OpenAI in Early Talks for Private Funding at About $1.2 Trillion Valuation

OpenAI is in preliminary discussions with major investors about a new private funding round that would target a valuation of about $1.2 trillion, a sharp increase from the $852 billion valuation at its March 2026 funding. The approach was initiated by investors, and whether the round proceeds and its timing depend on OpenAI's listing plans. In March 2026, OpenAI raised $122 billion at an $852 billion valuation. If the new round is completed, OpenAI's valuation would exceed that of rival Anthropic, which was valued at $965 billion after its May 2026 funding.

OpenAI's annualized revenue surpassed $40 billion in August 2026, up 20% month on month. OpenAI released GPT-5.6 in July 2026 and Astra earlier in September 2026, with market demand recovering accordingly. The company has previously said it has ample funds, but training large models still requires sustained heavy capital investment.

On September 15, 2026 local time, SpaceX Chief Executive Elon Musk said at the All-In Summit in Los Angeles that leading artificial intelligence companies could open their APIs to one another before model releases, allowing competitors to cross-test using their respective safety testing tools to identify potential risks such as biological weapons, nuclear weapons and deception. He said such a peer-review mechanism could start without waiting for new legislation; if identified problems are not resolved in time, competitors could publicly question the safety of a model release. Musk also said that for the mechanism to be truly effective, China should also participate. In August 2025, OpenAI and Anthropic opened their APIs to each other and jointly evaluated each other's released models using their respective safety testing and alignment assessment methods.

OpenAI has held discussions with Anthropic and Google on cooperating on artificial intelligence safety. An OpenAI spokesperson said the discussions began in July 2026, when Google DeepMind Chief Executive Demis Hassabis proposed establishing a US-led "standards body" to set artificial intelligence standards through public-private cooperation or an industry self-regulatory organization. OpenAI Chief Global Affairs Officer Chris Lehane said in September 2026 that the company is promoting "industry-led standards" and is willing to cooperate voluntarily with other companies. The three companies remain competitors in artificial intelligence and have recently faced pressure from artificial intelligence safety and cybersecurity issues.

On September 15, 2026 local time, Google released two real-time audio models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking. The former supports 97 languages, real-time visual understanding and asynchronous tool calls, allowing users to continue a conversation without waiting for a task to finish; the latter adds complex reasoning capabilities and can perform multi-step tasks in the background while maintaining real-time interaction. Google said Gemini 3.8 Live Extended Thinking ranked first on Artificial Analysis's Speech to Speech Quality Index with a score of 82.6, and had a completion rate of 68.6% on the τ-Voice agent task test. Both models are available to developers through the Gemini API and AI Studio, with audio input priced at $0.005 per minute and output at $0.018 per minute.

Google also said audio generated by its artificial intelligence products carries the invisible SynthID watermark.

On September 15, 2026 local time, Micron announced completion of demonstration validation for the world's first 512GB DDR5 RDIMM, which has been tested on multiple server platforms. The module uses TSV vertical stacking technology and reaches a maximum transfer rate of 9,200 MT/s, enabling dual-socket server memory capacity to exceed 12TB; compared with a four-module 128GB configuration, operating power consumption is reduced by more than 60%. Micron said that for memory-intensive workloads, the 512GB RDIMM delivers up to 1.4 times the performance of a 256GB DDR5 configuration in Spark SVM data analytics.