SuperAPI AI Directory
Partner with us中文AI newsFavorites
Skillsthinking-jensen-huang

thinking-jensen-huang

Agents & models

View repository

Updated 10-15 08:05

About this skill

Jensen Huang Thinking Framework

Core Thinking Models

Model 1: Accelerated Computing Paradigm

One-sentence definition: When general-purpose computing reaches its end, specialized acceleration is the only way forward—using hardware + software co-design to achieve order-of-magnitude performance leaps.

Applicable scenarios：

  • Technology architecture selection—general-purpose solutions vs. specialized solutions
  • Breaking through performance bottlenecks
  • Long-term technology bet decisions

Execution steps：

  1. Identify the endpoint of Moore's Law: Does this field still have a "free lunch" (performance gains from process improvements)?
  2. Room for computational specialization: Which computing patterns are repetitive and parallelizable?
  3. Hardware + software co-design: Do not optimize only hardware or only software—design them simultaneously
  4. 10-year-plus investment cycle: Accelerated computing is not a 1-year project, it is a 10-year bet
  5. Build ecosystem barriers: Make developers use your tools, your language, your platform

Classic case: CUDA began investment in 2006, with almost no returns in the first 5 years. But Jensen Huang insisted on the vision that "a GPU is not just a graphics processor, it is a general-purpose parallel computing engine." By 2012, AlexNet used CUDA to train deep neural networks, igniting the AI revolution—a 6-year bet finally paid off.

Model 2: Ecosystem Platform Thinking

One-sentence definition: Do not sell products, sell a platform; do not compete at a single point, build infrastructure that everyone depends on.

Applicable scenarios：

  • Product strategy—build a tool or build a platform
  • Competitive barrier design
  • Developer relations strategy

Execution steps：

  1. Identify "the underlying capability everyone uses": What is the foundational need across industries and applications?
  2. Be the "industrial revolution of AI": Do not build a particular AI application, build the "electric power company" for all AI applications
  3. Lower the barrier to use, increase the cost of migration: CUDA makes GPU programming simple, but once development is based on CUDA, switching platforms is extremely costly
  4. From hardware to software to systems to services—full stack: Each layer makes the ecosystem stickier
  5. Continuously invest in the developer community: University courses, research grants, developer conferences—future developers are your ecosystem moat

Classic case: NVIDIA does not merely sell GPU chips. CUDA is a programming language and platform, cuDNN/tensorRT is the foundation of AI frameworks, DGX is an AI supercomputer, and Omniverse is a 3D collaboration platform—each layer makes customers more dependent on the NVIDIA ecosystem.

Model 3: Vision-Driven Long-Term Betting

One-sentence definition: See the world 10 years from now, work backward from the endpoint to decide what to do today, and then persist until the whole world catches up.

Applicable scenarios：

  • Strategic planning—5-10 year technology roadmaps
  • Investment decisions—short-term pain vs. long-term gain
  • Persuading teams/investors to believe in an "impossible vision"

Execution Steps：

  1. Imagine the world 10 years from now: How far will technology have advanced? What will humanity need most?
  2. Work backward through key milestones: 10-year goal → 5-year checkpoint → 3-year validation → 1-year action
  3. Identify the "betting points": Which technology directions are the必经之路 to that future?
  4. All-in betting: Not diversified investment, but concentrating resources on 1-2 key directions
  5. Endure being mocked as "too early": CUDA was mocked for years, and through the wild swings of cryptocurrency mining, he stuck with the AI direction

Classic Case: When Jensen Huang launched CUDA in 2006, Wall Street analysts questioned, "Why make GPUs do general-purpose computing? Aren't gaming graphics cards selling just fine?" When DGX-1 was launched to OpenAI in 2016, many people did not understand. When ChatGPT was released in 2022, the whole world finally understood NVIDIA's bet from 16 years earlier.

Model 4: CEO as Chief Evangelist

One-sentence definition: The CEO's primary responsibility is not management, but making the whole world believe in your vision—if you are the only one who believes in an idea, it is nothing.

Applicable Scenarios：

  • Company strategic transformation
  • New market education
  • Brand building and culture building

Execution Steps：

  1. Use one simple core narrative: NVIDIA's core narrative—"accelerated computing is the infrastructure of AI"
  2. Repeat infinitely: Every speech, every interview, every earnings call—the same narrative, from different angles
  3. Let demos do the talking: Not a PPT saying "our technology is very powerful," but a live demonstration
  4. Wear the leather jacket: Brand recognition—you don't need to see the name; when you see the black leather jacket, you know it's Jensen Huang
  5. Translate complex technology into human language: Not "FP8 Tensor Core," but "the power plant of AI"

Classic Case: GTC evolved from a small GPU technology conference into the "Woodstock of AI"—Jensen Huang's keynote became an industry bellwether, not because the PPT was well made, but because year after year he made prophecies in the right direction.

Model 5: Crisis to Opportunity

One-sentence definition: Every industry crisis is an opportunity to reshuffle the deck—when others retreat, you advance.

Applicable Scenarios：

  • Strategic decisions during a market downturn
  • Seizing the moment when competitors make mistakes
  • Managing uncertainty during technological transition

Execution Steps：

  1. Keep investing during a crisis: While others cut R&D, you increase it
  2. Identify the structural problems exposed by the crisis: What caused this crisis? Which old models proved unviable?
  3. Double down on future directions: The collapse of the old order means space for a new order
  4. Stockpile "ammunition": Do not inflate in good times; maintain reserves so you have resources to attack in a crisis

Classic Case: During the 2008 financial crisis, NVIDIA's stock price plummeted, but Jensen Huang did not cut CUDA investment. When the 2018 cryptocurrency crash led to a backlog of GPU inventory, he continued to double down on the AI direction. During the 2022-2023 semiconductor industry downturn, NVIDIA launched H100 against the trend, and its market value soared in the AI wave.

Decision Framework

技术趋势/市场信号
    │
    ▼
[第1层：愿景对齐]
这个方向是否符合"加速计算是未来"的核心愿景？
    │
    ├─ 不符合 → 放弃，不论短期多赚钱
    │
    ▼
[第2层：10年价值评估]
10年后这个世界会怎么运转？这个方向在10年后的世界中位置如何？
    │
    ├─ 10年后不重要 → 不投入核心资源
    │
    ▼
[第3层：生态杠杆]
这个方向能让更多人依赖我们的生态吗？
    │
    ├─ 不能 → 考虑是否值得做
    │
    ▼
[第4层：全栈可行性]
我们能做全栈吗？（硬件+软件+系统+服务）
如果能全栈，护城河多深？
    │
    ▼
[第5层：投资节奏]
前期投入多少？容忍几年无回报？
里程碑验证点在哪？
    │
    ▼
[第6层：布道计划]
如何让行业相信这个方向？开发者大会、学术论文合作、标杆客户

Decision Principles：

  • Vision > Financials: Short-term profit is subordinate to long-term vision; do not make money that does not align with the vision
  • Platform > Product: Build the foundation everyone depends on, not just a single application
  • Full-stack > Single point: Each additional layer doubles the depth of the moat
  • Persistence > Pivoting: If the direction is right, all that remains is to persist until the whole world catches up

Classic Quotes

  1. "The more you're in the front, the more you're wrong. But you have to be wrong to be at the front." — Remarks on innovation and risk-taking in multiple interviews

  2. "We're in the business of building the future, and the future of computing is accelerated." — Core narrative of multiple GTC keynotes

  3. "I think NVIDIA's greatest contribution is we made it possible for software developers to use parallel computing." — On the long-term value of CUDA, mentioned in multiple interviews

  4. "We had to invent a new way of doing computing. And that's what CUDA was." — On the origin and vision of CUDA

  5. "AI is the most powerful technology force of our time." — GTC 2017 and multiple keynotes thereafter

  6. "Buy as many GPUs as you can. The more you buy, the more you save." — Advice to customers on multiple public occasions in 2023-2024, now an industry meme

  7. "We're building the infrastructure for the AI industrial revolution." — Core positioning that recurs in earnings calls and GTC keynotes

  8. "The company is not about making products. The company is about making a contribution to the world." — On NVIDIA's mission and culture

Practical Templates

Template 1: 10-Year Vision Backcasting Method

## 10年愿景倒推：[领域/项目]

### 10年后的世界（2036年）
- [领域]会变成什么样子？
- 什么能力会成为"基础设施"级别的需求？
- 谁在使用这些基础设施？做什么？

### 关键里程碑
| 时间 | 里程碑 | 验证指标 |
|------|-------|---------|
| 10年 | [愿景实现] | [指标] |
| 7年  | [阶段性成果] | [指标] |
| 5年  | [生态初步形成] | [指标] |
| 3年  | [技术验证] | [指标] |
| 1年  | [第一步行动] | [指标] |

### 押注点
- 核心押注：[描述]
- 投入资源：[估算]
- 容忍无回报期：[X年]

### 风险
- 如果愿景方向对了但时机不对：[应对]
- 如果竞争对手先到：[应对]

Template 2: Ecosystem Platform Design

## 生态平台设计：[项目/产品]

### 底层能力识别
- 什么是跨[行业]的基础能力？
- 现在谁在提供？效果如何？

### 全栈架构
| 层级 | 内容 | 护城河深度 | 竞争对手 |
|------|------|----------|---------|
| 硬件 | [描述] | [深/中/浅] | [对手] |
| 软件/SDK | [描述] | [深/中/浅] | [对手] |
| 开发框架 | [描述] | [深/中/浅] | [对手] |
| 服务/平台 | [描述] | [深/中/浅] | [对手] |

### 开发者粘性设计
- 学习成本：[高/中/低]（开发者入门门槛）
- 迁移成本：[高/中/低]（换到竞品的代价）
- 社区建设：[计划]

### 生态健康指标
- 开发者数量：[目标]
- 基于平台的应用数：[目标]
- 开发者NPS：[目标]

Template 3: Accelerated Computing ROI Analysis

## 加速计算ROI分析：[场景]

### 当前通用计算性能
- 吞吐量：[数值]
- 延迟：[数值]
- 成本：[数值]
- 功耗：[数值]

### 加速计算方案
- 架构：[描述]
- 预期吞吐提升：[X倍]
- 预期延迟降低：[X倍]
- 开发投入：[估算]
- 硬件成本：[估算]

### 摩尔定律终点评估
- 通用计算还有多少免费提升空间？[估算]
- 加速计算的理论天花板：[计算]
- 当前距离天花板：[百分比]

### 决策
- 短期ROI：[X倍，Y年回本]
- 长期护城河：[评估]
- 建议：[做/不做/条件性做]

Application Scenarios

When to activate this skill：

  1. Technology Roadmap Formulation: When you need to formulate a 3-10 year technology investment plan
  2. Platform Strategy: When considering a transformation from product to platform
  3. Long-term Bet Decisions: When you have a "too early" vision and need to assess whether it is worth persisting
  4. Ecosystem Building: When you need to build a developer ecosystem and a moat
  5. Opportunity Identification in Crisis: When the industry hits a trough and you consider whether to invest counter-cyclically
  6. Full-stack vs Single-point Decisions: When deciding which layer of the technology stack to build
  7. Evangelism/Narrative Design: When you need to make the industry believe in a new direction

Typical Trigger Scenarios：

  • "Will this direction still matter in 10 years?"
  • "Can we build a platform rather than just a product?"
  • "Everyone says it's not worth investing in, but I think the direction is right"
  • "How to build a certain moat in an uncertain future"

Anti-patterns

  1. ❌ Chasing short-term hype: During the cryptocurrency boom, Jensen Huang made it clear that NVIDIA's core is AI, not mining. "Mining is temporary, AI is permanent."

  2. ❌ Doing only hardware or only software: NVIDIA's success is not "good GPUs," but a systems engineering effort of "GPU + CUDA + ecosystem." Single-point competitors will always be defeated by full-stack players.

  3. ❌ Ignoring the developer community: A platform without developers is an empty shell. Jensen Huang spends hours personally doing demos at GTC every year—the CEO's time invested in developer relations.

  4. ❌ Cutting R&D during a crisis: Every market downturn is the time for NVIDIA to increase investment. "Competitors are hibernating, we are training."

  5. ❌ Short-sighted financials orientation: Jensen Huang has repeatedly abandoned directions that were highly profitable in the short term (such as going all-in on gaming graphics cards, going all-in on mining chips), because they did not align with the long-term vision of AI accelerated computing.

  6. ❌ Complex narrative: NVIDIA's core narrative has not changed in 20 years—"accelerated computing is the future." If you cannot explain your vision in one sentence, it means you have not thought it through clearly.

Compiled based on Jensen Huang's GTC keynotes (2013-2025), public interviews, earnings call transcripts, the biography "The Nvidia Way" (by Tae Kim), and an MIT Technology Review interview. Last updated: 2026-04-14