All articles tagged with #compute capacity
A leaked OpenAI memo portrays Anthropic as a growing rival, but argues OpenAI currently maintains a compute leadership, citing 2025 capacity of about 1.9 GW (with a target near 30 GW by 2030) versus Anthropic’s ~1.4 GW and a 5 GW TPU deal with Google and Broadcom for 2027; the memo follows Anthropic’s Mythos reveal and notes OpenAI’s own cybersecurity AI efforts, as both firms eye IPOs by year-end 2026.
Nvidia plans to invest up to $100 billion in OpenAI and deploy at least 10 GW of systems to support AI model training and inference, highlighting a close partnership driven by the growing demand for AI compute capacity. Analyst Gil Luria views Nvidia as a key investor in this expansion, though concerns remain about OpenAI's heavy resource commitments and Nvidia's role as a financial backer. Despite these challenges, Nvidia's stock is rated a strong buy with a potential 20% upside.
OpenAI CEO Sam Altman stated that the company is constrained by a lack of compute capacity, affecting the frequency of product releases. The complexity of AI models and the need to allocate resources wisely are key challenges. OpenAI, backed by Microsoft, is reportedly collaborating with Broadcom and TSMC to design its own chips. Despite these challenges, OpenAI plans to release new features later this year, though not a new major AI model like GPT-5. The company also addressed recent executive departures, emphasizing its strong and growing team.