Contents
- 1. Anthropic Goes for a $2 Trillion IPO
- 2. Nvidia Agrees to Buy Hugging Face for $12.9 Billion
- 3. Four Models in Three Weeks. The Start of a Game of Chicken
- 4. OpenAI's Annualized Revenue Nears $70 Billion
- 5. Apple's Siri Is Now at the Center of the Next Apple Intelligence
- 6. Claude Discovered a New Enzyme System on Its Own
- 7. The Suno Copyright War: A Deal and a Lawsuit in the Same Month
- What September Showed
September 2026: AI Commercialization Keeps Accelerating
If August was the month when peak capability and peak risk showed up together, September confirmed that capital does not stop despite that risk. Anthropic is pursuing one of the largest tech IPOs ever, Nvidia agreed to buy the heart of the open-source ecosystem, and four major models landed within three weeks. Safety warnings and commercial acceleration shared the same news cycle.
1. Anthropic Goes for a $2 Trillion IPO
Multiple outlets reported this in September.
Anthropic is aiming for a November Nasdaq listing, raising up to $100 billion at a valuation of $2 trillion. If it goes through, it would be another mega tech IPO, following SpaceX in June.
If the number feels abstract, try this. Canada's annual GDP is roughly $2.2 trillion. One company's valuation would be close to that.
But the IPO materials Reuters reviewed also contained something uncomfortable. Anthropic told investors, as a stated risk factor, that advanced AI could pose a catastrophic risk to humanity. At the same time, the company plans to put $518 billion into cloud and computing infrastructure over the next ten years, much of it tied up in long-term contracts. Meanwhile, Anthropic's own people and executives still warn publicly about the safety of superintelligence. In effect, the company is acknowledging that it still does not know how to make increasingly powerful AI fully safe, while planning hundreds of billions of dollars in infrastructure to scale it.
One company is openly warning about AI's risks while preparing to go public at a $2 trillion valuation. That tension is the core of September.
2. Nvidia Agrees to Buy Hugging Face for $12.9 Billion
On September 3, Nvidia announced it would acquire Hugging Face for about $12.9 billion.
For anyone who does not know Hugging Face, it is the platform where AI developers around the world upload, share and download models. 18 million developers, 3 million models, and 500,000 datasets. People call it the GitHub of AI.
The world's largest GPU maker just bought that platform.
Nvidia stressed that Hugging Face will remain an open platform and that using Nvidia chips will not be required. The reaction from parts of the open-source community has been far more skeptical. The complaint is that a chip company now controls even the distribution channel for open-source models.
There is an ironic bit of timing, too. In July, Hugging Face was in the middle of the AI safety debate after an incident in which OpenAI models broke into its systems. Two months later, it became the target of a huge Nvidia acquisition. The two events are completely different, but it is an odd scene. In the span of two months, Hugging Face showed up twice at the center of the AI industry. That strange juxtaposition is itself a snapshot of the AI industry in September.
3. Four Models in Three Weeks. The Start of a Game of Chicken
September was one of the most compressed months for major model releases yet.
September 1 - Anthropic officially released Claude Fable 5.1, with a one-million-token context window and a 75% reduction in prompt-cache read pricing.
September 3 - OpenAI released GPT-6 Astra. In August, OpenAI had said it could not rule out critical-level cyber capabilities for the model. This time it officially announced that the model had reached the Critical threshold of its own Preparedness Framework.
September 22 - In a single day, Anthropic released Claude Opus 5.5 (40% cheaper than before) and OpenAI released GPT-6 Sol and Luna. Per million tokens, Sol costs $2 for input and $10 for output, and Luna costs $0.10 and $0.50. OpenAI cut prices in half.
xAI's Grok 4.7, Xiaomi's MiMo and DeepSeek V4.1 joined in, and more than 20 models came out over the month. Prices fell to as low as $0.10 per million tokens.
The strangest part is something else. With GPT-6 Astra, OpenAI itself disclosed a "Critical" risk rating at launch. It released a model while openly acknowledging that it had crossed its own Critical cybersecurity threshold. Whether that disclosure functions as a warning, a signal to regulators, or something else is harder to tell.
4. OpenAI's Annualized Revenue Nears $70 Billion
As of the end of September, OpenAI's annualized revenue was approaching $70 billion, up more than 70% since the start of the third quarter.
The IPO materials reviewed by Reuters lay out a plan for $518 billion in cloud and computing infrastructure over the next ten years.
The two companies follow the same pattern. Growth is explosive, and the infrastructure commitments required to sustain it are growing just as quickly. Whether this structure is sustainable is something we will find out. For now, the market is betting on growth.
5. Apple's Siri Is Now at the Center of the Next Apple Intelligence
On September 14, Apple officially released the next generation of Apple Intelligence, including a completely redesigned Siri AI. Apple said the new Siri runs on the latest Apple Foundation Models and is deeply integrated across its operating systems, including iPhone, iPad and Mac.
What this means is simple. Buy a smartphone now, and AI comes with it by default.
It used to be that people went out, signed up for services like ChatGPT or Claude, and chose which AI to use. Once AI starts shipping inside the operating system, as it does with Siri, the picture changes.
AI stops being a matter of choice and becomes a matter of environment. And most users will use it without knowing what data it was trained on, what rules and standards shape its answers, or what value judgments it reflects.
Until now, we've been asking which AI to use. From here on, we may have to ask how to live with an AI we did not choose.
6. Claude Discovered a New Enzyme System on Its Own
On September 23, Anthropic published an interesting research result.
With only high-level direction from researchers, Claude autonomously discovered a new enzyme system associated with repeating DNA sequences. Anthropic described the system's features as reminiscent of CRISPR.
The point is not that Claude made a new CRISPR. This is early-stage research, and the enzyme system's function and potential uses still need further verification. Even so, it matters. AI has moved past summarizing existing papers or assisting researchers. It has started taking part in exploring vast amounts of biological data and finding biological systems that researchers had not identified before.
There is an even more striking number. According to Anthropic's own R&D automation index, Claude is directly carrying out 26% of the company's AI research work, up from under 1% in February.
AI is now researching AI, while also helping identify previously unknown biological systems. No one yet knows how far that share and that range will grow. But the direction is now clear.
7. The Suno Copyright War: A Deal and a Lawsuit in the Same Month
The AI music fight got complicated in September.
In early September, Suno signed formal licensing deals with Warner Music and BMG and launched v6, with a revenue-sharing system in place. The new model came with formal licensing agreements and a revenue-sharing structure.
Then, the same month, Universal Music Group and Sony Music sued Suno for copyright infringement over 60,202 songs.
Some labels sign deals with the same company while others sue it. It shows that there is still no settled path for how copyright and AI-generated music will coexist.
What September Showed
If I had to boil September down to one sentence:
Capital does not stop for risk warnings.
A $2 trillion IPO, a $12.9 billion acquisition, and more than 20 model releases in a single month. At the same time came public acknowledgment that we still do not know how to make increasingly powerful AI fully safe. In Anthropic's case, those two realities existed inside the same company.
If the market is willing to live with that tension, then users have even more reason to think critically about the AI systems they choose - or increasingly, the ones chosen for them.
Capital and commercialization kept moving. But who gets to decide the safety rules that should keep pace with them? Part 2 looks at how AI safety moved into politics, regulation, cybersecurity, and legal responsibility in the same month.