The "AI slowdown theory" can't stop the funding frenzy of model giants! Astra is launched, sparking heated discussions on AGI; OpenAI aims for a $1.2 trillion valuation

According to Zhitong Finance APP, ChatGPT's developer—OpenAI, the global leader in AI applications—is in preliminary discussions with potential investors about a new round of financing, with a proposed valuation exceeding $1.2 trillion. Media reports, citing people familiar with the matter, said that the discussions were initiated by institutional investors and that whether to proceed will depend on the timing of the IPO. OpenAI CEO Sam Altman has stated that the company is unlikely to go public before 2027.

This also means that investors are still willing to fight for more OpenAI equity before the company is officially priced by the public market. The business logic supporting this interest is that advanced models are expanding the scope of tasks they can perform, opening up higher growth ceilings for subscriptions, enterprise services, and API revenues. The strength of future valuations will depend on how quickly these demands are converted into income and cash flow.

OpenAI's consideration of a new financing round, possibly valuing it at over $1.2 trillion, alongside the movements of the other leading AI application giants—Anthropic and DeepSeek—further demonstrates the battle for AI model platforms among capital investors. In May, Anthropic announced it raised $6.5 billion, reaching a post-money valuation of $965 billion; subsequent media reports on September 11 said its potential IPO could target a valuation of about $2 trillion, raising as much as $100 billion—if achieved, this would surpass the $86.3 billion record set by SpaceX in June. The $2 trillion figure is more likely to be the potential IPO valuation mentioned in the reports and not an already realized post-funding valuation.

As for DeepSeek, the most cutting-edge AI model developer in China, media reported on September 9 that its ongoing financing is being negotiated at a pre-funding valuation of about $710 billion, up from about $52 billion in its first financing round. Capital from different markets is betting heavily on the growth prospects for AI applications, but the prices under negotiation are still subject to final deal confirmation.

Capital Increases Ahead of IPO: OpenAI Plans New Fundraising, Valuation Could Exceed $1.2 Trillion

OpenAI is in preliminary talks with investors about a new financing round that would push the ChatGPT developer's valuation above $1.2 trillion ahead of its initial public offering.

According to media reports, a person familiar with the discussions, who requested anonymity, said whether to proceed with the financing will depend on when OpenAI ultimately decides to go public. The talks are investor-initiated, the source said.

Such financing would pave the way for the long-anticipated IPO. OpenAI CEO Sam Altman previously told Fortune that the IPO plans are still moving forward, but not within this year. Meanwhile, if this valuation is achieved, OpenAI would surpass its main competitor Anthropic, which completed a funding round in May that brought its post-investment valuation to $965 billion.

The Financial Times reported on Tuesday local time about this financing, noting that this round would allow long-term OpenAI backers to increase their exposure before the company's IPO.

Anthropic is preparing for its own IPO and has chosen Nasdaq as its listing venue. According to Bloomberg, the Claude chatbot developer is seeking to raise funds in its IPO, potentially as early as October, on par with or exceeding SpaceX. SpaceX raised a record $86.3 billion in its offering in June.

OpenAI CEO Altman recently told Fortune that the company will likely not go public until 2027.

From a perspective of technology iteration and unit economics, the core reason OpenAI and Anthropic attract heavy institutional investment and high valuations is their ability to translate model capabilities into repeatable professional work. Pre-training provides fundamental abilities, reinforcement learning and task evaluation improve reasoning and execution, while tool calling, context management, and result verification help models complete longer workflows.

OpenAI's recent high-profile launch of the Astra model, which ignited heated AGI discussions, offers a concrete example. In OpenAI's OSWorld 2.0 latency simulation test, Astra achieved a task score of 72.6%, taking about 40 minutes per single task; the previous generation, GPT-5.6 Sol, had 65.7% and took about 75 minutes. This evidences improved task completion and efficiency under specific evaluation conditions. OpenAI’s Astra also achieved a "nuclear" result in other benchmarks: 98% on FrontierMath Level 4 and 99.9% on ARC-AGI-3, showing breakthroughs in high-difficulty tests. This is why NVIDIA CEO Jensen Huang recently declared on social media that the birth of GPT-6 Astra means "the AGI era has arrived."

Anthropic’s recently released series of multi-agent research systems also demonstrates the engineering value of planning, parallel retrieval, and multi-round tool invocation. If these advances continuously reduce rework and manual intervention in client projects, enterprises will have greater incentive to include AI in their daily budgets, and the commercial space for model companies will expand into software development, research, and other high-value professional services.

The sudden halt of Pro version subscriptions triggered by Astra revealed both the overwhelming business demand growth connected with advanced AI models and the constraints in computing power supply. According to media reports, OpenAI product lead Tibo said Astra’s demand is “unprecedented,” with the Pro tier causing the most system strain. Since September 10, the $200/month Pro 20x plan has paused new subscriptions and upgrades, though existing subscribers continue as normal.

From a technical perspective, complex agent tasks require repeated context reading, tool calls, plan generation, and result verification: input handling adds computational load, long conversations and high concurrency increase KV cache usage, and the generation phase can be limited by GPU memory bandwidth. Thus, expanding available computing resources and optimizing scheduling are directly tied to how much paid demand the platform can handle and at what cost tasks can be completed.

Undoubtedly, the demand pressure brought by Astra strengthens OpenAI’s motivation to expand. The key to supporting a trillion-dollar valuation is continuous increases in successful task delivery efficiency, ultimately turning subscription, API, and enterprise revenue growth into sustainable profit and cash flow.

When Trillion-Dollar Valuations Meet 5% Yields: AI Investment Focuses Even More on “Delivery Capability”

Meanwhile, as the "AI slowdown theory" sweeps the market and the 10-year U.S. Treasury yield surges, both the U.S. and other global markets are starting to focus on AI profitability realization rates under the AI boom. AI investment is becoming increasingly differentiated: capital continues to compete for the long-term growth opportunities of top model platforms, while public markets are raising expectations for the speed at which AI-related, especially software companies, can turn AI into profits.

The “AI slowdown theory”—with global AI leaders like Anthropic and OpenAI over the weekend calling to slow down frontier AI model development, combined with surging yields on 10-year and longer-term U.S. Treasuries—are together raising the bar for technology stock valuations. On September 12, Anthropic CEO Dario Amodei called for slowing the pace of frontier capability advancement to allow for safety measures, which Altman and others have supported.

On the first trading day after Amodei's call for slowing development, September 14, “AI chip superpower” NVIDIA fell about 3.4%, and the global semiconductor barometer, the Philadelphia Semiconductor Index, saw a rare drop of about 6%, reflecting market worries about AI slowdown risks. The market is now re-evaluating the scale of training investments, the pace of model releases, and expectations for future compute procurement. On September 14, the Philadelphia Semiconductor Index fell about 6%; South Korea’s KOSPI, known as a global AI compute investment bellwether, fell over 3% on Monday and then another 0.85% on September 15, closing at 6,627.26, down for the fourth consecutive session.

On September 15, the 10-year US Treasury yield once again broke through 5%, reaching a high not seen since 2007. This "anchor for global asset pricing" rose intraday to 5.012% on September 14, hitting the highest level since 2007, before pulling back to 4.960%. By the close of the US stock market on September 15, the yield had firmly stayed above 5%, closing near 5.04% and continuing to hit new highs since 2007. This long-term risk-free dollar rate is a key reference for global asset pricing, influencing corporate financing costs and the discount rates used to convert future profits to present value. Rising energy prices and inflation pressures are pushing up rates, exposing tech stocks to higher funding costs and adjusted growth expectations.

After the 10-year US Treasury yield hit its highest level since 2007, long-term Treasuries oscillated at high yields and the “AI slowdown theory” began to dampen market expectations for compute expansion and chip company profit growth. However, the surge in AI application frequency and end-user compute demand driven by Astra sent positive signals for AI application commercialization. Against this backdrop, OpenAI is still negotiating financing at a valuation above $1.2 trillion, indicating that some investors remain optimistic about the long-term potential for advanced models to permeate subscriptions, enterprise services, and professional workflows. The divide in capital markets is squarely on how quickly AI can realize revenue—and whether those revenues can support ever-growing investment in compute power.