Anthropic has entered into a historic agreement to lease the Colossus 1 supercomputer, equipped with 220,000 NVIDIA GPUs, potentially adding $3-4 billion in annual revenue for the partner. This dramatically shifts the balance in the race for AI leadership.

OpenAI's main competitor has gained unprecedented access to the computational power needed to scale enterprise AI solutions. Simultaneously, OpenAI has released GPT-5.5 Instant with a 24% improvement in accuracy and integration of context from previous chats. The White House is moving towards federal testing of AI models, and major companies are transitioning from model development to consulting and deployment.

Anthropic and the Colossus Supercomputer: A New Era of Scaling

The agreement between Anthropic and the owner of the Colossus 1 supercomputer represents a turning point in the AI industry. Colossus 1 is equipped with 220,000 NVIDIA GPUs, making it one of the most powerful computing complexes in the world. This deal addresses the critical issue of reliability and scalability for the enterprise AI market, where companies require guaranteed access to computational resources.

The potential value of the deal is $3-4 billion in annual revenue, with a profit of $2.5 billion. This is not just a commercial transaction but a strategic move allowing Anthropic to compete with OpenAI at the infrastructure level. Previously, Elon Musk criticized Anthropic, calling the company 'evil,' but business logic prevailed over personal disagreements. Access to such a volume of computational power will enable Anthropic to accelerate the development of new models and provide reliable service to enterprise clients without operational disruptions.

OpenAI GPT-5.5 Instant: A Leap in Accuracy and Contextual Memory

OpenAI has officially promoted GPT-5.5 Instant to the default model for ChatGPT, replacing the previous version 5.3 Instant. The new model demonstrates significant performance improvements: the AIM 2025 math test score increased from 65.4 to 81.2 points, representing a 24% increase in accuracy. This improvement is critical for enterprise applications where errors in calculations or data analysis can lead to financial losses.

The implementation of enhanced context management is particularly significant. GPT-5.5 Instant can now integrate information from previous chats, uploaded files, and even Gmail emails to personalize responses. Initially, this feature is available to Plus and Pro users, but it will soon be rolled out for enterprise plans. This approach allows AI assistants to better understand the user's work context and provide more relevant solutions. This is especially important for companies working with large volumes of data and requiring consistency in information analysis.

Federal AI Testing: A New Standard for Safety

The White House is initiating a federal testing program for AI models, joined by OpenAI, Anthropic, Microsoft, Google DeepMind, and other major companies. Microsoft emphasized in its blog that government testing is a necessary condition for building trust in advanced AI systems. As AI systems become more powerful, the requirements for security checks and testing must increase accordingly.

This initiative reflects a growing understanding that industry self-regulation is insufficient. Federal testing will establish uniform standards for safety, reliability, and ethics for all major AI models. Companies that agree to participate have the opportunity to influence the formation of these standards, giving them a competitive advantage. For enterprise users, this means that AI systems that have passed federal testing will have a higher level of trust and can be used in business-critical applications without additional checks.

Transition from Models to Consulting: A New AI Business Model

OpenAI and Anthropic are initiating a strategic shift in their business approach, moving from exclusive model development to consulting and deployment services. OpenAI has created a joint venture called Deployment Company (Deployco), which is in advanced negotiations for three major acquisitions. The company has attracted $4 billion from 19 major institutional investors, including TPG, Bain Capital, and Brookfield Asset Management.

This shift reflects the understanding that simply creating a powerful AI model is not enough. Companies need services for integrating, configuring, and optimizing these models for their specific needs. Deployment Company will provide engineering and consulting services, helping businesses effectively deploy AI solutions. This opens up a new revenue stream and allows AI companies to integrate more deeply into their clients' operations. For enterprise users, this means they will receive comprehensive solutions from model developers, including support during deployment and optimization.

Expansion of AI Tools into Office Applications and Agent Systems

Anthropic has deployed Claude in all major Microsoft office applications: Excel, Word, PowerPoint, and Outlook. The functionality is available in beta with a full release planned soon. In parallel, Google has released Gemini 3.1 Flash, positioning it as the fastest and most efficient model in the Gemini line. Google is also testing the Gemini Agent mode, adding a separate tab for advanced automation features.

Meta is in the internal testing phase of its new AI assistant Hatch, which is positioned as a competitor to open AI agents. These developments indicate a trend towards integrating AI agents directly into everyday work tools. Instead of switching between separate AI applications, users will be able to work with AI assistants directly in Excel, Word, or other familiar applications. For businesses, this means increased productivity, as employees will be able to automate routine tasks without needing to learn new interfaces. Google AI Studio has also released an editing mode, allowing direct editing of UI components and image replacement.

Что это значит для Казахстана

For Kazakhstan and Central Asia, these developments have direct implications. Companies in the region working in the financial sector, logistics, and manufacturing can gain access to more reliable and accurate AI models through cloud services. The federal AI testing initiated by Western countries will establish global safety standards that will be adopted in Kazakhstan. This means that local companies will be able to use certified AI solutions without concerns about compliance with international requirements. Companies like Alashed IT (it.alashed.kz), specializing in IT outsourcing, will be able to offer their clients integration of advanced AI tools into their business processes. The expansion of Claude into office applications and the development of AI agents create new opportunities for automation in medium and large-sized Kazakh companies. Access to computational resources through cloud services will allow local startups to develop their own AI solutions without the need to invest in expensive infrastructure.

Anthropic has gained access to the Colossus 1 supercomputer with 220,000 NVIDIA GPUs, potentially adding $3-4 billion in annual revenue.

The AI industry is entering a new phase where competition moves from the model level to the infrastructure and service level. Anthropic has gained critical access to computational power, OpenAI has improved the accuracy of its models by 24%, and the entire sector is moving towards federal testing and standardization. Companies are transitioning from model development to consulting and deployment, opening up new opportunities for integrating AI into corporate processes.

Часто задаваемые вопросы

What is the Colossus 1 supercomputer and why is it important for Anthropic?

Colossus 1 is equipped with 220,000 NVIDIA GPUs and is one of the most powerful computing complexes in the world. For Anthropic, this is critical as it allows for the scaling of AI model development and service, competing with OpenAI at the infrastructure level. Access to such power ensures the reliability and speed of data processing necessary for enterprise applications.

By how many percent has the accuracy of GPT-5.5 Instant improved compared to the previous version?

GPT-5.5 Instant showed a 24% increase in accuracy on the AIM 2025 math test: the score increased from 65.4 to 81.2 points. This improvement is critical for enterprise applications where errors can lead to financial losses. The model also gained improved context management, allowing it to integrate information from previous chats and files.

Which companies have joined the federal AI testing program?

OpenAI, Anthropic, Microsoft, Google DeepMind, and other major companies have joined the federal AI testing program. Federal testing will establish uniform standards for safety and reliability for all major AI models, increasing the trust of enterprise users in these systems.

How much money is OpenAI raising for its new consulting division Deployco?

OpenAI is raising $4 billion from 19 major institutional investors, including TPG, Bain Capital, and Brookfield Asset Management. These funds will be used to acquire and develop companies that help businesses deploy AI solutions, reflecting a strategic shift from model development to consulting.

Which office applications have received integration from Anthropic's Claude?

Anthropic has deployed Claude in all major Microsoft office applications: Excel, Word, PowerPoint, and Outlook. The functionality is available in beta with a full release planned soon. This allows users to work with the AI assistant directly in familiar applications without switching between separate tools.

Читайте также

Источники

Фото: Nathan Samani / Unsplash