Calling the AI bubble “burst” because inference costs are falling gets the economics backwards. OpenAI cut pricing on Luna and Terra, its more token-efficient models. It left Sol and Pro pricing unchanged. The lower prices come from models getting cheaper to run, using fewer tokens and serving workloads more efficiently. That opens the door to far more usage. Tasks that were too expensive to automate can now run constantly across coding, customer support, research, data analysis and agent workflows. A company may spend less on each individual request while making thousands more requests across its business. The semiconductor question comes down to total compute consumed. Price per token alone tells you very little. Cloud backlogs are growing, hyperscalers remain short on capacity, and infrastructure spending continues to rise. Model competition will pressure margins at the model layer. Falling inference costs expand the number of applications that can economically use AI, which keeps pushing demand toward chips, memory, networking and data centers. @jvisserlabs