Broadcom is expanding its footprint in artificial intelligence infrastructure through new agreements with Google and Anthropic, underscoring the growing demand for compute power behind generative AI systems. The company said it will develop future versions of Google’s AI chips while also supporting a major expansion of Anthropic’s access to computing capacity. The updates, disclosed in a regulatory filing, pushed Broadcom shares up about 3% in extended trading.

At the center of the announcement is Broadcom’s continued work on Google’s tensor processing units, or TPUs, custom chips designed to train and run AI models at scale. While the companies have collaborated for years, the latest agreement signals a deeper alignment as competition intensifies among chipmakers and cloud providers. Custom silicon is becoming increasingly important as AI companies look for alternatives to general-purpose graphics processing units.

Broadcom is also scaling its relationship with Anthropic, one of the fastest-growing AI startups. The expanded deal will provide the company with access to roughly 3.5 gigawatts of compute capacity, primarily powered by Google’s TPU infrastructure. That marks a sharp increase from earlier deployments. Broadcom CEO Hock Tan recently said the company had already begun supplying around 1 gigawatt of compute to Anthropic, with demand expected to exceed 3 gigawatts by 2027.

Anthropic’s rapid growth helps explain the scale of the investment. The company said its annualized revenue has surpassed $30 billion, up from about $9 billion at the end of last year. It now counts more than 1,000 enterprise customers spending over $1 million annually, a figure that has doubled in just two months. Its Claude chatbot also saw a surge in popularity earlier this year, briefly becoming the most downloaded free app in Apple’s U.S. App Store.

The broader opportunity for Broadcom could be substantial. Analysts at Mizuho estimate the company may generate $21 billion in AI-related revenue from Anthropic in 2026, potentially doubling to $42 billion in 2027. While Broadcom did not disclose financial terms, the projections highlight how central large AI customers are becoming to semiconductor revenue growth.

A Shift Beyond GPUs

The deals also reflect a wider shift in how AI infrastructure is built. For years, companies like Anthropic and OpenAI have relied heavily on Nvidia GPUs accessed through cloud providers such as Amazon, Google, and Microsoft. That model is now evolving.

Broadcom is working with multiple AI developers, including OpenAI, on custom silicon tailored to specific workloads. At the same time, OpenAI has committed to using large volumes of AMD GPUs, signaling a diversification of suppliers. This mix of custom chips and alternative hardware suggests the AI ecosystem is moving toward more specialized and distributed infrastructure strategies.

Scaling the AI Backbone

The expansion of compute capacity into the gigawatt range highlights the industrial scale of modern AI. Training and deploying advanced models now requires vast energy, data center space, and specialized hardware. Much of Anthropic’s new infrastructure is expected to be located in the United States, reflecting both capacity needs and strategic considerations around data and supply chains.

For Broadcom, the partnerships reinforce its transition from a traditional semiconductor supplier into a key enabler of AI platforms. For the industry, they illustrate how the race to build and control AI infrastructure is becoming as critical as the development of the models themselves.

Cursor has introduced Cursor 3, a major update to its AI-powered development platform, reflecting a broader shift toward agent-driven software engineering. The release positions AI agents as central participants in the coding process, moving beyond traditional manual workflows.

The company describes software development as entering a new phase where autonomous agents handle a growing share of code generation and iteration. However, current workflows remain fragmented, with developers managing multiple tools, terminals, and agent interactions. Cursor 3 aims to address this by consolidating these processes into a single interface.

The new workspace is built around agent coordination rather than file-level editing. It allows developers to operate at a higher level of abstraction, focusing on outcomes while still retaining the ability to inspect and modify underlying code when needed.

A key feature is the ability to run multiple agents in parallel. These agents can operate across different environments, including local machines and cloud infrastructure, while remaining visible within a centralized interface. This reduces the need to track separate sessions or switch between tools.

Unified Interface and Workflow Integration

Cursor 3 introduces a multi-repository layout that enables teams and agents to collaborate across projects within the same workspace. The platform also supports seamless transitions between local and cloud environments, allowing developers to move agent tasks depending on performance needs or availability.

For example, long-running processes can be shifted to the cloud, while local environments can be used for testing and iteration. This flexibility is designed to improve efficiency and reduce interruptions during development workflows.

The update also includes improvements to code review and deployment processes. A redesigned diff view simplifies reviewing changes, while integrated tools allow users to stage, commit, and manage pull requests directly within the interface.

Cursor continues to build on its foundation as a standalone development environment, originally derived from a fork of Visual Studio Code. Cursor 3 expands this approach with additional features such as an integrated browser for testing applications, support for plugins through its marketplace, and enhanced navigation tools for exploring codebases.

The launch highlights a broader trend in developer tools, where AI systems are evolving from assistants into active collaborators. As models improve and agent capabilities expand, platforms like Cursor are focusing on orchestration and usability, aiming to make complex multi-agent workflows more accessible.

Cursor said it will continue investing in both its agent infrastructure and traditional IDE features, as it works toward a future where software systems are increasingly built and maintained through coordinated AI-driven processes.

Chinese semiconductor companies are reporting record revenues as demand for artificial intelligence infrastructure accelerates and U.S. export restrictions reshape global supply chains. The combined effect has boosted domestic chip production and strengthened Beijing’s push for technological self-sufficiency.

Semiconductor Manufacturing International Co. (SMIC), China’s largest chipmaker, reported a 16% year-over-year revenue increase to $9.3 billion in 2025, with projections exceeding $11 billion in 2026. Hua Hong also posted record quarterly revenue, reflecting strong demand across multiple chip segments.

The growth is being driven in part by domestic technology firms investing heavily in AI infrastructure. With limited access to advanced U.S. chips due to export controls, Chinese companies are increasingly turning to local suppliers to meet computing needs.

U.S. restrictions, particularly on high-performance GPUs and advanced semiconductor equipment, have accelerated China’s efforts to develop its own chip ecosystem. Analysts describe the restrictions as a catalyst that has intensified demand for domestically produced components across industries including AI, electric vehicles, and data centers.

Companies such as Moore Threads are benefiting from this shift, with the firm projecting more than 200% annual revenue growth as it works to position itself as a local alternative to global GPU leaders.

Memory Shortages and Technology Gaps Persist

In addition to logic chips, Chinese memory manufacturers are seeing significant gains. ChangXin Memory Technologies (CXMT) reported a sharp rise in revenue, driven by global shortages and rising demand for memory used in AI systems and consumer electronics.

High-bandwidth memory, a critical component for AI workloads, remains dominated by global players such as Samsung, SK Hynix, and Micron. However, export restrictions have created opportunities for domestic firms like CXMT to supply the Chinese market, even with older-generation technologies.

Despite strong revenue growth, Chinese semiconductor firms continue to lag behind global leaders in advanced manufacturing capabilities. Companies such as SMIC and Hua Hong are unable to produce cutting-edge chips at scale due to limited access to advanced lithography equipment from suppliers like ASML.

Efforts to build a fully domestic semiconductor supply chain are ongoing but face significant technical and financial challenges. China is attempting to replicate large portions of the global chip ecosystem, a process expected to take years.

While current growth is supported by import substitution and strong domestic demand, analysts warn of potential overcapacity in mature-node chips. Sustained progress will depend on whether Chinese firms can advance into higher-value segments, including next-generation memory and advanced logic chips, which are critical for long-term competitiveness in AI infrastructure.