Meta has launched Muse Glimmer, a compact artificial-intelligence model designed to operate directly on compatible personal computers using a single graphics card.

Unlike popular AI services that process most requests through large data centres, Muse Glimmer can perform certain tasks locally on a Mac or Windows PC. This could give developers and businesses greater control over sensitive information while reducing their dependence on costly cloud-based AI services.

The model was released on Monday as Meta chief executive Mark Zuckerberg called for wider access to powerful AI technology and warned against concentrating control in the hands of a few corporations and governments.

What Can Muse Glimmer Do?

Muse Glimmer is designed primarily for “agentic” tasks—jobs in which an AI system receives an objective and completes several connected actions on the user’s behalf.

Its potential uses include:

• Assisting with computer programming
• Organising files and documents
• Reading and understanding screenshots
• Managing schedules and routine administrative work
• Analysing documents and charts
• Operating software tools through commands

The model is smaller than the most advanced systems developed by Meta, OpenAI, Google and Anthropic. Its principal advantage is not that it is the world’s most powerful AI, but that it can perform useful tasks without requiring an expensive network of data-centre chips.

AI That Can Work Without a Constant Internet Connection

Running an AI model locally could be particularly useful for organisations that handle confidential information.

Hospitals, financial institutions, legal firms and government departments may not want sensitive documents uploaded to an external AI provider’s servers. A locally deployed model allows information to remain within the organisation’s own computer or network.

Local processing can also make some tasks faster because every instruction does not have to travel to a distant server and back. It may additionally allow certain AI features to remain available when an internet connection is unavailable or unreliable.

However, users will still require sufficiently powerful hardware. The phrase “runs on a PC” does not mean that Muse Glimmer will perform smoothly on every ordinary laptop.

Is Muse Glimmer Open-Source?

Meta describes Muse Glimmer as an open-weight model. This means developers can access its trained model weights, run it on their own hardware and customise it for particular applications.

“Open-weight” and “open-source” are sometimes used interchangeably, but they are not necessarily identical. A fully open-source system generally provides broader access to its source code, training methods and associated components.

The new model is expected to appeal particularly to independent developers, startups and businesses seeking an alternative to paying cloud providers for every AI request.

Meta Promises a More Powerful Model

Meta has also said it plans to release the weights of Muse Spark 1.2, described as its most advanced AI model, in the near future.

Zuckerberg argued that wider distribution of AI would encourage innovation and prevent the technology’s benefits from being controlled by only a small number of companies.

The move also signals Meta’s attempt to regain momentum in the open-model market, where Chinese developers such as DeepSeek, Alibaba and Moonshot AI have emerged as major competitors.

Privacy Benefits Come With Security Risks

Downloadable AI models offer greater privacy and customisation, but they also create concerns. Once a powerful model is available for local use, its developer has less control over how it is modified or deployed.

Cybersecurity experts have warned that unrestricted models could be adapted for generating malicious computer code, automating scams or assisting cyberattacks.

Supporters of open-weight AI counter that public access enables independent researchers to find weaknesses, develop safeguards and prevent a few companies from privately controlling the technology.

Meta’s new model therefore enters the market at the centre of a larger debate: should advanced AI be widely distributed, or should access be restricted because of its potential risks?

For ordinary users, Muse Glimmer points towards a significant shift. The next generation of AI assistants may not always live inside a distant data centre—they could operate directly on the computer sitting on a user’s desk.

Microsoft has launched its largest data-centre hub in India, significantly expanding the computing infrastructure needed to run artificial intelligence services, banking platforms and other critical digital operations within the country.

The new India South Central cloud region, located in Hyderabad, is Microsoft’s fourth cloud region in India. The company already operates regions in Pune, Chennai and Mumbai.

With businesses rapidly adopting generative AI, the Hyderabad facility is expected to give Indian companies faster access to Microsoft Azure, AI tools and cloud services while allowing eligible data and workloads to remain within India.

Why Microsoft’s Hyderabad Data Centre Matters

AI systems require enormous computing power to train models, analyse information and respond to users. When cloud infrastructure is located closer to customers, digital services can generally deliver lower latency, improved reliability and better compliance with local data regulations.

The Hyderabad region consists of three separate Availability Zones. These are physically independent data-centre locations within the same region, designed to ensure that services continue operating even if one facility suffers a disruption.

This arrangement is particularly useful for banks, government departments, healthcare providers and other organisations that cannot afford prolonged downtime.

Microsoft said the infrastructure has been designed according to India’s regulatory requirements and the region’s seismic conditions.

HDFC Bank and Adani Group Among Early Users

Several large Indian companies have already joined the facility’s early-access programme.

The initial customers include:

• HDFC Bank
• Adani Group
• Bajaj Finserv
• PB Pay

HDFC Bank plans to use the Hyderabad region as a disaster-recovery location complementing its existing cloud operations in Central India. The arrangement is expected to improve business continuity and keep banking services available during technical disruptions.

Bajaj Finance said the region would help strengthen data residency, resilience and scalability while supporting the company’s cloud and AI adoption.

Zero Water to Be Used for Cooling

Large data centres traditionally require considerable amounts of water to cool servers that generate heat while processing information.

Microsoft says its new Hyderabad region will use enhanced mechanical cooling and air-cooled chillers designed to consume zero water for cooling.

The claim is significant because the rapid construction of AI data centres has raised concerns around the world about their electricity and water requirements.

The company has also signed agreements supporting more than 1,000 megawatts of new solar, wind and hybrid-energy projects in India. More than 630 MW of this capacity is already operational, according to Microsoft.

Part of Microsoft’s $20.5-Billion India Commitment

The Hyderabad facility forms part of Microsoft’s combined $20.5-billion commitment to expand cloud and artificial-intelligence infrastructure in India.

The company announced an investment of $3 billion in January 2025, followed by another $17.5 billion in December that year.

Microsoft now claims the largest hyperscale-cloud presence in India, with four regions of its own and two additional data centres operated in partnership with Jio.

Its global infrastructure includes more than 80 cloud regions and over 500 data centres.

India’s Demand for AI Services Surges

Microsoft says its Azure business in India has registered strong double-digit growth during the past two years.

According to the company, more than 90% of Nifty 100 companies included in its analysis are using Microsoft 365 Copilot. Infosys, TCS, Wipro and LTIMindtree have collectively signed up for more than 400,000 Copilot seats.

The expansion indicates that Indian enterprises are beginning to move beyond small AI experiments and deploy the technology across everyday business operations.

Market researcher IDC expects spending on public-cloud services in India to reach $45.7 billion by 2030, growing at an annual rate of 22.2%. Spending on AI is projected to expand at approximately twice that pace.

Amazon and Google Also Expanding in India

Microsoft is not alone in pursuing India’s growing AI market.

Amazon and Google are also investing heavily in Indian cloud and data-centre capacity. The country offers more than one billion internet users, a large pool of technology professionals and fast-growing demand for digital services.

The competition could give Indian organisations more choice while improving access to high-performance computing. However, the expansion will also increase scrutiny of data centres’ electricity consumption, environmental impact and handling of sensitive user information.

Microsoft’s Hyderabad launch shows how the global AI contest is moving beyond chatbots and software. The next phase will depend on who can build the physical infrastructure—the chips, servers, electricity networks and data centres—needed to keep AI running at scale.