A New Phase in AI Chip Competition
The emerging AI chip competition between Nvidia and Euclyd describes a race to design processors that run advanced artificial intelligence workloads locally on personal computers, replacing dependence on cloud data centers and reshaping how users interact with software and hardware across everyday devices. Nvidia, a $5tn US semiconductor giant, has introduced RTX Spark, a “superchip” that brings AI capabilities directly into consumer laptops and desktop computers. According to The Guardian, this development could even “replace the mouse and keyboard in how people use computers,” signaling a profound change in user interfaces. At the same time, Euclyd, a Dutch startup, is building a rival architecture that claims far greater energy efficiency. Together, these efforts signal a shift away from remote servers toward local AI processing, where PCs and future AI PCs become the primary platforms for running large language models, AI agents, and creative tools.
RTX Spark: Turning Everyday PCs into AI Machines
RTX Spark is Nvidia’s attempt to make advanced AI a standard feature of consumer hardware rather than a cloud-only capability. The chip enables laptops and desktops to run complex AI models locally, including large language models, AI agents, and creative AI applications, without sending data to distant servers. This promises faster response times, better privacy, and less reliance on constant internet connections. TechDigest reports that RTX Spark will appear in systems from Dell, Lenovo, Asus, and HP, paired with Microsoft Windows, turning mainstream PCs into AI PCs by default. The market reaction has been positive, with the new platform forming “a new standard” for AI PCs, according to IOPlus. For Nvidia, RTX Spark also serves as a bridge to its more advanced Vera Rubin architecture, intended to power future AI factories and, eventually, consumer machines with far greater performance.
Vera Rubin vs Craftwerk: Competing Architectures for Local AI
Behind RTX Spark sits Nvidia’s next architecture, Vera Rubin, which is designed to deliver higher performance for both data centers and future AI PCs. IOPlus describes Vera Rubin as a modular system that pairs powerful Vera processors with Rubin graphics cards, built on a 3-nanometer platform and featuring 288 gigabytes of HBM4 memory for handling vast AI workloads. Nvidia targets massive AI “factories,” with mass production planned for late 2026, and expects consumer devices to benefit soon after. Euclyd counters with its Craftwerk architecture, which it claims could be up to one hundred times more energy-efficient than Vera Rubin for AI inference tasks. Craftwerk uses 16,384 tightly coupled processors to reach an expected 32 petaflops of computing power, aiming directly at the energy and cost challenges of AI inference at scale. If Euclyd’s claims hold, the balance between raw power and efficiency in AI chip design could shift dramatically.
Euclyd’s Bid to Break an AI Hardware Monopoly
Euclyd positions itself as a challenger to Nvidia’s dominance in AI inference, especially in markets hungry for efficient local AI processing. Founded by Bernardo Kastrup, the company is working to commercialize its Craftwerk architecture and is in talks to raise €100 million in growth capital, according to IOPlus. Its planned first commercial production in 2028 gives it a longer runway, but also more time to refine and validate its bold efficiency claims. By focusing on an alternative architecture rather than copying GPU-style designs, Euclyd aims to address the mounting power demands of AI factories and edge devices alike. Success would not only diversify the AI chip landscape but also offer PC makers, cloud providers, and enterprises a second major option for local AI processing hardware, potentially reducing reliance on a single dominant supplier.
From Cloud-First to Local AI Processing
The contest between Nvidia and Euclyd reflects a broader shift from cloud-first AI deployment to local AI processing on consumer devices. RTX Spark enables PCs to run sizeable AI workloads without constant cloud access, keeping data on-device and easing pressure on data centers already facing noise, environmental, and infrastructure concerns, as highlighted by reports on data center impact in Sterling, Virginia. In parallel, AI tools such as OpenAI’s GPT-5.5 Cyber for banks show that security-critical workloads are still likely to rely on specialized cloud systems, at least for now. Over time, more AI tasks—interface control, productivity, creative work, and some security analysis—are expected to move onto PCs and edge devices. As Nvidia and Euclyd compete on performance and efficiency, their designs will shape how much AI happens near the user, how much stays in the cloud, and who leads the next generation of AI chip architecture.
