In Sweden, a consortium consisting of Ericsson, AstraZeneca, Saab, and SEB recently announced partnerships with Nvidia to construct an AI factory driven by the latest Grace Blackwell GB 300 systems. Within Europe, the EU's AI Factory initiative, supported by the EuroHPC Joint Undertaking, is establishing a network of at least 15 AI-optimized supercomputing hubs and several AI Gigafactories to assist startups, small and medium-sized enterprises (SMEs), and research institutions across fields such as healthcare, climate, and manufacturing throughout Europe. In the U.S., Nvidia has declared its intention to domestically produce AI supercomputers for the first time, collaborating with partners like TSMC, Wistron, Amkor, and SPIL to generate up to $500 billion worth of AI infrastructure over the next four years. Likewise, in Asia, both the private and public sectors are investing in constructing AI factories, focusing on business innovation and training AI talents.

The era of AI factories has arrived, bringing along economic opportunities and enhanced productivity but also safety risks and energy challenges, along with the diminishing of tangible human connections.

Why Enterprises Require AI Factories

Enterprises across various industries are incorporating AI into their research and development processes. At Pfizer, AI and supercomputing played crucial roles in the creation of PAXLOVID, the company’s COVID-19 oral treatment. AI expedited molecule discovery, halved the time required for clinical data analysis, and cut production cycle time by two-thirds. In collaboration with Nvidia and Classiq, BMW is investigating how quantum-AI synergies can enhance vehicle engineering and manufacturing logistics. Similarly, Goldman Sachs employs LLMs to simplify legal and compliance operations, while Dell’s enterprise clients are utilizing agentic AI to deploy digital workers across cloud and edge environments. Goldman Sachs has introduced its GS AI Assistant to over 10,000 employees—bankers, traders, and asset managers alike. Engineered to progress from basic task automation (such as proofreading and code translation) to agentic behavior, the system is trained to mimic a seasoned Goldman employee—consulting relevant data, applying appropriate algorithms, and reflecting the firm’s values and decision-making protocols.

As computing becomes embedded into daily workflows, drafting documents, generating code, performing statistical modeling, even designing new products, enterprise demand for compute is surging. McKinsey has reported that over the next three years, 92 percent of companies intend to boost their AI investments. To keep pace, organizations require not just more GPUs, but entire supply chains that provide compute at scale, speed, and security.

This is where AI factories come into play. Unlike conventional data centers or chip fabrication facilities, AI factories combine three critical layers: chip production and packaging, supercomputer system assembly, and data center-scale deployment. Dell Technologies, for instance, in partnership with Nvidia, has released Dell AI Factory 2.0, an end-to-end enterprise platform tailored for agentic AI, facilitating federated data access, semantic search, and feedback loops optimized for large-scale digital workforces.

By consolidating these functions, AI factories eliminate delays between design, construction, and deployment. They grant companies greater direct control over how their models are trained, where their data is stored, and how resilient their infrastructure is against global disruptions.

AI Factory Redefined

Nvidia now offers a fully integrated AI factory platform aimed at accelerating this expansion. It encompasses not only hardware components but a complete software and orchestration stack. The TensorRT ecosystem supports high-performance AI inference, while Nvidia Dynamo and NIM microservices streamline deployment and model optimization. Data Flywheel facilitates continuous customization based on real-world feedback, ensuring that AI systems improve over time.

Nvidia’s full-stack platform enables AI factories to function akin to industrial assembly lines for intelligence. Raw materials consist of foundation models, customer data, and toolkits. These inputs are molded by high-performance training, fine-tuning, and inference, while ongoing feedback loops allow systems to adapt and enhance in real-time.

The Nvidia Blackwell Ultra-based GB300 NVL72 solution provides 50 times higher output for AI reasoning than previous iterations. Nvidia’s DGX SuperPOD integrates this computational power into a ready-to-use system, while DGX Cloud delivers the same AI infrastructure virtually via major cloud providers. Both options empower enterprises to scale rapidly without sacrificing performance.

To harmonize all these elements, Nvidia’s Omniverse Blueprint provides a means to simulate entire factories as digital twins, optimizing design, layout, and workflows in a virtual environment.

Transforming the Nature of Work

Unlike traditional manufacturing plants, AI factories do not promise mass employment on the factory floor. Instead, they herald a transition towards high-skill, cross-disciplinary roles including: 1) digital twin architects who model industrial workflows in simulations; 2) AI-operations engineers who oversee and refine factory performance; 3) specialists who program and maintain automated systems.

These positions are fewer in number but higher in complexity and wage potential. In many respects, the AI factory mirrors a broader shift from routine cognitive tasks to advanced cognitive skills. This shift necessitates the integration of expertise across computer science, statistics, manufacturing, and process management to address problems comprehensively.

At GTC 2025 earlier this year, Jensen Huang unveiled “Blue,” a robot capable of learning tasks by modeling physical environments, a breakthrough underpinned by the same infrastructure AI factories now support. With mass production capabilities, AI factories will not only power digital models but also fabricate the next generation of physical robots for logistics, manufacturing, healthcare, and domestic use.

AI factories thus fulfill a dual role: they sustain the software revolution while concurrently driving an AI hardware transformation.

The Environmental Challenge

However, these advantages do not come without cost. AI factories are energy-intensive and, if unmanaged, carbon-heavy. With Oracle planning a $40 billion investment in Nvidia chips for its Abilene, Texas-based Stargate facility, expected to deliver 1.2 gigawatts of compute, sustainability is a key concern.

To alleviate this, Nvidia and its partners are deploying cutting-edge technologies. Liquid immersion and direct-to-chip cooling systems are being implemented to significantly reduce energy expenditure on thermal management. Nvidia’s Blackwell GPUs, utilized in the GB200 NVL72 systems, offer up to 25 times more energy-efficient inference compared to previous generations, enhancing compute output per watt.

Meanwhile, some AI factories are co-located with renewable energy sources to decrease reliance on fossil fuels. Sweden, for instance, is aiding AI factories through green energy partnerships leveraging its abundant hydroelectric power. Microsoft’s datacenters in Finland have similarly prioritized low-carbon electricity sourcing.

Sustainability is now a design prerequisite, not an afterthought. The future of AI factories may depend not only on how much compute they deliver, but on how responsibly they deliver it, balancing energy efficiency with AI capabilities. As AI transitions from the cloud into an ecosystem of factories, the organizations and countries that take the lead in the long-term development of this domain will shape the contours of the global economy.

The above is the detailed content of Nvidia Partners With Swedish Consortium: AI Factory Redefines Economy. For more information, please follow other related articles on the PHP Chinese website!