Weekly Cheonlian: Week 1 of October

Pokute

Nov 23, 20257m ago

1. AI Market and Investment Trends

+ AI investment scale and data center boom

With the AI craze, data center construction investment is reaching an all-time high. According to U.S. Census Bureau data, annualized spending on data centers is projected to reach $41 billion by July 2025, nearly equal to the total cost of all private office space in the United States. This represents a staggering 2,200% increase since July 2014, and the acceleration of investment is evident following the launch of ChatGPT in late 2022. Investment in the AI infrastructure layer is not over yet, with opportunities across energy optimization, chips, hardware, models, and software.

+ AI market bubble controversy

Warnings continue that the AI industry is in an 'industrial bubble' state.

Amazon founder Jeff Bezos has predicted that while AI is in an industrial bubble, the technology is "real and will transform every industry" and will bring "huge" benefits to society.

He noted that during a bubble, all ideas, good and bad, are funded, making it difficult for investors to distinguish between good and bad. Bezos explained that, just as the biotech bubble of the 1990s ultimately led to the development of life-saving drugs, industrial bubbles can ultimately have positive social consequences.

While prominent investor David Einhorn warned that massive data center investments could lead to "enormous capital destruction," OpenAI CEO Sam Altman took a different view, saying the current pace of investment is insufficient and that the planned multi-billion dollar investment will "look slow later."

The industry predicts three potential "tipping points" related to AI overinvestment.

The limits of scaling laws: Achieving AGI (artificial general intelligence) may take longer than expected, and the incremental value of model improvements may be less than the cost (hundreds of billions of dollars).

Oversupply: Demand currently exceeds supply, but it is uncertain whether demand will smoothly match supply, which will increase 23-fold in 1836 months.

Negative-Margin AI Startups Running Out of Funds: AI model wrapper companies are struggling financially, and VCs are hesitant to invest.

However, hyperscalers such as Google, Amazon, Meta, and Microsoft are considered to be in a "don't catch the tail, don't miss the head" situation where AI spending is discretionary and doesn't significantly impact their balance sheets, and their AI investments are evaluated as rational choices.

2. AI Architecture and Agent Innovation

+ The Future of Agent-Based Computing

It's becoming clear that the industry is moving towards an "agent-first" world.

The roles will shift, with humans focusing on setting goals and intentions, and AI agents handling the 'chores' by executing, reasoning, and even collaborating directly with each other.

Early examples of these agent-based solutions include Clarify, a CRM solution that autonomously classifies emails, creates follow-up drafts, and infers deals, and Dropzone, a cybersecurity company that deploys swarms of agents to detect and respond to threats.

+ The End of RAG and the Contextual Revolution

Retrieval-Augmented Generation (RAG) was a useful band-aid during the "context-poor era," but it's now reaching its end, according to analysis. Early LLMs (GPT-3.5, GPT-4) had limited processing capacity (4K-8K tokens), so the dominant architecture was the RAG, which retrieved the most relevant document chunks from the knowledge base and fed them to the LLM.

However, RAG has fundamental limitations, such as the structural relationships of documents being destroyed during the chunking process, or the embedding failing to understand numbers or concepts.

The recent explosion of context windows (200K for Claude Sonnet 4, 1M for Gemini 2.5, and 2M for Grok 4-fast) is making this RAG architecture obsolete. This has given rise to a new paradigm, "agentic search." Agentic search uses traditional filesystem tools (e.g., Grep) and intelligent navigation to read and infer entire documents, eliminating complex chunking, embedding, and reranking steps. As context becomes richer, search shifts from "retrieval" to "navigation," allowing agents to follow references and fully understand the context, much like human analysts.

+ AI autonomy measurement indicators

"99% step-length" has been proposed as a key metric for measuring the autonomy of AI systems. This refers to the number of actions an AI system can perform continuously with 99% confidence without human intervention.

Current leading systems reliably manage about 100 steps at this threshold, but by 2029, they could exceed 10,000 steps, at which point AI could operate without supervision for weeks or even months. A 99% reliability rate is considered a feasible threshold for autonomous operation, while lower success rates, such as 90%, require continuous human monitoring.

+ Corporate AI adoption strategy

While companies are making significant progress in implementing generative AI into their operations, adoption remains uneven, and the return on investment (ROI) remains unproven. Simply distributing AI across teams isn't enough; a centralized AI team is needed to define a shared roadmap, playbook, and data foundation. Success hinges on deeply integrating AI into the business context and enabling systems, teams, and processes to continuously learn and evolve.

3. The Rise of AI Commerce and Market Restructuring

+ ChatGPT's Instant Checkout and Agentic Commerce

OpenAI ushered in the era of AI commerce with the release of "Instant Checkout," a feature co-developed with Stripe based on the "Agentic Commerce Protocol."

Currently, ChatGPT users in the US can purchase items directly from Etsy sellers within the chat window, and integrations with Shopify merchants (over 1 million, including Glossier and SKIMS) are planned to expand. This protocol is an open AI commerce standard that enables AI agents, individuals, and businesses to collaborate to complete purchases.

When users ask shopping-related questions, ChatGPT presents organic (non-sponsored) product results ranked by relevance. If they match the user's purchase intent, they can complete the purchase with just a few taps, without leaving the chat window. OpenAI charges merchants a small commission, but this rate is estimated at around 2%, significantly lower than traditional platforms (Amazon 15%, Google Shopping 12% in the past).

+ Commerce Economics Analysis and Market Impact

ChatGPT's low commission rate (approximately 2%) appears to be a market-entry pricing strategy, offering a clear value proposition compared to existing competitors. Merchants retain control over payments, systems, and customer relationships.

According to standard economic theory, the economic burden of fees is determined by their relative elasticity, regardless of who legally pays the fee (the seller). At ChatGPT's current low 2% fee rate:

•

Merchant: You will likely absorb 70-90% of the commission burden through margin reduction.

•

Consumers: The burden of price increases will be minimal (5-15%), as a 2% price increase can result in a decrease in profits for most products.

•

Existing competitors: Competitors like Google or Amazon can absorb 5-25% of the burden through market share loss and weakened pricing power. This constitutes burden transfer through competitive displacement.

LLM commerce is structurally different from traditional search-based commerce, and the source of market dominance is shifting from indexing to intelligence. If OpenAI gains a competitive edge through data accumulation and user lock-in in the long term, its commission rate could rise to the 4-6% range. However, if competition intensifies, it could remain below 2%.

4. Generated AI Content and Policies

+ Sora's copyright and profit distribution plan

OpenAI is learning how Sora is used and gathering feedback from copyright holders and other stakeholders.

Key changes include:

Enhanced Control over Character Creation: Similar to the "Likeness" opt-in model, we plan to provide copyright holders with more granular control over character creation. Many copyright holders are excited about new forms of engagement, such as "interactive fan fiction," and want to be able to dictate how their characters are used.

Shared Revenue Model: With more videos being created per user than expected, we're exploring sharing a portion of the revenue generated from video creation with the copyright holders who allow users to create their own characters.

OpenAI anticipates a very rapid pace of change, and plans to experiment with different approaches through Sora and consistently apply successful models to other products.

+ Controversy over the introduction of AI actors and their impact on the labor market

Hollywood is facing a new wave of disruption with the arrival of AI-generated actress Tilly Norwood. Norwood's creator claims she may soon be represented by an agent, championing AI actors as a new artistic tool, akin to CGI or animation.

However, this raises concerns about actors' livelihoods, the deepening of existing inequalities, and the potential silencing of human voices. Actors have criticized the situation, where they are forced to compete on equal terms with AI actors created by synthesizing the faces of countless other actors.

These changes reflect the inevitable trajectory of explosive content demand, cost reductions (elimination of travel, daily expenses, and union restrictions), and advancements in generative media technologies. This could push human actors into "premium niches," occupying areas where emotional depth, spontaneity, live presence, and human vulnerability are valued.

Key challenges include establishing regulations and policies on attribution and consent, labor and compensation, and transparency.

5. Digital Rights and State Control

+ Privacy invasion and censorship controversy

Telegram founder Pavel Durov condemned the French authorities' investigation as "Kafkaesque" and "absurd," warning of efforts across Europe to undermine digital privacy by using the guise of child protection and election integrity to justify surveillance and censorship. He reiterated that Telegram would not grant any government, including French intelligence agencies, access to users' private conversations through a backdoor.

Citing the response to 9/11, where governments exploited fear to expand their surveillance powers, Durov warned that "the cure is worse than the disease." He also expressed concern that Western governments are repeating the same patterns that led to rights violations in authoritarian regimes, and that "suppressing voices that don't align with the narrative is now the norm."

+ Russia's national digital ID system

Russia is piloting a state-linked digital ID feature through the Max app, a messaging platform developed by VKontakte and closely aligned with the government. Max is being built as an all-in-one hub (similar to China's WeChat) that integrates communications, payments, government portals, and commercial services. This feature requires users to connect to the Gosuslugi (National Services Portal) via biometric data or passport uploads, and serves as an official digital ID via a dynamic QR code.

Max's terms allow user data to be shared with government agencies, raising concerns that essential services could be tied to a single, state-controlled app, potentially leaving users with sensitive personal information without meaningful safeguards. This comes at a time when Russia is tightening its online control.

6. Domestic and international business and financial trends

+ Naver-Dunamu Web3 Financial Integration Promotion

News of a merger between Naver, Korea's largest platform company, and Upbit (Dunamu), Korea's leading virtual asset exchange, has sent shockwaves through the ICT industry. Naver aims to accelerate "Web3 finance" amidst the growing mainstream adoption of stablecoins in the global payment market, and needs to secure funding for AI investments and global expansion. Dunamu, which has struggled to expand its business due to regulatory barriers, will gain access to Naver's 30 million user base and payment infrastructure.

+ London's status as a financial hub declines

London's status as a global financial hub has suffered a blow, falling from the top 20 global IPO markets to 23rd place. It was overtaken by Mexico and Singapore, and IPO volume this year plummeted 69% to $248 million, the lowest in 35 years. Third-quarter results further deteriorated, with $42 million, an 85% drop from the same period last year.

+ Expanding the brand experience of AI services

AI search engine Perplexity.ai has demonstrated its connected experience design by opening an offline space, "Cafe Curious," on Dosandaero in Seoul. This marks the first cafe directly operated by an AI company worldwide.

A connected experience is one designed to provide customers with a sense of consistency and continuity across various channels and touchpoints. Perplexity integrates this experience across three dimensions: digital (Comet Browser Agent, free access to Pro services via QR codes), physical (harmony between metal structures and nature, and a spatial presentation inspired by "intellectual exploration"), and human (integrating AI technology into everyday culture and strengthening connections between users). Through this, Perplexity aims to move beyond mere technology to become a culture.

7. New social/political structure

+ Limitations of the network state theory

Noah Smith points out two fundamental problems with Balaji Srinivasan's concept of the "Network State," which posits that online communities evolve into state-like entities providing services like education, insurance, and private dispute resolution.

The problem of public goods provision: Network states find it difficult to provide physical public goods, such as national defense, courts, and physical infrastructure (roads, sewers). If network state members do not pay sufficient taxes to cover these costs and instead use existing state infrastructure, this creates a free-riding problem.

Conflict with established states: Network state members negotiating special "privileges" within their home states can provoke resentment and hostility from neighboring populations, who often question fairness. Network state citizens essentially become "rootless cosmopolitans," which can clash with traditional state structures and ultimately lead to violent ouster.

Therefore, if the network state does not provide any benefits to the traditional state and exists in a parasitic form, it is analyzed that it will have difficulty surviving due to the superiority of organized violence in the existing state form.

References

Funds Flood into AI Data Centers... $41 Billion Annual Investment] (NEW SOURCE: Excerpts from "Funds Flood into AI Data Centers... $41 Billion Annual Investment")

[Durov Slams France Over "Kafkaesque" Telegram Probe](NEW SOURCE: Excerpts from "Durov Slams France Over "Kafkaesque" Telegram Probe")

[Highlights From the 2025 IA Summit](NEW SOURCE: Excerpts from "Highlights From the 2025 IA Summit")

[I was curious, so I asked - Marginal REVOLUTION](NEW SOURCE: Excerpts from "I was curious, so I asked - Marginal REVOLUTION")

[Jeff Bezos: AI in an 'industrial bubble' but society will benefit](NEW SOURCE: Excerpts from "Jeff Bezos: AI in an 'industrial bubble' but society will benefit")

[Network State, or a Network of States? - by Noah Smith](NEW SOURCE: Excerpts from "Network State, or a Network of States? - by Noah Smith")

[Reader Response to "AI Overinvestment"](NEW SOURCE: Excerpts from "Reader Response to "AI Overinvestment"")

[Russia Pilots State-Linked Digital ID Through Max App](NEW SOURCE: Excerpts from "Russia Pilots State-Linked Digital ID Through Max App")

[Sora update #1 - Sam Altman](NEW SOURCE: Excerpts from "Sora update #1 - Sam Altman")

10.

[The RAG Obituary: Killed by Agents, Buried by Context Windows](NEW SOURCE: Excerpts from "The RAG Obituary: Killed by Agents, Buried by Context Windows")

11.

[Tilly Norwood](NEW SOURCE: Excerpts from "Tilly Norwood")

12.

[UK fact of the day - Marginal REVOLUTION](NEW SOURCE: Excerpts from "UK fact of the day - Marginal REVOLUTION")

13.

[Can the Naver-Dunamu Big Deal Break Through the Regulatory Gray Area?] (NEW SOURCE: Excerpts from "Can the Naver-Dunamu Big Deal Break Through the Regulatory Gray Area?")

14.

[Chat GPT, Beyond Shopping to Payments... The Era of AI Commerce Has Arrived] (NEW SOURCE: Excerpts from "Chat GPT, Beyond Shopping to Payments... The Era of AI Commerce Has Arrived")

15.

[Perplexity's Connected Experience Design: A Visit to 'Cafe Curious' | Yojeom IT] (NEW SOURCE: Excerpts from "Perplexity's Connected Experience Design: A Visit to 'Cafe Curious' | Yojeom IT")

16.

[🔮 The only AI curve that matters](NEW SOURCE: Excerpts from "🔮 The only AI curve that matters")

Subscribe to 'Dogma13'

If you subscribe to the site, you can be the first to receive the latest updates, including new posts, via notifications and email.
Join Slashpage and subscribe to 'Dogma13'!