From OpenAI’s financing talks at a valuation of more than $1.2 trillion to the financing race between Anthropic and DeepSeek, competition for leading model platforms continues. However, the combination of the 10-year U.S. Treasury yield breaking above 5% and the “AI slowdown theory” is raising the valuation threshold for listed technology stocks. Whether the trillion-dollar valuation can be realized still depends on the speed at which AI commercialization translates into revenue and cash flow.
OpenAI is in preliminary talks with investors over a new round of financing, with a proposed valuation of more than $1.2 trillion. According to people familiar with the matter, the talks were initiated by institutional investors, and whether they move forward depends on the timing of a listing. OpenAI CEO Altman said the company will most likely not go public until 2027. This means that before formally accepting public market pricing, investors are still willing to secure more OpenAI equity.
While OpenAI considers a new round of financing, the financing moves of Anthropic and DeepSeek, two other leading AI application players, further reflect capital’s competition for model platforms. Anthropic announced in May that it had completed $65 billion in financing, with a post-money valuation of $965 billion; since then, media reports have said its potential IPO may be prepared at a valuation of about $2 trillion, raising up to about $100 billion. As for DeepSeek, media reported on September 9 that its ongoing financing is being negotiated at a pre-money valuation of about $71 billion, compared with a valuation of about $52 billion in its previous first round of financing.
OpenAI’s recently launched Astra large model has triggered heated discussion of the “AGI topic.” In the OSWorld 2.0 delayed simulation evaluation, Astra scored 72.6% on tasks, with each task taking about 40 minutes; the previous generation GPT-5.6 Sol scored 65.7% and took about 75 minutes. OpenAI’s published Astra results also include 98% on FrontierMath Level 4 and 99.9% on ARC-AGI-3. Nvidia CEO Jensen Huang recently said on social media that the arrival of GPT-6 Astra means “the AGI era has arrived.”
The sudden suspension of new subscriptions to the Pro version triggered by Astra simultaneously reveals the strong growth in commercial demand for AI large models and the constraints on computing power supply. OpenAI product lead Tibo called demand for Astra “unprecedented,” saying the Pro tier places the greatest pressure on the system; starting September 10, the company suspended new subscriptions and upgrades for the $200-per-month Pro 20x package, while existing subscriptions continue to renew normally. Complex agent tasks require repeatedly reading context, calling tools, generating plans, and verifying results. Expanding available computing power and optimizing scheduling are directly related to how much paid demand the platform can handle and at what cost it can complete tasks.
At the same time, as the “AI slowdown theory” sweeps the market and the 10-year U.S. Treasury yield surges, U.S. stocks and even global stock markets have begun to compete over the ability to deliver earnings growth amid the AI boom. On September 12, Anthropic CEO Amodei called for slowing the pace of advancing frontier capabilities so that safety measures can keep up, with Altman and others expressing support. On the first stock market trading day afterward, Nvidia shares fell about 3.4%, and the Philadelphia Semiconductor Index dropped sharply by about 6%. On September 15, the 10-year U.S. Treasury yield once again broke above 5%, touching its highest level since 2007.
Divergence in the capital market is now focused on how quickly AI commercialization can deliver revenue and whether that revenue can support continuously expanding computing power investment. OpenAI is still negotiating financing at a valuation of more than $1.2 trillion, which can be understood as some investors remaining optimistic about the long-term space for advanced models to penetrate subscriptions, enterprise services, and professional workflows.
