Anthropic’s IPO puts artificial intelligence at the heart of a battle between giants
When Anthropic makes its Nasdaq debut, some of the investors buying its shares will also be competing for the company’s customers. Google sells Gemini and provides computing power for Claude, while Microsoft integrates Anthropic’s models into Copilot to compete in document, spreadsheet, and coding tasks. Both companies are participating in the $26 billion private placement that precedes the IPO. Among Claude’s financiers are companies that want to occupy the same market space.
The launch, scheduled for Thursday, 19th, will assign a public value to one of the AI sector’s main bets. Anthropic has moved from the research phase to achieve $17 billion in revenue in the third quarter. Now, the company needs to prove that it can maintain growth while its models face rivals from the United States, China, and Europe—and while the applications that distribute its technology also start choosing between different providers.
The battle for customer control
The next contract may be decided by a small difference for those who only look at the conversation interface. One model may write better code, another may cost less to classify thousands of documents, and a third may be easier to install in a bank’s infrastructure. Those who already control the text editor, the cloud, or the customer relationship have another advantage: the possibility of putting AI in the path of a task that the user already performs daily.
Model and agent: pieces of the same puzzle
Claude, ChatGPT, Gemini, and Copilot appear side by side in consumer comparisons, but the names represent different combinations of technology and distribution. A model processes information and generates results. The application combines this model with resources such as files, memory, and connections to other programs. An agent adds an execution flow, in which the system selects tools, performs steps, and checks the result before moving on.
That is why a model capable of interpreting a spreadsheet may have different results depending on the environment in which it is used. To modify the file, it needs access to the editor and tools to manipulate the cells. To send the final version, it needs a connection to email and authorization. The result depends on the model’s capabilities and how the product organizes these steps.
The dispute has advanced precisely to this set. Claude Code operates within software projects. Codex, from OpenAI, also combines models and tools for programming. Google offers Antigravity, and Microsoft brings Copilot closer to corporate files and applications. Meta entered this market with Muse, focused on personal and professional tasks.
Anthropic and OpenAI need to turn capability into recurring use
Anthropic comes to the IPO with a position built on programming and document work. Its catalog includes models with different combinations of capability, speed, and price. The newly launched Fable 5.2 and Mythos 5.2 target more demanding technical projects, while the company recommends Opus 5.5 as a starting point for most everyday programming and intellectual work tasks.
The customer can pay for a subscription or for the use of the API, which allows incorporating the models into other products. OpenAI follows a similar path with its own broader distribution through ChatGPT. On Tuesday, 17th, it launched GPT-6.2 Astra and GPT-6.2 Sol, focused on professional work, programming, and computer use. These models arrive at ChatGPT Work, Codex, and the API. For both companies, a function performed repeatedly within a team creates a different business opportunity than an occasional question in a chatbot.
This revenue also brings a recurring bill. Each task uses computing, and an agent may make several queries to the model until it completes a request. A system that costs less per query may need more attempts. A more expensive model may compensate if it finishes correctly with fewer interventions. The relevant cost for the customer becomes how much they spend to get a result they can use.
Cloud giants dictate market rules
Google combines Gemini models, productivity applications, cloud, and its own chips. Gemini 4 Carbon, released this month, was introduced for multi-step tasks in programming, research, and digital security. For a company that already uses the group’s infrastructure, the contract can happen within an existing business relationship.
Microsoft uses the same starting point in corporate work. Copilot connects to documents, meetings, emails, and development tools. Its strategy allows working with models from different vendors. Anthropic can earn revenue when Claude is chosen within this environment, while Microsoft maintains the interface, integration, and customer relationship.
Amazon combines its Nova models with the Bedrock platform, which also offers third-party models, and AWS infrastructure. It can sell access to Claude and charge for the capacity that keeps it operational. Alexa+ is another front, aimed at consumer interaction and everyday tasks.
The $26 billion private investment linked to the IPO illustrates these relationships. Google, Nvidia, Microsoft, and other companies have committed to investing in Anthropic in a private placement conditioned on the IPO closing, while at the same time participating in the infrastructure chain or competing in parts of the market where it operates. For Claude’s owner, these partnerships provide capital and distribution, but also leave some customer access in the hands of groups that have alternatives to its product.
Europe and China bring different alternatives
Alibaba has a combination similar to Google and Amazon: it develops models, operates a cloud, and brings tools to a business base. Qwen4-Max, launched this week, works with text and images and has a context window of up to 1 million tokens. The company also promises to release the weights of smaller versions of the Qwen4 family, allowing developers to run them on their own infrastructure.
DeepSeek emphasizes efficiency. Its documentation describes changes in V4.1 Flash to reduce memory and cache storage during execution. Moonshot AI competes in programming and multi-step tasks with Kimi K3.1 Pro. Z.ai combines GLM models with the ZCode programming environment. ByteDance develops the Doubao family and offers models through its Volcengine cloud platform.
French company Mistral combines models, the Le Chat assistant, and tools for businesses. With Mistral Large 4, the company highlights the possibility of deployment under European control. For organizations that need to meet localization and data governance requirements, this choice enters the comparison alongside capability and price.
What does the market expect from the IPO
Anthropic will sell shares for $83 in the indicative range disclosed for the offering. The price determination is scheduled for Wednesday, 18th. Claude’s revenue growth shows that the company has found buyers. The prospectus also exposes long-term commitments to ensure the computing power needed for the coming years.
To sustain this expansion, Claude’s owner will have to continue being chosen both within and outside its own applications. Competitors arrive with different advantages: some already control the software in which the customer works; others offer their own deployment or try to reduce execution costs. The IPO price will incorporate an expectation of how long Anthropic can maintain its position in this set of choices.
Frequently Asked Questions
What is the difference between an AI model and an agent?
The model processes information and generates results. An agent uses models and tools to execute a sequence of actions, such as querying data, modifying a file, and checking the result.
Why might Claude appear within a competing product?
Platforms like Copilot and cloud services can offer third-party models. The model provider is paid for usage, while the platform maintains the interface and integrations with the customer.
What does it mean for a model to have available weights?
It means that the trained parameters can be obtained for execution outside the developer’s service, subject to license and technical requirements. The availability of weights varies between models from the same company.








