Dylan Patel

SemiAnalysis launches InferenceMAX, an open-source automated benchmark that continuously re-benchmarks inference frameworks and hardware in real-time to address rapid software evolution that makes static benchmarks obsolete.

“Co-author of the InferenceMAX benchmarking article”

“Co-author of SemiAnalysis's GDPx benchmark publication evaluating frontier AI models across high-value work domains.”

“Speaker at Fireworks Forge event.”

“SemiAnalysis analyst forecasting 50GW of new data center capacity in 2027 with chips and memory as bottlenecks.”

“Co-authored systematic analysis of 857 Chinese AI model releases and safety evaluation practices by nine leading developers.”

“Co-author of SemiAnalysis subscription value analysis for AI models.”

“Co-authored comprehensive report on Chinese datacenter infrastructure and AI capacity expansion.”

“Invested in Positron AI's Series C and joined its board, noting the company's memory-first architecture addresses real inference constraints.”

“Co-authored article on South Korea's sovereign AI buildout and infrastructure investment.”

“Discussed lab compute economics, projecting that OpenAI and Anthropic could control most usable AI compute by 2028 due to superior monetization of FLOPs.”

“SemiAnalysis founder theorizing that Mythos 2 was used to train Mythos 3 based on model release rumors.”

“SemiAnalysis analyst co-authoring detailed analysis of reinforcement learning bottlenecks.”

“Forecasts Claude hitting 20%+ of GitHub commits by end of year and discussed energy concerns for datacenters.”

“Author analyzing Anthropic's revenue growth versus compute scaling and implications for AI token economics and frontier model access.”