OpenAI, under the leadership of Sam Altman, is not merely advancing artificial intelligence models; it is meticulously shaping their very persona and the infrastructure upon which they operate. Recent revelations from a private dinner with executives and reporters in San Francisco underscore a strategic pivot towards nuanced AI-human interaction and an unprecedented commitment to scaling computational capacity.
During an exclusive dinner in San Francisco, OpenAI CEO Sam Altman spoke with CNBC’s MacKenzie Sigalos and a select group of reporters about the company’s latest product developments and ambitious future plans. The discussion centered on the recent GPT-5 rollout, the burgeoning demand for AI compute, and OpenAI's expanding strategic horizons.
A striking revelation was OpenAI’s deliberate last-minute adjustment to GPT-5’s tone. Just an hour before the dinner, the company pushed an update to make the model’s responses "much warmer." This indicates a conscious effort to imbue AI with more approachable, human-like qualities, moving beyond raw intelligence to prioritize emotional resonance and user experience. Altman further articulated a future vision where users would gain explicit control over the AI's "tone and voice," suggesting a new frontier in personalized AI interaction. This focus on emotional intelligence and user customization marks a significant evolution in AI product design, aiming to foster deeper engagement and utility.
Altman was candid about the significant capacity constraints OpenAI faces. He stated, "We have better models, and we just can't offer them because we don't have capacity."
This shortage, primarily due to being "maxed out" on Microsoft’s Azure, is driving OpenAI to engage with "all the hyperscalers" and prepare for an astronomical investment. Altman disclosed plans to spend "trillions" on data center construction, an eye-popping figure that highlights the immense capital expenditure required to fuel the next generation of AI. This financial commitment underscores the escalating infrastructure race, where access to and control over vast computational resources will dictate the pace of AI advancement and market leadership. The demand for compute, he noted, always outstrips supply.
