Gavin Baker shares insights on the current state of AI, emphasizing the significance of scaling laws demonstrated by Gemini 3. He views this as a key data point, suggesting that upcoming Blackwell models could markedly improve performance when released in 2026. Baker critiques GPT-5, noting it was designed for cost efficiency rather than performance, indicating it doesn't contradict the scaling laws that remain intact. He argues that advancements in reasoning are reshaping the economics of frontier models, allowing for a feedback loop where user-generated data enhances product quality and user acquisition. The competitive landscape is tightening, with four major players — Gemini, OpenAI, Anthropic, and xAI — controlling advanced AI models. Baker points out that Meta has a slim chance to catch up with Chinese open-source models, which lag behind. He believes Blackwell will widen the gap between American and Chinese technologies, particularly as the U.S. invests in domestic semiconductor production. The impact of power shortages on AI infrastructure is also a concern. Baker suggests these shortages may actually benefit leading firms like Blackwell, as efficiency in tokens produced per watt will become the decisive factor in revenue generation. Baker highlights the importance of return on invested capital (ROIC) for hyperscalers, which has remained strong despite significant capital expenditures on GPUs. He notes the first signs of AI's impact on financial performance from multiple S&P 500 companies, indicating a shift towards AI-driven productivity. Yet, he acknowledges growing anxiety around OpenAI, which has lost market share and key personnel, placing it behind competitors in model quality. Despite these jitters, Baker believes overall demand for AI tokens will remain robust, driven by customer ROI rather than the fortunes of any single company.