OpenAI faces distribution, compute limits
An internal OpenAI memo highlighted friction with Microsoft while promoting an alliance with Amazon and said Microsoft had 'limited our ability' to reach clients. (cnbc.com) Separately, reporting says AI demand has been metered and some consumer features were cut back to free compute for coding and enterprise workloads as providers prioritize scarce compute. (hindustantimes.com)
OpenAI is pushing harder into Amazon’s cloud as it says its Microsoft tie-up has constrained sales to big corporate customers. (cnbc.com) In a memo sent Sunday, chief revenue officer Denise Dresser told staff that Microsoft had “limited our ability” to meet enterprises “where they are,” including on Amazon Web Services’ Bedrock platform, according to CNBC. She said customer demand since the Amazon partnership was announced at the end of February had been “frankly staggering.” (cnbc.com) The memo came less than two months after Amazon said it would invest up to $50 billion in OpenAI as part of a strategic partnership. Microsoft, which has invested more than $13 billion in OpenAI since 2019, declined to comment to CNBC. (cnbc.com) The dispute is about distribution as much as money. Amazon Web Services sells access to major artificial intelligence models through Bedrock, while OpenAI is trying to win more large-company contracts at a time when enterprise customers often want models inside the cloud stack they already use. (cnbc.com) OpenAI is making that push while computing capacity is tight across the industry. McKinsey wrote in December that artificial intelligence is now the main growth engine for United States data centers, with demand for power capacity projected to rise from about 30 gigawatts in 2025 to more than 90 gigawatts by 2030. (mckinsey.com) That squeeze shows up in the products companies choose to emphasize. OpenAI said on March 5 that GPT-5.4 was built for “professional work,” added state-of-the-art computer use in Codex and the application programming interface, and cut token use versus GPT-5.2 to deliver faster responses at lower cost. (openai.com) OpenAI’s own revenue mix is shifting in the same direction. Dresser told CNBC earlier in April that enterprise now accounts for 40% of OpenAI revenue and is on track to reach parity with consumer revenue by the end of 2026. (cnbc.com) The company is also chasing those contracts against stronger competition in offices and software teams. CNBC reported that Anthropic’s Claude has established itself as a leader in enterprise accounts, while Google is competing aggressively for the same customers. (cnbc.com) Microsoft’s deal with OpenAI still stands, and Azure remains the exclusive host for OpenAI’s core application programming interfaces, according to GeekWire. But the memo shows OpenAI wants more than one road into the market while the industry is short on chips, power, and room to grow. (geekwire.com)
