‘AI Factories’ With Millions of Agents and Robots Working With Humans
At Computex 2025 in Taipei this week, Nvidia CEO Jensen Huang offered a extra aggressive imaginative and prescient of his firm’s future, emphasizing synthetic intelligence’s want for devoted infrastructure—and Nvidia’s dedication to offer the world what it wants.
“AI is now infrastructure, and this infrastructure—similar to the web, similar to electrical energy—wants factories,” Huang advised the viewers. “You apply power in, and it produces one thing extremely helpful: tokens.”
The corporate launched the concept of those information facilities as new “AI factories,” reimagining their objective as amenities particularly optimized for synthetic intelligence workloads.
As a primary step towards important enlargement, Nvidia introduced NVLink Fusion, opening its “interconnect” expertise to different producers. This technique permits prospects to combine processors from different firms alongside Nvidia’s GPUs.
“A tectonic shift is underway: for the primary time in a long time, information facilities should be essentially rearchitected—AI is being fused into each computing platform,” Huang stated in an official press launch. “NVLink Fusion opens Nvidia’s AI platform and wealthy ecosystem for companions to construct specialised AI infrastructures.”
This improvement is especially important because it supplies builders with extra {hardware} decisions, even from third-party suppliers, for coaching their fashions. Corporations can now construct semi-custom AI programs tailor-made to particular workloads, optimizing for price, efficiency, or particular architectural benefits.
Companions together with MediaTek, Marvell, Alchip Applied sciences, Astera Labs, Synopsys, and Cadence are among the many first to undertake this expertise.
One other necessary improvement, introduced individually from Computex, is that Nvidia expanded its partnership with Neocloud suppliers. Such suppliers focus nearly completely on renting high-end GPUs, comparable to NVIDIA’s H100, H200, A100, and upcoming Blackwell chips, for AI-specific workloads like coaching giant language fashions (LLMs), laptop imaginative and prescient, and inference. A few of these new “cloud companions” embody CoreWeave, Nebius, Crusoe, and Lambda.
As a part of its imaginative and prescient of building “sovereign AI” improvement, the corporate introduced partnerships with Saudi Arabia and the United Arab Emirates to develop large infrastructure for AI coaching and deployment, primarily pushed as a part of Trump’s chip accords.
The market has responded positively to those bulletins, with Nvidia’s inventory rising nearly 60% since its dip on April 7. On Monday, although, the inventory was comparatively secure—because it has been for the final 4 days—going from $132.39 to $135.35. NVDA completed Tuesday’s buying and selling day just under the $135 mark.
Picture: Tradingview
Nvidia additionally emphasised that NVLink permits compatibility between Qualcomm processors and Nvidia GPUs. This might result in AI computer systems that includes energy-efficient CPUs paired with Nvidia’s high-performance GPUs, and even cellular {hardware} with Nvidia GPUs for sooner inference and enhanced gaming experiences.
The corporate additionally revealed plans for a brand new native headquarters referred to as Nvidia Constellation in Taipei’s Beitou Shilin Science Park. Nvidia can be increasing to Taiwan, the place the corporate plans to broaden its quantum operations.
For particular person builders, Nvidia launched DGX Spark, a compact AI workstation constructed by companions together with ASUS, Dell, and MSI. Powered by the GB10 Grace Blackwell Superchip, it gives 1,000 AI TOPS with 128GB reminiscence, supporting fashions as much as 200 billion parameters.
Only for context, Nvidia’s strongest client GPU, the RTX 5090 packs as much as 32GB of reminiscence, and begins at $2,499.
“Direct descendants of the DGX-1 system that ignited the AI revolution, DGX Spark and DGX Station are created from the bottom as much as energy the subsequent technology of AI analysis and improvement,” Huang famous.
Reservations for the DGX Spark begin at $3,999 per unit.
Picture: Nvidia
Magnificence at nighttime facet of AI?
Addressing issues about AI’s influence on employment, Huang offered a unique perspective. Somewhat than inflicting a job disaster, he sees AI as the answer to an impending workforce scarcity.
“We’ll have a scarcity of staff by 2030—about 30 to 50 million folks quick. It is truly limiting the world’s potential to develop,” Huang stated. “Now now we have these digital brokers that may work with us. One-hundred % of Nvidia software program engineers now have digital brokers working with them to assist develop higher code extra productively.”
Trying additional into the long run: “The rationale why humanoid robotics are so necessary is as a result of it’s the solely type of robotic that may be deployed nearly anyplace,” he stated, including that “this entire new trade goes to show us to large alternatives forward.”
