The source material (the newsletter ‘AINews’) presents that a major paradigm shift is underway in the AI industry in terms of business structure and economic impact.

Specifically, it maps out the current state and future outlook of the AI economy from the following 6 key perspectives.

1. Ultra-rapid corporate growth and radical change in the “data vendor/model evaluation market”

In the AI market, revenue growth and valuation increases are progressing at speeds far surpassing those of traditional SaaS and IT companies.

• Mercor’s record-breaking scale: Data vendor Mercor expanded its annualized revenue (run-rate revenue) from $1 million ($1M) to $3 billion ($3B) in just 27 months and raised $750 million in Series D financing at a valuation of $20 billion ($20B). This is the fastest pace ever recorded for reaching $3B, and major tech companies such as NVIDIA, Salesforce, and Zoom have joined the ranks of investors alongside Benchmark and General Catalyst.
• Enormous total addressable market (TAM): Companies such as Mercor organize domain experts (at a scale of 5 million) and provide advanced AI model evaluations and training data. There are more than 800 occupations in the United States alone, and model evaluation and optimization within the workflows of each occupation remain largely unexplored, leaving vast room for growth.

2. Structural impact on the macroeconomy: The arrival of an era in which “inference costs > labor costs”

Among the particularly noteworthy economic indicators in the source is the “surge in AI inference costs and the reversal of labor productivity”.

• Structural shift in which inference budgets exceed salaries: In Mercor’s case, an amount equivalent to “three times employees’ salaries” is already being invested in AI inference (Inference) costs. This enormous inference cost can “fully pay for itself (Pay for itself)” by dramatically improving the team’s operational efficiency and productivity.
• Ripple effects on GDP growth: If a ratio such as “concentrated investment in inference generates returns greater than labor costs” spreads across the broader economy, analysis suggests that macroeconomic growth over the next five years could produce double-digit annualized GDP growth at roughly three times the conventional rate.

3.Vast Revenue and IPOs from Frontier AI Companies, and the Race for Infrastructure Investment

The business scale of leading companies developing frontier models (such as Anthropic and OpenAI) has reached levels comparable to those of major IT giants.

• Anthropic's Mega-Scale IPO: Anthropic has begun an IPO roadshow anticipating a valuation of approximately $1.24 trillion to $1.38 trillion ($1.24–1.38T). As of the end of October 2026, its run-rate revenue stands at $81 billion, while revenue for Q3 alone reached $17 billion.
• OpenAI's Revenue Scale: OpenAI's Q3 revenue also reached approximately $10 billion, as U.S. frontier companies amass vast revenue and capital while competing in infrastructure investment.

4. Token Economics (Tokenomics) and Cost-Optimization Trends

As frontier models become more powerful, rigorous comparative evaluations of “compute costs (API fees)” and “cost-effectiveness (ROI)”have become a major business focus.

• Polarization Between High-Cost Models and Cost-Efficient Models:
  • Google's Gemini 4 Carbon can handle advanced development tasks and, according to Perplexity's evaluation, reportedly achieves results equivalent to competing Claude Fable 5.2 at a per-task price of $5.86 and at 40% of the cost. It also ranks first on the Vals Index for cost-effectiveness in finance, law, and tax.
  • Meanwhile, Anthropic's Claude Fable 5.2 demonstrates high performance but consumes an extremely large number of tokens, resulting in a high-cost structure: it uses 4.3 times as many output tokens as the competing model (Astra) and costs 7.5 times as much as Gemini 4 Carbon. Indeed, one example of game-prototype development using Fable 5.2 reportedly required 8.5 billion tokens (API costs of approximately $15,000).
• Cost Reduction through “Decision Models”:
  • The use of task-specific decision models, rather than invoking a fully capable general-purpose LLM every time, is gaining traction. For example, Amazon switched its internal ticket-labeling process from Sonnet to a decision model, maintaining equivalent accuracy while achieving a 4.5-fold speedup and a 44-fold cost reduction.

5. The Economic Rationale for Hybrid Operations and Local/Edge Inference

Against the soaring cost of cloud AI, a hybrid economic ecosystem combining local environments and small models is rapidly taking shape.

• 80% Cost Reduction through Cloud Instructions × Local Execution:
  • A multi-agent operating method has been reported that delegates only planning and reviews to the cloud’s top-tier model (Fable 5.2), while distributing actual code generation and task execution to local open models (such as three parallel instances of Qwen 3.8 27B). This has successfully cut development time in half compared with cloud-only operation, while reducing API costs to approximately one-fifth ($15.77 → $2.98).
• Advances in Memory-Efficient, Low-Cost Technologies:
  • With the Ternary (3-bit) quantization technique (TTF), the memory required for training and conversion is being reduced from the conventional 16 bytes per weight to less than 1 byte, making it increasingly possible to train and fine-tune 27B-class models on general-purpose consumer GPUs (such as a 24GB RTX 3090).
  • The economic incentives for local inference in on-premises and private environments are growing, as demonstrated by a 40-unit cluster of inexpensive mini PCs costing around $5,000 each (running GLM 5.3 Flash) and an app called “Harbor” that pools unified memory through Apple’s Thunderbolt 5 connections.

6. Enterprise Integration and Ecosystem Monetization

AI vendors are shifting their monetization strategies from simply providing chat interfaces to deep integration into enterprises’ day-to-day workflows.

• Platformization and B2B Authentication: Through features such as “Businesses in ChatGPT” and “Sign in with ChatGPT,” the verification of company profiles and integration with external tools such as Notion and v0 are advancing.
• Vertical Integration of Productivity Tools: Claude Whiteboards, Sheets, Data, and the MCP (Model Context Protocol) standard are driving the development of enterprise platforms that support everything from data analysis to meeting participation.
• Spillover into Hardware and the Supply Chain: In the semiconductor sector, alongside demand for NVIDIA GPUs, massive capital investment continues at the infrastructure foundation layer, including the race to develop next-generation memory such as “HBM5,” which doubles bandwidth.

Summary

The core of the AI industry’s business and economics, as shown by the sources, is a shift from “competition over the performance of individual models” to “optimization of return on investment (ROI) that takes inference and operating costs into account”.

• On the macroeconomic front, inference investment is expanding to levels that exceed labor costs, driving a productivity revolution significant enough to suggest the possibility of double-digit GDP growth.
• On the microeconomic and corporate-management front, the enormous costs of cloud frontier models ($15,000 per project and high API unit prices) have made cost-optimization technologies and structures—such as specialization in decision-making models, hybrid parallel operation with local models, and memory-efficient training important factors in determining a business’s success or failure.