AMD and Intel Push 'Agent Computer' as AI Shifts From Cloud to Desktop

A quiet battle is emerging over the future location of artificial intelligence: not in distant data centers, but on the desk in front of you.

AMD has begun promoting what it calls the "Agent Computer," a new product category designed to run AI models locally rather than relying on cloud services. The company is positioning its latest Ryzen AI Max processors as purpose-built for users who want to run AI agents on their own hardware, citing open-source projects that already enable laptop-based AI workflows. The timing is deliberate: AMD announced the initiative days before Nvidia's annual GTC developer conference, where the GPU giant is expected to discuss its own roadmap for local AI hardware.

Intel is pursuing a parallel strategy. ASUS announced its NUC 16 Pro Mini PC, powered by Intel's Core Ultra X9 processor with integrated Arc B390 GPU and a 50 TOPS neural processing unit. The compact system is marketed as an "Edge AI-ready" machine capable of handling videoconferencing, analytics, and language translation without internet connectivity. ASUS's AI SuperBuild software allows users to run customized large language models locally, a feature the company emphasizes for handling sensitive data like patient records or financial information.

Nvidia itself has entered the local AI market. Last year it began selling the DGX Spark, a $3,999 mini PC supporting up to 128GB of RAM, with partners including Dell developing their own versions. A more powerful DGX Station is slated to arrive this spring.

(The push toward local AI processing reflects growing concerns about data privacy, latency, and the cost of continuous cloud inference. It also represents a strategic opening for chipmakers who have struggled to compete with Nvidia in the data center market.)

The competitive landscape remains fluid. While Nvidia dominates AI training workloads with its GPUs, inference—the continuous process of running deployed models—is emerging as a distinct battleground. Startups including SambaNova, which raised $350 million after acquisition talks with Intel collapsed, and Tenstorrent, valued at $2 billion, are building alternatives they position as cheaper and more efficient than traditional GPUs. Intel subsequently announced a multiyear collaboration with SambaNova and invested in its Series E round.

The shift toward local AI hardware also carries geopolitical weight. U.S. export controls have restricted Nvidia's ability to sell advanced chips to China, a constraint that Nvidia CEO Jensen Huang has warned will only accelerate Chinese domestic chip development. Meanwhile, the broader AI hardware ecosystem has become increasingly tangled, with companies like Broadcom simultaneously competing with Nvidia on chip design while supplying the networking technology that connects Nvidia's GPUs.

Whether "Agent Computer" becomes industry vernacular or fades as marketing jargon remains uncertain. But the underlying trend is clear: as AI matures beyond experimental chatbots, the question of where intelligence resides—centralized in cloud infrastructure or distributed across millions of devices—is becoming a defining strategic choice with implications for privacy, sovereignty, and market power.