Nvidia has announced a strategic partnership with Marvell Technology, backed by a $2 billion investment, to expand its AI infrastructure ecosystem and accelerate development of next-generation computing systems.

The collaboration connects Marvell’s hardware capabilities to Nvidia’s NVLink Fusion platform, a rack-scale architecture designed to support custom AI infrastructure. The partnership reflects growing demand for scalable systems capable of handling increasingly complex AI workloads.

The move comes as companies across the industry race to build “AI factories,” large-scale computing environments optimized for training and deploying advanced models.

Expanding the NVLink Ecosystem

Under the agreement, Marvell will develop custom XPUs and networking technologies compatible with Nvidia’s NVLink Fusion platform. Nvidia will provide core components including its Vera CPU, NVLink interconnect, Spectrum-X switches, and networking solutions such as ConnectX NICs and BlueField DPUs.

The integration allows customers to design semi-custom AI systems while remaining fully compatible with Nvidia’s broader ecosystem. This approach supports heterogeneous computing environments, where different types of processors and accelerators work together within a unified architecture.

By enabling tighter integration between custom silicon and Nvidia’s infrastructure, the companies aim to provide greater flexibility for enterprises building specialized AI systems.

Focus on Networking and AI at Scale

The partnership also includes joint development of advanced networking technologies, particularly in silicon photonics and optical interconnects. These components are critical for improving data transfer speeds and reducing latency in large-scale AI deployments.

In addition, Nvidia and Marvell plan to collaborate on AI-RAN technology, which applies AI to telecommunications networks, including 5G and future 6G systems. The goal is to transform telecom infrastructure into distributed AI computing platforms.

As AI workloads increasingly rely on distributed systems, high-speed connectivity and efficient data movement are becoming central to performance and cost efficiency.

Strategic Bet on AI Infrastructure

Nvidia’s investment underscores the importance of partnerships in scaling AI infrastructure beyond standalone chips. As demand for inference and model deployment grows, companies are focusing on building integrated systems that combine compute, networking, and storage.

The collaboration with Marvell positions Nvidia to strengthen its role not only as a GPU provider but as a broader infrastructure platform. It also highlights the increasing role of custom silicon and specialized hardware in meeting the needs of enterprise AI applications.

AI & Machine Learning, Cloud & Infrastructure, News, Startups & Investment

Nebius has announced plans to build a large-scale AI data center in Finland, as Europe accelerates efforts to expand computing infrastructure needed for artificial intelligence.

The facility will be located in Lappeenranta and is expected to reach a capacity of up to 310 megawatts, making it one of the largest AI data centers in the region. Nebius said the site is scheduled to begin initial operations in 2027.

The project forms part of the company’s broader strategy to scale global AI infrastructure and meet rising demand for compute resources.

Europe Accelerates AI Infrastructure Buildout

The announcement comes amid a wave of data center investments across Europe, as governments and companies seek to reduce reliance on external infrastructure and support domestic AI development.

French startup Mistral recently secured $830 million in debt financing to operate a data center near Paris, while additional projects have been announced in Sweden and other parts of the region. Meanwhile, companies including Nvidia and institutional investors have backed large-scale AI campuses, highlighting the growing importance of compute capacity.

Nebius, headquartered in the Netherlands and listed in the United States, has positioned itself as a key provider of AI infrastructure in Europe. The company said it is targeting more than 3 gigawatts of contracted power globally by the end of the year, with over 750 megawatts already secured across Europe, the Middle East, and Africa.

Energy and Scaling Challenges

Despite strong momentum, Europe faces structural challenges in building AI infrastructure. Energy costs remain higher than in the United States, and developers often encounter delays related to grid access and permitting.

These constraints have pushed companies to carefully select locations that offer reliable energy supply and supportive regulatory environments. Finland, with its access to renewable energy and established data center ecosystem, has become an attractive destination for such projects.

Nebius is also expanding beyond Europe, with plans for a gigawatt-scale AI data center in Missouri, reflecting a dual strategy of regional expansion and global diversification.

AI & Machine Learning, Cloud & Infrastructure, News

Starcloud has reached a $1.1 billion valuation following its latest funding round, positioning the startup among the fastest to achieve unicorn status after graduating from Y Combinator.

The company raised approximately $170 million in its Series A round, led by Benchmark and EQT Ventures, bringing total funding to around $200 million. The investment reflects growing interest in space-based computing as demand for AI infrastructure accelerates and terrestrial data center expansion faces constraints.

Starcloud is developing orbital data centers designed to process AI workloads in space, a concept that aims to address limitations related to energy, land use, and regulatory barriers on Earth.

Building AI Infrastructure in Orbit

The company launched its first satellite in November 2025, equipped with an Nvidia H100 GPU. According to Starcloud, the satellite successfully trained an AI model in orbit and ran a version of Google’s Gemma, marking an early demonstration of space-based compute capabilities.

A second satellite, Starcloud 2, is scheduled for launch later this year. It will include more advanced hardware, such as Nvidia’s Blackwell chip and an AWS server module, alongside additional computing systems.

The company’s long-term focus is on Starcloud 3, a three-ton spacecraft designed to function as a full-scale orbital data center. With a planned capacity of 200 kilowatts, the system is intended to be deployed using SpaceX’s Starship rocket and could represent the first space-based compute platform capable of competing with terrestrial data centers on cost.

High Risk, High Infrastructure Costs

Starcloud’s business model depends heavily on future reductions in launch costs. CEO Philip Johnston estimates that orbital data centers could become cost-competitive if launch prices fall to around $500 per kilogram.

However, this assumption relies on the successful commercialization of next-generation launch systems such as Starship, which is not expected to operate at scale until the late 2020s. Until then, Starcloud plans to continue deploying smaller satellites using existing rockets.

The company is also addressing significant technical challenges, including power generation, thermal management, and synchronization between distributed computing nodes in orbit. These factors are critical for scaling from small inference tasks to larger workloads such as AI model training.

Emerging Competition in Space Compute

Starcloud is part of a growing group of companies exploring space-based computing, alongside efforts from startups and large technology firms. The concept has gained attention as AI models require increasing amounts of energy and compute capacity.

Despite early progress, the scale of space-based infrastructure remains limited. Only a small number of advanced GPUs have been deployed in orbit, compared to millions used in terrestrial data centers.

At the same time, larger players are exploring similar ambitions. SpaceX has proposed deploying large-scale compute infrastructure in orbit, potentially creating direct competition in the long term.

Starcloud’s rapid rise highlights both the opportunity and uncertainty in this emerging sector. While the company has demonstrated early technical feasibility, the viability of orbital data centers will depend on advances in launch economics, hardware design, and sustained demand for distributed AI computing.

AI & Machine Learning, News, Space & Future Tech

Anthropic has introduced a new “computer use” capability for its Claude Code CLI, enabling AI agents to directly control a user’s computer, interact with applications, and execute tasks through graphical interfaces.

The feature, currently available as a research preview on macOS, allows Claude to open apps, click through interfaces, type inputs, and capture screenshots. It represents a significant step toward fully autonomous AI agents capable of performing real-world tasks beyond text-based interactions.

Computer use is available to users on Pro and Max plans and requires the latest version of Claude Code in an interactive session.

From Terminal to Full System Control

The new capability extends Claude’s functionality beyond traditional command-line operations. Instead of relying solely on APIs or scripts, the AI can now interact with software in the same way a human user would.

This includes building and testing applications, navigating user interfaces, and debugging visual issues. For example, Claude can compile a macOS app, launch it, click through its interface, and verify functionality within a single workflow.

The system prioritizes more precise tools when available, such as APIs or command-line operations, but defaults to computer control when tasks require direct interaction with graphical environments.

This approach enables automation of tasks that previously required manual input, including working with proprietary software, simulators, or tools without programmatic access.

Safeguards and Controlled Access

Anthropic has implemented several safeguards to manage risks associated with granting AI access to a user’s system. Access is controlled on a per-application basis, with users required to approve each app before Claude can interact with it.

The system also enforces a session-based control model, allowing only one active instance to operate the computer at a time. Users can interrupt actions at any moment using keyboard commands, ensuring they retain control.

Certain applications, such as system settings or file managers, trigger additional warnings due to their broader access permissions. The terminal itself is excluded from screenshots to prevent feedback loops or prompt injection risks.

These controls highlight the challenges of balancing autonomy and security as AI agents move closer to operating directly within user environments.

Expanding the Agentic AI Model

The introduction of computer control reflects a broader shift toward agent-based AI systems that can execute complex, multi-step workflows across different tools and environments.

Anthropic’s approach aligns with similar developments across the industry, where companies are building AI agents capable of acting on behalf of users in real time. By enabling direct interaction with operating systems, Claude moves closer to functioning as a general-purpose digital assistant.

The feature also complements other recent updates focused on integrating AI into everyday workflows, including support for messaging platforms and developer tools.

AI & Machine Learning, News