OpenAI is redirecting resources from experimental research to ChatGPT development, triggering senior staff departures as the $500 billion company faces intense competition from Google and Anthropic.
The San Francisco-based AI firm has reallocated computing power and funding away from long-term projects to focus on improving the large language models that power its flagship chatbot, according to 10 current and former employees cited by the Financial Times.
Three senior researchers left OpenAI in recent months over the strategic shift. Vice-president of research Jerry Tworek departed in January after seven years, model policy lead Andrea Vallone joined rival Anthropic last month, and economist Tom Cunningham exited last year.
The departures signal a fundamental transformation for OpenAI, which launched ChatGPT as a research preview in 2022 before sparking the generative AI boom.
Under CEO Sam Altman's leadership, the company is evolving from a research laboratory into one of Silicon Valley's major corporations.
This transition brings pressure to generate revenue that justifies OpenAI's $500 billion valuation to investors. The company must demonstrate commercial returns while competing against Google's Gemini 3 model and Anthropic's Claude, both of which have made significant technical advances.
In December, Altman declared an internal "code red" following Google's release of Gemini 3, which outperformed OpenAI's systems on independent benchmarks. Staff were instructed to prioritize rapid ChatGPT improvements, with some reassigned from autonomous agents, advertising, and e-commerce projects.
Teams working on non-language model projects reportedly struggle to secure computing resources. Video and image generation models Sora and DALL-E feel neglected compared with ChatGPT work, according to sources close to the company.
Multiple non-language model initiatives have been wound down over the past year.
Tworek's departure followed a standoff over research direction, with his requests for additional computing power and staff reportedly denied by leadership. He wanted to pursue continuous learning research that allows models to adapt without losing prior knowledge.
Vallone faced what sources described as an "impossible" task of safeguarding ChatGPT users' mental health before joining Anthropic. Cunningham reportedly left concerned that OpenAI was moving away from impartial research to focus on company-driven projects.
OpenAI's chief research officer Mark Chen rejected the characterization that long-term work is being sidelined.
The company now treats language model development as an engineering challenge focused on scaling compute, algorithms, and data, according to one person familiar with its research culture. While this approach delivers rapid gains, it makes original "blue-sky" research harder to pursue.
Former employees describe the AI landscape as a "crazy, cut-throat race" where companies spend unbelievable amounts to maintain quarterly advantages. OpenAI's 800 million ChatGPT users provide a strategic moat, but competitors with more advanced models leave little room for slowdown.
Jenny Xiao, a partner at Leonis Capital and former OpenAI researcher, believes the company's advantage has shifted.
"The moat has moved from research to user behavior," Xiao said. "That's a much stickier advantage than having the best model."
The reorganization reflects broader industry pressures as AI companies balance foundational research against commercial imperatives. OpenAI's pivot from blue-sky experimentation to product-driven engineering marks a new phase in its evolution from startup to technology giant, even as the company continues to launch new products like the Codex desktop app for Mac.
