security state strategy?
10. What are the long-term implications of China’s demographic decline on their tech ambitions?

Course Logistics

Teaching Team:

• Ali Keshavarzi (Instructor)
• Joe Malchow (Co-creator, VC and software+semi+energy background)
• Gaurav Thareja (Collaborator, Applied Materials)
• Prof. Srabanti Chowdhury (Collaborator, EE — compound semiconductors, power electronics)
• Prof. Tom Lee (Collaborator, EE — circuit design, RF chips)

Guest Lecturers Include:

• Steve Blank (Technology, Innovation, and Great Power Competition)
• Mark Papermaster (AMD CTO)
• Prof. Robert Burgelman (GSB — “Strategy is Destiny,” Intel/Andy Grove case study)
• Industry experts from Intel, TSMC, Synopsys, and others

Related Course:

• MSE 292: Technology, Innovation, and Great Power Competition (Steve Blank)
• Starts from policy perspective, works down to technology
• EE 292P starts from technology, works up to policy
• Highly complementary

Class Format:

• 110-minute sessions (to accommodate panels and multiple speakers)
• Mix of lectures, invited talks, panel discussions
• Feel free to take breaks during long sessions
• Attendance-based (sign-in sheet)

Previous Quarter (EE 291N):

• Deep dive into latest semiconductor industry advances
• Speakers from Intel, Micron, ASML, Western Digital/SanDisk, Sony
• Topics: Transistors, DRAM, EUV, 3D NAND, packaging, silicon photonics, power/RF

Press enter or click to view image in full size