\n\n
Key developments: The price of running a GPT‑3.5‑level model dropped from $20 per million tokens in late 2022 to just $0.07 by October 2024—a 280‑fold decline, according to Stanford’s 2025 AI Index. This deflation continues today, with the cheapest production API at $0.02 per million input tokens (Llama 3.1 8B Instruct) and flagship‑class DeepSeek V4.1 Flash at $0.15 in/$0.60 out. Anthropic’s Claude Mythos 5.1 also cut its rate by 83% to $10.25 per million tokens.
Why it matters: Lower token costs make high‑quality generative AI affordable for startups and enterprises, accelerating adoption across SaaS, content creation, and internal tooling. At the same time, massive compute investments such as NVIDIA’s Firebird AI factory in Armenia (300 MW capacity by 2027) and Microsoft’s upcoming Maia 300 inference chip signal that scaling infrastructure remains capital‑intensive, shifting competitive advantage toward firms that can combine cheap APIs with private compute.
What to watch next: 1) How cloud providers price next‑gen GPUs and the impact of NVIDIA’s 70,000‑GPU rollout in the CIS. 2) Whether new low‑cost models (e.g., DeepSeek, Llama 3.1) sustain performance parity with premier models, influencing enterprise procurement. 3) Regulatory and sustainability scrutiny of mega‑scale data centers as AI’s electricity demand surges. [[1, 2]]