OpenAI is weighing 40 to 60 percent price cuts on its GPT-4.5 API to match Anthropic's aggressive pricing, according to a report in The Information published June 10. The cuts would be the largest in OpenAI's history and would compress margins at the company that pioneered commercial AI inference. The AI price war just became a margin war. [1]

The pricing pressure originates with Anthropic. The company slashed Opus pricing by 67 percent with the 4.6 release in November 2025, bringing its per-token costs close to OpenAI's. Claude Fable 5, released in June, priced input tokens at $10 per million and output tokens at $50 per million — less than half what Anthropic charged for comparable models a year ago. The pricing trajectory is clear: AI inference is becoming a commodity. [2]

The question is whether this is a race to sustainability or a race to zero. OpenAI's current pricing generates significant revenue but insufficient margin to cover its infrastructure costs. The company spends billions on GPU clusters, data centers, and energy. Lower prices mean lower revenue per token, which means higher volume is required to maintain the same revenue. The math works only if demand scales faster than prices fall. [3]

X's frame treats the price war as a race to zero. Users noted that OpenAI and Anthropic are burning cash at current pricing, and a true price war ends only when one company cannot sustain the losses. The Chinese competitors — DeepSeek and Qwen — are already operating at a fraction of the cost. A reliable Shenzhen source reported that Chinese models will halve their prices again by year's end using newer Huawei GPUs. The US companies are not just competing with each other. They are competing with a cost structure they cannot match. [4]

The structural problem is that AI inference has no natural floor. Unlike manufacturing, where raw materials and labor create a minimum cost, AI inference is electricity and compute — both of which are getting cheaper. The price trajectory is downward with no obvious bottom. OpenAI's 40 to 60 percent cuts would bring GPT-4.5 pricing close to what open-source models charge. The question is whether proprietary models can survive at open-source prices. [5]

MSM coverage treated the cuts as competitive dynamics. The Information focused on the internal deliberations — which models would be affected, what the timeline might be, how Microsoft would react. The Wall Street Journal framed it as a response to competitive pressure. Neither outlet addressed the structural question: what happens to the AI industry when inference becomes a commodity. [6]

The Microsoft relationship adds complexity. Azure AI revenue depends on OpenAI's pricing. If OpenAI cuts API prices by 60 percent, Azure's revenue per token drops by the same amount. Microsoft's AI strategy — built on the assumption that proprietary models command premium pricing — would need to be recalculated. The price war is not just an OpenAI problem. It is a Microsoft problem. [7]

X users pointed to the broader implication. AI companies have raised hundreds of billions of dollars on the assumption that proprietary models would command premium pricing for years. A price war undermines that assumption. If inference becomes a commodity, the moat is not the model — it is the distribution, the integration, and the data. OpenAI has distribution. Anthropic has integration. Neither has a durable cost advantage. [8]

The broader pattern is the commoditization of AI. Every technology platform goes through the same cycle: innovation, premium pricing, competition, price compression, commoditization. AI is entering the compression phase. The companies that survive will be those with the deepest pockets, the broadest distribution, or the most differentiated products. OpenAI has distribution. Anthropic has differentiation. The question is which matters more when prices fall. [9]

The paper's position is that the margin compression is the story. AI companies built their valuations on the assumption that inference would remain expensive. The price war proves that assumption wrong. The gap between valuation and margin is the risk the market has not yet priced. [10]

-- THEO KAPLAN, San Francisco