newtype AI Business, Industry and Investment Observations

452 posts + BIG NEWS timeline. Starting from the 2024-03 OpenAI vs Musk open letter, ending with NVDA acquiring Groq in 2026-05.

The theme of this wiki = business / industry / investment / individual economy. For tool reviews, see newtype Model Evaluation, Fine-tuning and Hardware Evolution, for the tool stack see newtype · AI Tool Stack Practice and Evolution, for the knowledge system see newtype Knowledge System and the Full Evolution of Life OS.

0. Reading Route

If you want to understand big company wars and mergers & acquisitions → §2 (Industry Events Timeline)
If you want to understand the model layer vs application layer debate → §3 (The Three-Faction Debate over Scaling Law)
If you want to do AI startups / independent development → §4 (Lao Huang's methodology) + §5 (Personal business experiences)
If you want to do investment → §6 (Three investment logics: NVDA / Starlink / Apple)
If you want to understand the AI-era landscape → §7 (Three-layer framework / Page Rank → Agent Rank / The end of intermediary monopolies)
If you want to know what to do in 2026 → §8 (Three things Lao Huang recommends)

1. Core Beliefs / Immutable Anchors

Lao Huang's business judgments over two years can be distilled into 7 immutable anchors:

1.1 Do not build a house on someone else's foundation (2024-03 Post 4 / Zhu Xiaohu)

"I always tell domestic entrepreneurs,never build a house on someone else's foundation. "

→ Domestic large model companies insist on building both models and applications themselves, so entrepreneurs are even more afraid to use them.

1.2 Models are eating applications; position first, then wait for the wave (2024-05 Post 230 / Classic)

"Since large models are extremely centralized operating systems, they will definitely eat up many, many application tracks. For entrepreneurs, perhaps they can only get a share after this monster has eaten almost everything. So, don't rush to make a move.

You need to position yourself and wait for the wave."

1.3 LLM = OS (2024-05 Post 227 / Classic)

LLMs have all the components an OS should have:

OS Component	LLM Equivalent
Memory management	Context window
File system	Conversation history + knowledge base
Drivers	Function Call
User interface	Natural language interaction (replacing CLI / GUI)

Stringing together OpenAI's updates along the way,it truly follows the logic of an OS.

1.4 General at the bottom + vertical at the top (2024-08 Post 436)

"For the general large models at the bottom, use whichever is good or cheap. For the vertical small models at the top, truly invest in fine-tuning and optimization,to become a moat. "

1.5 Context Matters (2025-07 Post 1495 / Lao Huang's business philosophy)

When asked about "choosing a product track", he answered:

"Which direction to go depends on what you believe in. For example, I personally firmly believe:Context Matters! The Prompt House and Obsidian MCP I have already released, and the plugins I am about to build, are all developed based on this philosophy. These products are purely carriers to express my philosophy."

1.6 Pro Users definitely use multiple tools, so Prompt management is a real demand (2025-11 Post 1947)

"Pro Users definitely use multiple AI tools at the same time, and will definitely have demands for Prompt management and invocation."

→ The commercial moat of Prompt House.

1.7 Data is the barrier, models are consumables (2025-08 Post 1516)

"The Claude model is hard currency right now.Whoever provides more model quota will have a product that takes off(Kiro). Whoever cannot bear the cost and starts to scale back will go downhill (Cursor)."

→ Fulfilled: Cursor faced mass unsubscriptions in 2025-07; Anthropic / OpenAI / Google burned their own cash to grab market share.

2. Industry Events Timeline (BIG NEWS Compilation)

2.1 2024 H1: Musk vs Altman War + Multimodal Breakthroughs

Date	Event	Lao Huang's Observation
2024-03-06	Musk sues OpenAI / OpenAI counters with open letter	"Open" redefined = universal access rather than open source
2024-03-25	Amazon adds another $2.75 billion to Anthropic, bringing total investment to $4 billion	One of the largest generative AI funding rounds
2024-04-22	Mistral Mixtral 8x22B open-sourced	MoE goes mainstream
2024-04	xAI valued at $18 billion / Sequoia invests	"Understand the true nature of the universe"
2024-05	GPT-4o free + Apple-OpenAI negotiations	iOS 18 integration
2024-05	Microsoft Copilot+PC	On-device AI-ification
2024-06	OpenAI and Apple officially partner	GPT-4o integrated into iOS/iPadOS/macOS
2024-06	Former NSA director Paul Nakasone joins OpenAI board	Military-political background involvement
2024-06	Claude 3.5 Sonnet + Artifacts	Lao Huang "will not renew ChatGPT Plus next month"

2.2 2024 H2: o1 + 5 Levels of AGI + Startup Disillusionment

Date	Event	Lao Huang's Observation
2024-07-09	OpenAI stops API access for unsupported countries	Domestic developers switch to Microsoft Azure / OpenRouter
2024-07-14	OpenAI 5 Levels of AGI(Post 370 / Classic)	1.Chatbot 2.Reasoner 3.Agent 4.Innovator 5.Organization
2024-08	NVDA Blackwell B200 launch delayed	Design flaw / 2025-Q1
2024-08-06	Musk restarts OpenAI lawsuit	Sequel to the protracted battle
2024-08	Canva acquires Leonardo.Ai	Visual AI consolidation
2024-09	OpenAI o1 / o1-mini	Era of slow thinking
2024-09-22	Microsoft restarts Three Mile Island nuclear plant ($1.6 billion refurbishment)	20-year partnership with Constellation Energy
2024-09	Meta: Llama 3 uses 20,000 GPUs, Llama 4 over 100,000, Llama 5 even more	Computing power arms race
2024-10	OpenAI secures another $4 billion revolving credit facility	Total deployable capital reaches $12.6 billion
2024-10-23	Lao Huang asserts OpenAI chose the wrong core users(Post 562)	Code worse than Claude / Canvas worse than Artifacts / Search worse than Perplexity / GPTs dispensable
2024-10-27	Microsoft discusses $14 billion equity investment in OpenAI	Profit-priority repayment structure exposed
2024-11	Former Meta VR/AR head joins OpenAI	Signal of hardware pivot
2024-11	Microsoft open-sources Magentic-One multi-Agent system	5-layer architecture (Orchestrator/WebSurfer/FileSurfer/Coder/Terminal)

2.3 2025 H1: DeepSeek Moment + Claude Code Takes Off

Date	Event	Lao Huang's Observation
2025-01-12	DeepSeek APP launches	National-level hit
2025-01-07	NVDA Project Digits / GB10 $3000	Lao Huang "definitely getting one"
2025-01-09	Kai-Fu Lee interview: "Scaling Law is slowing down"	Accelerated commercialization + accelerated elimination
2025-02-19	Perplexity valued at $18 billion	$100 million funding / NVDA + SoftBank participate
2025-02	Claude 3.7 Sonnet	Number one in AI Coding
2025-03	Manus goes viral in Chinese-speaking circles	First wave of general-purpose Agents
2025-03	DeepSeek-V3-0324 at rock-bottom price	Roughly 1/10 the cost of Claude with similar performance
2025-04	Gemini 2.5 Pro is god-tier	Scans codebases + long articles
2025-04	OpenAI Codex CLI open-sourced	5000 GitHub stars
2025-05	Anthropic Claude 4(SWE-bench 72.5%)	Can work continuously for 7 hours
2025-05	OpenAI Codex(Based on codex-1)	AI Coding enters the cloud Agent stage

2.4 2025 H2: Agent Era + Claude Skills

Date	Event	Lao Huang's Observation
2025-06	Apple opens Apple Intelligence base model + Foundation Models Framework	On-device AI opened up
2025-07	Perplexity valued at $18 billion + $100 million funding	NVDA + SoftBank
2025-07	Alibaba Qwen3-Coder open-sourced	Domestic programming model
2025-07	ByteDance Trae 2.0 SOLO mode	Domestic AI IDE localization
2025-07	Lao Huang "AI coding is a battleground for every company" (Post 1402)	Not just a track, it relates to the future of models
2025-08	Claude Code Sub-agents	Model tiers (Haiku/Opus)
2025-10-19	Anthropic Claude Skills appear	Lao Huang builds Super Analyst / Super Writer
2025-10	OpenAI ChatGPT Atlas + Perplexity Comet	Dual giants of AI browsers
2025-11-03	Domestic open-source model Skills testing (by Lao Huang)	Kimi K2 > GLM-4.6 > Minimax M2
2025-11-09	Kimi K2 Thinking → Lao Huang launches newtype AI OS(Post 1965)	Inflection point for practical use of domestic models
2025-11-18	Gemini 3 Pro	Two-option (Flash / Pro) streamlined product
2025-11-19	OpenAI ChatGPT Group(Group chat feature)	Lao Huang "OpenAI's most promising product"
2025-11-30	Claude Opus 4.5(SWE-bench 80.9%)	Surpasses Gemini 3 Pro / GPT-5.1
2025-12-02	Lao Huang "NVDA 10-year holding logic"(Post 2076 highlight)	Accelerated computing + digital twins + AI for Science
2025-12-14	GPT-5.2(Three tiers: Instant / Thinking / Pro)	"Reaches human expert level for the first time"
2025-12-21	Gemini 3 Flash (3 times faster than 2.5 Pro)	Cost-performance ratio drops again

2.5 2026 Q1: OpenCode Black Card + Clawdbot

Date	Event	Lao Huang's Observation
2026-01	OpenCode Black Card $200 sells out in seconds	Lao Huang "OpenRouter should also release one"
2026-01-09	Anthropic Cowork	"Claude Code applicable to other work"
2026-01-16	Codex plugin marketplace	Lao Huang revises newtype OS submission
2026-01-19	Coze version Skills + skill store	New opportunities for domestic ecosystems
2026-01-24	NVDA acquires Groq for $20 billion(Post 2318 / inflection point signal)	"When a giant acquires a technological heretic company, it signals an era inflection point"
2026-02	Clawdbot / OpenClaw self-evolution framework	Lao Huang believes "it is just a toy until privacy and security are resolved"
2026-02-05	US stock SaaS sector sees violent turbulence(Post 2373)	"The end of intermediary monopolies"
2026-04	Anthropic requires identity verification (passport + selfie)	Lao Huang announces switch to OpenCode
2026-04	GitHub pauses Copilot registrations	Agent computing power cannot keep up
2026-04	DeepSeek V4	Strong reasoning
2026-04	Claude Opus 4.7 GitHub Copilot promotion	Lao Huang's current main tool
2026-05	Perplexity Personal Computer	$200-tier investment tool
2026-05	Zsxq officially releases Skill	Domestic platforms follow Anthropic

3. The Three-Faction Debate over Scaling Law (Classic Topic)

3.1 Faction A: Zhu Xiaohu (2024-03 / Realism)

"Right now OpenAI's technology iteration curve is still relatively steep; open source is definitely one year or even a year and a half behind non-open source. But once the non-open source technology iteration curve slows down, open source will catch up."

Core judgment:

lies in "whether a hundred thousand cards can actually be deployed"
Large models are already in a price war (tens of millions for deployment in early 2023 → five million in June → less than one million by year-end)
Large model companies cannot survive alone; they must compete on applications + ecosystems
Midjourney cannot hold its ground, because the consumer-facing market has too low usage frequency
Sora is a new approach, but "technology iteration cannot stay steep forever"
Before controllable nuclear fusion is achieved, I do not believe the Earth can truly reach AGI

3.2 Camp B: Cohere CEO Aidan Gomez (2024-10 / Pragmatist)

"Scaling Law still holds, and will remain valid for a considerable period of time. But the reality is,most companies' businesses cannot yet support the deployment cost of GPT-4-level models. "

The trend for a long time to come = general models and vertical models coexisting:

First use general large models for prototype design
After the model works, do vertical fine-tuning

3.3 Camp C: Kai-Fu Lee (2025-01 Post 708 / Evolved Pragmatist)

"1. Scaling Law is slowing down. 2. Only big tech companies can continue to build super large models; the success probability for startups is extremely low. 3. The models that AI applications need are small enough, fast enough, cheap enough, and capable enough, not necessarily the largest and most general. 4. If super large models (such as Claude 3.5 Opus) cannot be made cheap, they do not need to be released; keep them internally as 'teachers' to train 'students'. 5. In 2025, AI will accelerate commercialization and also accelerate elimination."

Golden quote:"When opportunities come, you must make decisions bravely; the same applies when opportunities disappear."

3.4 Lao Huang's comprehensive judgment (2025-03-27 Post 1040)

"Many people in China believe that even if the models are not good enough, they can still compete on applications. But the current situation is,models are evolving into complete applications, devouring various vertical tracks and traditional positions. "

→ Model = Application, they are not separate. This is a further extension of LLM = OS.

4. Entrepreneurship Methodology in the AI Era (Lao Huang's Theory)

4.1 Two-layer flywheel of personal capability (2025-03-20 Post 1010 / 30 likes / Classic)

Lao Huang's reflection on "why he is so capable":

Outer flywheel:

Information gathering and rapid learning (12 years in communication and marketing)
Strategic capability (defining problems → finding solutions)
Narrative capability (content capability / huckstering capability)
Product-building capability (a byproduct of two failed startups)

Inner flywheel (key):

"My greatest strength is my core, the gradually improving and complete personality. This is the result of more than ten years of continuous【seeking inward】. This core, combined with those external capabilities, forms two flywheels. Once they start spinning, everything you want will come along, without extra effort."JUST DO IT. "

4.2 Advance by all means + Do not be bound by tools

Lao Huang cannot do professional UI, does not understand code, and cannot use professional software.Solutions:

Cannot draw UI → Look at a lot on Dribbble + copy
Cannot use software → Draw UI in Keynote
One principle:Advance by all means

4.3 Non-technical independent development process (2025-05 Post 1207 / Featured)

When asked "how to manage code / testing / progress":

Weigh your own capabilities and resources — One person, non-technical, spend a little money on AI products, never touch large projects
Is technology necessarily the success factor of a project? Not necessarily — Even if a technical barrier is needed, I can turn it into a business problem, an arithmetic problem, rather than a technical problem. For example,Can't I just buy ready-made APIs? Wholesale first, then retail
Marketing / operations / product are my strengths; I structure the entire project starting from these areas

4.4 The three-piece set of Cursor + Claude 4 + v0 (2025-06 Post 1376)

"99% of it was done with Cursor. Claude 4 is extremely capable; it writes all kinds of records on its own. During the entire development process of Prompt House, I did not write any documentation; everything was done in Cursor.

Front-end UI uses v0.dev. In this regard, it is even more capable than Claude 4.

There are a large number of Serverless products overseas, which are very suitable for early-stage product trial and error. The entire ecosystem is far stronger than that in China."

4.5 Don't compete with giants on the "input box" (2025-05 Post 1229)

"Chatbots like ChatWise, if they cannot find a narrow niche to survive, will eventually be crushed by big companies. Because you are essentially competing with big companies to build the most basic, most general'input box'. This is the biggest taboo in AI entrepreneurship."

→ Lao Huang built Prompt House not as an "input box", but as "Prompt management + MCP protocol scheduling" (atomized middle layer).

4.6 GUI Agent is work for giants (2025-03 Post)

"GUI Agent is work for platforms; for example, Apple and Google will definitely do it. For developers and startups, there is no advantage in going in this direction, and it is easy to fail."

4.7 On-device AI is still early, do not bet on it (2025-04 Post 1077)

"My consistent view is that on-device AI, especially on mobile,has significantly fallen behind the broader industry. If I were a developer, I would not focus too much on AI at this stage, and would return to application thinking."

4.8 Two key points for mobile: permissions and computing power (2025-04 Post 1077)

Permissions: Held by phone manufacturers, they will definitely not release them
Computing power: Mainstream phones run 3B, Qualcomm can reach 7B, targeted fine-tuning is mandatory

→ AI on mobile must wait until phone manufacturers + chip manufacturers have taken their share.

4.9 Experience from Zhu Xiaohu's investment in fancy tech (2025-05 Post 1320)

"In all likelihood, this company does not have any particularly awesome technology. It is all integrated application of ready-made technologies and products.

PR companies and advertising firms of all sizes are in this track, all researching and practicing. On the other hand, startups like Lovable have also joined in.Therefore, technology is secondary; a Leader who understands technology, has sufficient industry experience, and has decision-making authority, plus a small elite team, is the key to starting out. "

4.10 No need to take courses to learn prompts (2025-05 Post 1223)

"What is the essence of prompts? Logic, methodology. Is this some special course? Isn't it just breaking down a task and giving a clear brief? If a new person joins the company and I need to train them, that's exactly how I teach it.

So, if I really want to improve my prompt-writing skills,I would just study logic directly; even books like 'Pyramid Engineering' or similar would be more useful. "

4.11 Data is the moat, not RAG / fine-tuning (2025-04 Post 1004)

"For the model foundation of AI assistants,the key issue is how to fine-tune according to the business characteristics of the enterprise or vertical industry itself. Therefore, for AI implementation in vertical industries,the key lies in the enterprise's business data and its quality. So for vertical industries, data is the key."

4.12 Creator methodology in the AI era (2025-12 Post 2258)

"Creators in the AI era do not only output content such as text and video; also,this kind of middle-layer stuff is also an excellent carrier!"

→ Lao Huang's newtype Profile / Skills / Prompt House are all "carriers" of the middle layer.

5. Personal Business Experience (Lao Huang Build in Public)

5.1 Key startup milestones

Time	Event	Input / Output
2023-05	First SaaS product (Chatbot+RAG+online search)	Self-funded front-end + back-end hire / Before Cursor existed / Shut down on New Year's Day 2024
2024-03	Zsxq community launched (newtype)	1k followers → estimated 1500+ by 2026-05
2024-09	newtype.pro Blog launched	Hand-built with v0 + Cursor
2025-04	Registered US company via Stripe Atlas($500)	Delaware LLC / 2 working days / $100/year Registered Agent
2025-04	Mercury Bank Registration successful	Approved in 2 days
2025-06	Tax ID (EIN)Received (waited 1 month)	Can withdraw Stripe income
2025-06	Prompt House Mac version launched at $9.98	Electron / Stripe / Resend / Supabase
2025-06	Prompt House sold over 100 USD → filed taxes	Stripe Taxually starting at $90/month
2025-06	Apple Developer application ($98.99)	Requires D-U-N-S Number / US company entity
2025-08	Prompt House Google traffic picking up	SEO taking effect
2025-09	First paid article published on WeChat Official Account	24,500 followers, first article views < 500, paid < 10 → fluctuating growth
2025-09	Substack replaced Hugo($50 domain binding)	newtype.pro switched backend to Substack
2025-11	Substack 200 subscribers + 10 paid	Content indexed by Perplexity/ChatGPT/qwen
2025-11	US Capital One secured credit card $200	Started building US credit
2026-04	wiki.newtype.pro launched (Karpathy LLM Wiki approach)	SEO + GEO secondary site
2026-04	Codex took over SEO / GEO automation	No need to pay for SEO consultants
2026-05	Bing Webmaster indexing of wiki took effect (comparison pages)	Comparison pages scrape GEO data

5.2 Overseas identity infrastructure checklist

Asset	Tool / Service	Purpose
US SIM card	JD.com physical card $3/month	SMS
UK SIM card	giffgaff 10£ Pay as you go	SMS
eSIM	Hongcha (Austrian number)	Overseas service registration
US company	Stripe Atlas $500 / Delaware LLC	Receive Stripe payments / tax filing
US address	Stable (Stripe partner)	Receive and forward mail
Bank account	Mercury (approved in 2 days)	Corporate account
Credit card (corporate)	Mercury	Register for overseas services
Credit card (personal)	Capital One secured $200	Personal overseas use
Tax	Stripe + Taxually $90/month	Sales tax automation
Apple Developer	$98.99/year	App Store listing
Google Fi	T-Mobile alternative	5G/SMS outside the US

5.3 Lao Huang's current (2026-05) content platform strategy

Platform	Role	Content
YouTube	Long-termism, high-quality information source	Weekly video updates
Substack(newtype.pro)	Overseas home base	Long-form articles / Newsletter / Thread / Note / Chat
wiki.newtype.pro	Structured knowledge map + GEO + long-tail search	Karpathy LLM Wiki approach
X(Twitter)	Moments + overseas social	Blue verified badge / Grok integration
Reddit	First stop for going overseas	Obsidian MCP views 5K+
Medium	Accumulate long-form articles	Backup
Zsxq	Community-exclusive	Domestic home base
WeChat Official Account	Paid reading (launched 2025-10)	50 WeChat Beans per article
Patreon	Overseas Newsletter	Synced with Zsxq content
Video Channels	Domestic traffic	Not used as home base

5.4 Domestic platform vs overseas platform strategy (2024-12 Post 689)

"This is why I do not take domestic platforms as my home base. They often disgust you, restrict you for inexplicable reasons, and force you to self-censor."

→ 5 overseas platforms + 4 domestic platforms, with overseas as the home base.

6. Investment Logic (Lao Huang's Views)

6.1 Logic for holding NVDA for 10 years (2025-12 Post 2076 / Featured)

Lao Huang's view:If I could only hold one stock for 10 years, I would choose NVDA.

Two key insights:

Most people equate accelerated computing with AI, but Nvidia has been deeply engaged in HPC for 15 years, and CUDA was not created for AI
The end of computing power is physical simulation, not language generation

Why ASIC/TPU cannot replace NVDA:

Dimension	ASIC/TPU	NVDA GPU + CUDA
Precision	FP16/FP8/INT4 (high precision removed for AI)	FP64+ (required for scientific computing)
Strength	Matrix multiplication accelerator	Complex geometry + graph theory + matrices
Applicable to	Text generation	Protein folding / chemistry / meteorology / energy / life sciences
Migration cost	-	Rewriting the software infrastructure of entire disciplines(Scientists cannot switch)

6.2 TPU is not for ordinary companies (2025-11 Post 2066)

"TPU cannot replace GPUs, because currently only giants have the capability and eligibility to use it.When you buy TPU, you are not buying chips, but entire cabinets, entire cabling solutions, and you even have to renovate your data center for it. If you don't have extremely large training demands, TPU is simply not cost-effective for you. On top of that, you need a talent pool that understands XLA, otherwise you can't make it work at all."

6.3 Starlink IPO Investment Logic (Post 2116, 2025-12)

"SpaceX is turning 'first-mover advantage' into 'physical monopoly'. It is not just a rocket company; it will become 'the sole monopolist of humanity's future space-based infrastructure.'"

Lao Huang: "I will definitely buy some when it IPOs next year, get a seat at the table first."

6.4 Signals from NVDA's $20 Billion Acquisition of Groq (Post 2318, 2026-01-22)

"There is an unwritten rule in Silicon Valley:when giants start acquiring startups with 'heretical' technology roadmaps, it often means an inflection point of an era has arrived. NVIDIA announced the $20 billion acquisition of Groq — this inference chip startup has never turned a profit commercially, and its product shipments are less than a tiny fraction of its competitors'. But this deal..."

6.5 Apple AI Strategy Observations

Apple's self-developed models are formidable (Post 324, 2024-06 / dual-model architecture: small on-device + large cloud)
But the repeated back-and-forth in strategic choices and partnerships with OpenAI / Tencent / ByteDance shows an unclear roadmap
Apple Intelligence is "total garbage" (Post 1354, 2025-06 / castrated mainland China version)

6.6 Positioning of Alibaba / Tencent / ByteDance (Post 2094, 2025-12)

Company	Positioning	Analogy
Alibaba	King of computing power + models	A hybrid of AWS + Meta (open source)
ByteDance	King of applications	Most likely to produce mass-market AI products
Tencent	King of connectivity + implementation	Embedded in China's national-level instant messaging (IM) apps

"At the infrastructure and B-end ecosystem level, Alibaba has a very high probability of coming out on top. But at the level of national-level AI applications, the probability is relatively low."

6.7 Bets on ByteDance + Meta (Post 704, 2025-01)

Two things Lao Huang was most bullish on at the start of 2025:

ByteDance successfully builds China's first phenomenal AI product(likely music / video, not text)
Meta successfully builds the world's first phenomenal AI hardware(glasses, in partnership with Ray-Ban)

6.8 Investment Tools (Lao Huang's current setup)

Primary:Perplexity Max $200(Computer + Personal Computer)
Auxiliary: Grok 4.2 Expert (investment assistant / exclusive X data / straightforward style)
Verification: Codex / Claude Code for Fact-Check
Report template: Market Insights GEM (Post 2135, 2025-12 / full prompt see Full Evolution of the newtype Knowledge System and Life OS)
Betting: Polymarket (Post 2782, 2026-05 Lao Huang "won both Iran war situation 4/30 NO and 6/30 YES")

7. Core Judgments on the AI Era Landscape

7.1 Three-Layer Framework (Post 2360, 2026-02 / highlight / most complete)

Lao Huang presents a three-layer framework for understanding the AI era:

First layer: Technology stack

Bottom layer: Model capabilities (Pre-training → Reasoning → Long Horizon)
Middle layer: Infrastructure (Framework → Harness → Memory System)
Top layer: Application scenarios (Coding → SRE → Research → Vertical)

Second layer: Value chain

Supply side: AI creates new capabilities (Long Horizon Agent)
Transmission layer: New capabilities restructure Workflow (GitHub from repository → dispatch system)
Demand side: New Workflows create new demand ("writing code with AI" becomes the norm)

Third layer: Competitive landscape

General layer: Model vendors (OpenAI / Anthropic) compete for entry points → winner-takes-all
Infrastructure layer: Independent players (LangChain / MemOS) provide transferable capabilities → diversified coexistence
Application layer: Vertical companies (Factory / Rogo) build moats in specific scenarios → rapid iteration

Strategic implications for different roles:

Role	Strategy
Entrepreneurs (bottom layer)	Pursue generality, build technical moats, target infrastructure
Entrepreneurs (middle layer)	Pursue best practices, iterate rapidly, target de facto standards
Entrepreneurs (top layer)	Pursue scenario-specific fit, accumulate memory assets, target domain moats
Investors (bottom layer)	Look at technical moats + possibility of "encirclement"
Investors (middle layer)	Look at engineering capability + ability to turn model capabilities into products
Investors (top layer)	Look at scenario depth + unreplicable memory assets
Individuals (cognition)	Shift from "demand side" → "supply side", from "what do I need" → "what can I create"
Individuals (capability)	Coding becomes the new "literacy"; you don't have to be a programmer, but you need to understand code logic
Individuals (assets)	Your collaborative memory with AI = personal assets, consciously manage and accumulate them
Organizations (strategy)	Judge between "working for AI" vs "being transformed by AI"
Organizations (culture)	Shift from "humans invoking software" → "AI dispatching humans"

7.2 Page Rank → Agent Rank (Post 1317, 2025-05 / highlight)

"If Page Rank becomes Agent Rank, many old tool websites will have to be rebuilt.

After search engines evolve into question-answering engines, the traffic distribution rules of the entire ecosystem will be rewritten.

Over the past decade or two, many tool websites have been consistently recommended by search engines simply because they launched early. For example, sites for checking IP addresses, converting JPG to PNG, etc., are very low-end, completely free, but have earned enormous amounts through ad monetization.

Imagine what will happen next when search engines have fewer display slots and users make fewer clicks? Some tool capabilities will be internalized by models, but there will still be a large number that are not worth internalizing and can be satisfied by calling a small, micro Agent.

This could be an unassuming big opportunity. "

7.3 The End of SaaS Intermediary Monopolies (Post 2373, 2026-02)

"The US SaaS sector has undergone violent turbulence. Those once highly sought-after 'long-termism' targets seem to have lost their halo overnight.

I believe this is not just a market correction, but also a tectonic-plate-level restructuring of business logic. In one sentence, the real narrative behind the SaaS crash is:the end of intermediary monopolies. "

7.4 Service Providers Do Not Face Users Directly (Post 1277, 2025-05)

"Various service providers no longer face users directly, nor do they build websites to showcase themselves; instead, they sell services to AI companies, or provide APIs for AI to access for a fee."

→ AEO (Agent Experience Optimization) is about to replace SEO.

7.5 Whose Account Is Powerful? (Post 253, 2024-05 / classic)

"Accounts follow data. Whoever has more data, and more valuable data, has the more powerful account.WeChat is the endpoint, because WeChat is [identity], while other platforms are merely [business cards].

In the AI era, a large model that is widely accepted and used by most people will emerge. To maximize the value of this large model, users will actively choose to hand over all their data to it.The platforms that own this data will most likely include Google and Apple, followed by upstarts such as OpenAI. The account systems of these old and new giants will become the most foundational accounts, or called [digital ID cards].

Therefore, accounts / identity are the business of giants. Everyone else should stop thinking about it, and think about businesses that giants cannot do and that will not be crushed under their wheels."

7.6 Command-Line Agent = General Agent Skeleton (Post 1461, 2025-07 / highlight)

Why Why are Claude / Gemini both launching command-line AI programming tools?

"Command-line Agent = massive existing tool ecosystem + classic Unix composition philosophy + modern AI orchestration capabilities (such as ReAct)"

So What What future does the command-line Agent point to?

"Command-line Agent + mature MCP ecosystem = general Agent skeleton

Perception: Agent fully understands intent and system state through MCP

Thinking: LLM brain reasoning

Action: Agent executes through CLI interface

Loop verification: Check the file system again through MCP to confirm the task"

7.7 Individual Economy: Dan Koe = Gen Z Naval (Post 2183, 2025-12)

"Dan Koe plays the role of 'Gen Z version of Naval' to a large extent:

Naval = Dao (underlying philosophy, worldview, first principles)

Dan Koe = Shu (techniques/methods)(how to specifically implement and monetize in the creator economy)

Naval's wealth mainly comes from venture capital (technology → entrepreneurship → investment, extremely high barrier to entry)

Dan Koe's wealth comes from content creation, courses and software (writing → design → personal brand → selling knowledge products) — for Gen Z who have no capital but time, this is a more replicable path. "

7.8 Medici Family vs Contemporary Super Individuals (Post 2211, 2026-01 / highlight)

Lao Huang conducted a multi-angle analysis based on Kimi K2 Thinking and concluded:

"The survival model of the Medici family is very inspiring for today's super individuals.

It is not about owning real estate everywhere, but having influence assets everywhere. 'Being able to flee at any time' is not 'being able to book a flight ticket at any time', but 'no matter where I am, there is a value network I have sown in advance.'

Identity should be spectral and context-activated. It is not 'I am a creator and also an investor', but: when writing, the investor identity gives me an economically independent perspective; when investing, the creator identity lets me see through the essence of narratives; when communicating in communities, the superposition of the two identities creates unique value.

What you write is not just articles, but an interface that records how this era thinks. What you build is not just a community, but a stronghold against systemic stupidity.

Coping with the 'sense of drift' does not rely on 'the heart is where home is', but on creating something more important than a place to return to—— a legacy of civilization."

8. 2026 Action Checklist (3 things Lao Huang recommends / Post 2163, 2025-12)

"In 2026, everyone can start with these three things:

Apply for a bank card in Hong Kong, the United States or Europe.

Open a US stock account and start investing in AI-related stocks or the Nasdaq index.

Be able to use the most advanced AI products on Earth anytime, without barriers, on both desktop and mobile. "

9. Things Lao Huang Sees But Cannot Do Himself

Lao Huang repeatedly warns entrepreneurs not to do these things:

GUI Agent — Platform plays (2025-03)
Basic Chatbot — Competing with giants on the "input box" (Post 1229, 2025-05)
Writing a 10,000-word novel and letting AI— Output length bottleneck (2024-08 / see newtype Model Evaluation, Fine-Tuning and Hardware Evolution)
Paid GPT relay stations — "Putting a skin on it to pretend it's official, purely scamming newbies" (Post 1897, 2025-10)
OpenClaw toys — Not counted as productivity before privacy + security are resolved (Post 2367, 2026-02)
Traditional knowledge bases (RAG route) — Cline / Cursor / Claude Code don't need RAG / MCP+ embedding is better (Post 1083, 2025-04)
WeChat groups / private domains — Lao Huang doesn't build groups, can't manage them; discussions are all on Xingqiu (Post 2322, 2026-01)
Training courses on prompt engineering — "The vast majority on the market are useless; it's more useful to learn the logic directly" (Post 1223, 2025-05)

10. Related Wiki

Full Evolution of the newtype Knowledge System and Life OS — Full evolution of the knowledge system (including full Market Insights GEM prompt)
newtype Model Evaluation, Fine-Tuning and Hardware Evolution — Model evaluation (including the three-faction debate on Scaling Law §2.1)
newtype Mindset · Super Individual Methodology — Mindset / Unfair Advantage / Just Do It
newtype · Vibe Coding Independent Product in Practice — Productization Path for Prompt House / Obsidian MCP / LearnFromYouTube / newtype OS
newtype · Agent Architecture and Multi-Agent Orchestration — Commercial Implementation of Agent / Multi-Agent / Workflow
newtype · MCP Protocol and Ecosystem in Practice — Commercial Significance of the MCP Protocol Ecosystem
newtype · AI Coding in Practice and Toolstack Evolution — Commercial Competition Among Cursor → Claude Code → Codex → OpenCode
newtype · AI Toolstack in Practice and Evolution — Toolstack Evolution

Source of this wiki: 452 posts on business / industry / investment topics from the newtype Knowledge Planet community from 2024-03 to 2026-05, fully scanned without sampling. Includes merged BIG NEWS timeline mini-buckets. Focuses on business logic, industry landscape, investment judgment, and the individual economy perspectives; for tool evaluations, see newtype Model Evaluation, Fine-Tuning and Hardware Evolution; for technical details, see Full Evolution of the newtype Knowledge System and Life OS.

2026-06 Supplement: Industry bottleneck hypotheses and operating infrastructure are recorded separately

This section compiles investment and operating writings from the community in June. It records the judgments of the author and members at the time, and does not represent current facts about market conditions, financial reports, or taxation; this content has not been externally verified this time, and does not constitute instructions for buying, selling, or opening accounts.

Industry views must retain their status as “predictions”

The author expanded his focus from prominent computing power leaders to resources required for storage, power, cooling, interconnection, and manufacturing, and stated on the 13th that he would not buy SpaceX in the short term and would prioritize storage. This represents a phased change from the earlier plan on this page of “buying at IPO,” and the old record need not be erased. HBM, NAND, HDD, enterprise storage systems, and CXL are different observation directions listed in the posts; they cannot all be treated as assets with the same cycle and risk.

Figures such as NBIS earnings on June 2, MRVL’s gain, and Micron’s gross margin on the 25th are all self-reported or relayed in the posts, and positions and financial reports have not been checked. The helium supply situation and the impact of geopolitical conflicts on the 20th likewise lack independent verification and are not upgraded to confirmed supply chain facts. The long posts about Micron cooperation, the “three gates” of optical interconnection, the profit migration map, and cryptocurrency only have opening previews available, and the latter halves cannot be reconstructed from the titles.

A counterexample that can be directly retained is the weather Bot: on June 2, the author found that the AI-designed strategy implicitly assumed unlimited capital and time, and even if the theoretical probability was ultimately favorable, it was not applicable to real budgets. Therefore, before “letting AI implement it,” humans must first review the resource assumptions of the strategy; this is a methodological boundary, not proof that the strategy is already profitable.

Companies, payment collection, and personal life arrangements are problems at different layers

On the 16th, the author explained that the original purpose of opening a U.S. company was to collect payments for Prompt House and Substack through Stripe, and that matters such as corporate accounts, mailing addresses, and personal tax IDs came later. He stated that registration in Delaware took two or three days and that obtaining an EIN and related items took about two months; these timelines are not fully consistent with earlier notes and should be regarded as personal recollections, with no promise of processing times for others.

On the same day, he explained that the domestic company is used for commercial income settlement and his own salary and social security arrangements, and emphasized in replies that cross-border corporate payments must be assessed based on the nature of the real transaction. Establishing a company does not make all transfers or investments inherently feasible. Posts about brokerages, banks, cards, and subscription payments can only serve as archives of experience at the time; applicable conditions and fees still require specific verification.

In particular, members’ posts about account opening and taxation on June 7–8 extrapolated from “changing IP” to being unregulated, and from asset scale to a zero probability of taxation, and also involved false statements and underreporting of taxes. These are unsubstantiated and dangerous claims; this library does not adopt them as operational steps or legal conclusions, and the original files are retained for review. A single successful operation does not prove compliance with service rules or filing obligations.

Signals of distribution and commercialization do not equal a proven, working model

Two posts about the Kimi co-branded card described envisioned benefits such as redeeming points for computing power and priority access to models, and comments also mentioned temporary failures of the reservation page; the idea of “connecting consumer benefits with work tools” can be studied, but this does not prove that applications are currently available. When members recommended investment bloggers and brokerage invitation links, those posts were promotional in nature, and their claims of “replicable returns for ordinary people” are not taken as conclusions.

The publishing plan, from submission for review and negotiations to the approval at the review meeting on June 24, reflects stage-by-stage status and cannot be written as the physical book having already been published. The announcement on the 27th that Episode 3 of the course had been updated only proves the release record at that time and does not mean this batch contains the full text of the course.