ai economics

Bilingual

Show HN: Wage Against the Machine – MacWages Index for AI Tasks
2 months ago
- The 'Wage Against the Machine' tool compares AI task purchasing power across countries based on hourly wages.
- Higher hourly wage countries like Switzerland ($55.60/hr) and the USA ($35.10/hr) can purchase more AI tasks per hour worked.
- Lower wage countries such as India ($2.20/hr) and Brazil ($4.30/hr) can purchase fewer AI tasks per hour.
- The benchmark includes multiple AI models (e.g., Opus 5 Max, Luna Max) and evaluates tasks per hour worked.
- The data shows significant disparities in AI purchasing power, with Switzerland leading and India trailing.
Moore's Law for Everything (2021)
2 months ago
- AI will rapidly replace human labor in many fields, leading to massive wealth creation but also potential inequality if policy does not adapt.
- To distribute this wealth fairly, the author proposes taxing capital (companies and land) instead of labor, with a specific plan called the American Equity Fund.
- The American Equity Fund would tax companies 2.5% of market value annually and land 2.5% of value, distributing shares and cash to all adult citizens.
- This system aims to align incentives, reduce poverty, and enable everyone to benefit from capitalism as owners, fostering collective support for growth.
- Implementation challenges include preventing tax avoidance, valuing land accurately, and ensuring the tax does not stifle growth, with gradual adoption to ease transition.
No, it doesn't cost Anthropic $5k per Claude Code user
2 months ago
- Forbes claimed Claude Code Max users cost $5,000 in compute, but this is based on retail API prices, not actual inference costs.
- Actual compute costs are roughly 10% of API prices, as shown by open-weight models on OpenRouter (e.g., Qwen 3.5 at $0.39/MTok input vs Opus $5/MTok).
- Heavy users cost Anthropic ~$500 per month, not $5,000—a $300 loss, not $4,800.
- Most users are break-even or profitable; average Claude Code user consumes $180/month API-equivalent, costing $18 to serve.
- Cursor faces the real $5,000 cost because they pay retail API prices for Anthropic models.
- The narrative that AI inference is a money pit is misleading; it discourages competition and justifies high API markups.
GLM 5.2 and the coming AI margin collapse (part 1)
2 months ago
- DeepSeek's R1 caused market panic but misread AI economics: training is a fixed cost, while inference scales with demand and carries high margins.
- Frontier labs like Anthropic/OpenAI spend heavily on training then profit from inference, with gross margins estimated at ~90% on compute cost.
- GLM 5.2 from Z.ai is a genuine open weights competitor to Opus and GPT, offering similar quality but with weaknesses: slowness, no vision support, and poor web search.
- Switching costs are very low due to OpenAI/Anthropic compatible endpoints, making migration trivial compared to traditional vendor lock-in.
- Cost savings are significant: GLM 5.2 costs ~$4.40/MTok (less than 20% of Opus price), with expected further reductions as serving stacks optimize.
- Data privacy concerns with Z.ai's terms exist, but open weights allow on-premise deployment or alternative providers with better contractual protections.
- The article hints at a coming inference margin collapse that could reshape the AI industry, referencing Bezos's 'your margin is my opportunity'.
Winners and losers in the coming AI margin collapse (part 2)
2 months ago
- The AI market is bifurcating into expensive frontier models and cheap, 'good enough' models for many agentic tasks, with aggressive pricing from competitors like Grok 4.5 and GLM5.2.
- Winners include semiconductor companies and the hardware supply chain, as value shifts from software to hardware due to supply constraints and increased demand from cheaper models.
- Coding agents (e.g., Cursor) benefit from cheap models, enabling profitable operations and access to valuable user data that can train future models.
- End users and consumers are major winners, gaining high-quality AI intelligence at low prices compared to earlier models like GPT4.
- Frontier AI labs may lose revenue if users switch to cheaper models, but they could respond by restricting access to top models via managed platforms to avoid commoditization and protect against distillation.
- The B2C market remains a wildcard, with potential for LLM-adjacent advertising to monetize large user bases like ChatGPT's 1 billion MAUs, shifting focus back from B2B.
- Overall, margins in pure model inference are heading to zero, with value captured by hardware supply chains and users, while frontier labs must maintain innovation leads to avoid commoditization.
Is Open Weight AI Decelerationist?
2 months ago
- Dean Ball of OpenAI claimed open-weight AI models are inherently decelerationist, sparking debate.
- Decelerationism opposes accelerationism, which views rapid tech proliferation as beneficial.
- The economic argument: open weights may deter capital expenditure by compressing future revenues.
- China's ability to subsidize AI infrastructure and overbuild capacity amplifies the threat to profits.
- Current training paradigms are transitory; future AI may be more efficient, reducing upfront capex.
- Value may shift from base tokens to applications, integration, and services, like electricity or Linux.
- Open weights lower barriers to participation, expanding innovation beyond a few labs.
- The article concludes open weights are not inherently decelerationist but a bet on a broader value ecosystem.
Who's Afraid of Chinese Models?
2 months ago
- The author initially believed tech, especially software's zero marginal costs, was fundamentally different, but later realized zero marginal costs underpin Aggregation Theory, leading to centralized value chains.
- AI is reintroducing old business principles like marginal costs, challenging the notion that open-weight models are free; they involve significant COGS (cost of goods sold) for inference.
- Tokens are not fungible commodities; intelligence derived from tokens is fungible. COGS for intelligence depends on factors like model footprint, inference efficiency, and token efficiency.
- Commodity markets operate where all sellers charge the same price based on supply and demand, with profits depending on cost structures; suppliers with higher marginal costs risk bankruptcy.
- Currently, the AI intelligence market isn't fully commoditized due to high demand and compute shortages, giving frontier labs like Anthropic and OpenAI high margins and low costs per unit of intelligence.
- Frontier labs may fear Chinese models due to training cost dominance, the non-perfect commoditization of intelligence (where data from inference improves models), and competition in customer experience integration.
- China's strategy involves commoditizing AI complements through open-weight models to boost its physical world dominance and weaken U.S. frontier labs, leveraging distillation attacks for cost advantages.
- Distillation attacks allow Chinese labs to use frontier models as teachers for rapid improvement, disadvantaging Western open-weight model makers who must follow stricter terms of service.
- Cybersecurity concerns arise as defenders may need to use Chinese models due to restrictions on U.S. models, highlighting the need for policy changes to ensure equal competition and access.
- The author argues against overreacting to Chinese models from an economic perspective but emphasizes cybersecurity risks and calls for U.S. policies that support innovation and defense capabilities.
Open-Source Communism
2 months ago
- The author addresses concerns about open-weight models being 'decelerationist' and threatening American labs' business models.
- Open-weight models are compared to open-source programming languages, where ecosystem value outweighs raw code.
- American labs' success will depend on developer ecosystems, integrations, and tools, not just superior model capabilities.
GLM 5.2 and the coming AI margin collapse
3 months ago
- DeepSeek's R1 model caused a market reaction due to low training costs, but AI economics are more complex.
- Training costs are fixed, while inference costs scale with demand and have high gross margins for frontier labs.
- GLM 5.2 is a strong open-weights competitor to models like Opus and GPT, with near-equivalent quality.
- Drawbacks of GLM 5.2 include slower speed for interactive use, lack of vision support, and poor web search capabilities.
- Migration to open-weights models like GLM 5.2 is easy due to OpenAI/Anthropic-compatible endpoints, with low switching costs.
- Cost savings are significant: GLM 5.2 is priced at ~$4.40/MTok, less than 20% of Opus, and may become even cheaper with optimizations.
- Enterprise concerns about data privacy with providers like Z.ai can be mitigated by using other providers or on-premises hosting.
- Inference margin collapse could reshape the AI industry, benefiting those who leverage lower costs.
AI inference is obviously profitable
3 months ago
- The belief that AI inference is unprofitable due to high costs is disputed, as evidence suggests it can be profitable.
- Rough cost estimates indicate inference costs around $1 per million tokens, with GPT-5.4-mini charging $4.50, implying high profit margins.
- Open LLMs like DeepSeek report over 80% profit margins on inference, with market costs near $0.87 per million tokens, supporting profitability.
- AI labs like OpenAI and Anthropic may use high inference margins to subsidize training costs, explaining their higher prices.
- Even if major AI labs fail, inference providers without training costs can profitably serve models, showing the inference business is sustainable.

ai economics

Show HN: Wage Against the Machine – MacWages Index for AI Tasks

Moore's Law for Everything (2021)

No, it doesn't cost Anthropic $5k per Claude Code user

GLM 5.2 and the coming AI margin collapse (part 1)

Winners and losers in the coming AI margin collapse (part 2)

Is Open Weight AI Decelerationist?

Who's Afraid of Chinese Models?

Open-Source Communism

GLM 5.2 and the coming AI margin collapse

AI inference is obviously profitable