Microsoft took the wraps off the third-generation of its in-house AI chip, as part of a plan to reduce its reliance on buying processors from Nvidia.

The Maia 300 chip will be used for the tech giant’s multi-gigawatt deployments in 2027, making it the primary inference system for Microsoft products and services. Reuters reported it plans to potentially deploy more than 1 million of the chips next year.

According to a blog post from EVP Scott Guthrie, the company expects Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems without naming Nvidia.

The Maia 300 is built as a single system, with software, silicon, physical rack infrastructure, networking, and data centres co-designed to operate as a single platform.

Guthrie noted a 96-chip rack offers seven times the rack-scale token throughput of Microsoft’s Maia 200. He said the new chip is “designed with an inference-first architecture optimized for long-context, memory-intensive and decode-heavy AI workloads”.

The Maia 200 chip powers the company’s production workloads, including Microsoft 365 Copilot, as well as OpenAI models running on Microsoft’s Azure cloud computing service.

The post Microsoft unveils Maia 300 custom AI chip appeared first on Mobile World Live.

Indian IT firm HCLTech struck an agreement with Anthropic to establish a separate business unit designed to accelerate enterprise adoption of agentic AI, with a focus on safety, cybersecurity and governance.

HCLTech stated its services team will build industry solutions and offerings powered by Anthropic’s Claude, Cowork and Claude Code, while its HCLSoftware business unit will embed these capabilities into select enterprise products.

The company will further establish a dedicated “Anthropic Business Unit and Centre of Excellence” to accelerate use of Claude in enterprise environments, providing secure sandbox environments to enable development.

Clients will also be aided by the partnership with the adoption of AI-native engineering practices, training workforces and building organisational capabilities to operate agentic AI safety at scale.

As what is known as an Independent Software Vendor (ISV) partner, HCLSoftware will be given access to advanced models. It cited select products across cybersecurity, automation, business process application and application development as services that will benefit from advanced Claude capabilities.

Claude will also be used across IT support and for digital employee services, workflows supported by policy guardrails, exception handling and human oversight for high-impact changes.

CEO and managing director at HCLTech, C Vijayakumar, said few technology companies can combine the scale of a global systems integrator with the product depth of an ISV.

“That gives us a rare advantage to securely engineer and deploy Agentic AI at scale and deliver the business outcomes that matter most to our clients.”

HCLTech’s deal with Anthropic follows an agreement worth $1.1 billion in July with a major unnamed European company to deploy AI services and boost enterprise services for the business.

The post HCLTech taps Anthropic to boost AI enterprise play appeared first on Mobile World Live.

Nvidia is reportedly in talks to either acquire or increase its investment in open-source start-up Reflection AI, as it seeks to strengthen its position in open-weight AI models and challenge lower cost Chinese rivals.

Sources told the Financial Times (FT) discussions remain at an early stage and could lead to a full acquisition, further equity investment or a deal covering chips and computing capacity. An acqui-hire, under which Nvidia would recruit staff and license technology, is another option.

A deal could apparently be reached in the coming weeks, though the terms under discussion remain unclear.

According to the report, Nvidia has already invested $800 million in Reflection AI, which was valued at $25 billion in a March funding round. Founded in 2024 by former Google DeepMind researchers, the start-up develops AI models and tools for software development.

Reflection launched its first open-weight model, Beam, last week, targeting coding and agentic tasks as it looks to compete with Chinese models such as DeepSeek and Kimi.

Open-weight models allow businesses and developers to customise AI systems and run them on their own infrastructure, potentially cutting costs compared with closed services such as OpenAI’s ChatGPT and Anthropic’s Claude.

Although Nvidia has its own Nemotron family of open-weight models, FT said it lacks a frontier-scale model of its own. In an essay published in July, CEO Jensen Huang called for a domestic open-weight “ecosystem” across the US, arguing it was essential for “creating opportunities for innovation and prosperity across the country”. “In fact, openness may be one of the most important paths to AI safety and security,” he argued, adding “relying solely on closed models is not inherently safe” as “they can be breached, misused, or fail in ways that outsiders cannot detect”.

Reflection AI has also secured substantial computing capacity from SpaceX. The deal, struck in June, gives the start-up access to Nvidia’s GB300 chips at SpaceX’s Colossus 2 facility in Memphis, Tennessee, for $150 million per month through 2029.

The post Nvidia weighs Reflection AI acquisition appeared first on Mobile World Live.