A Silicon Valley startup just secured major backing to solve one of AI's biggest economic problems: the idea that only hyperscalers with massive cloud infrastructure can afford to deploy artificial intelligence. Quadric, a Burlingame, California-based chip company, announced the second close of its Series C funding round, bringing total capital raised to $90 million. The new $46 million injection, led by the International Finance Corporation (the private sector arm of the World Bank Group), signals growing confidence that on-device AI inference represents a genuine alternative to cloud-dependent systems.

The funding milestone matters because it reflects a fundamental shift in how the AI industry thinks about computing. For years, running large language models and complex AI workloads meant sending data to remote servers, paying per-token fees, and accepting latency delays. Quadric's approach flips that model: its Chimera general-purpose neural processing unit (GPNPU) architecture allows AI models to run directly on devices, from smartphones and laptops to automotive systems and industrial equipment.

What Makes Quadric's Chip Different From Other AI Hardware?

Most AI chips are designed with fixed features locked in years before they ship. By the time a new processor reaches the market, AI models have already evolved, leaving the hardware behind. Quadric's programmable architecture solves this by allowing the same physical chip to adapt to new models through software updates, rather than requiring entirely new silicon.

The Chimera GPNPU can scale from 1 to more than 3,200 TOPS (tera operations per second) in multi-chiplet configurations. That range means the same core architecture works for everything from wearables and edge devices to enterprise servers. The chip supports convolutional neural networks, transformer-based models, on-device large language model (LLM) inference, and emerging vision-language-action world models.

"A chip feature set is locked years before it ships, and AI models change every few months, so an operator-centric, fixed-function NPU arrives behind the models and only falls further back. Quadric is a living platform: because the stack is software, the same chip runs new models and gets faster long after it ships," said Veerbhan Kheterpal, CEO and Co-Founder of Quadric.

Veerbhan Kheterpal, CEO and Co-Founder of Quadric

The company's software toolchain converts AI models into C++, and system-on-chip design teams can write code in Python or C++. This flexibility allows engineers to deploy both AI inference workloads and classic digital signal processing (DSP) and control algorithms on the same hardware.

How Does This Change the Economics of AI for Smaller Businesses?

The World Bank's investment signals recognition that on-device AI could help close the digital divide. Cloud-based AI services charge per token, meaning small and medium enterprises (SMEs) in emerging markets face ongoing costs that can price them out of AI adoption entirely. With local inference, businesses own the hardware and run models without per-token cloud bills.

"Powerful AI cannot remain the exclusive domain of hyperscalers if emerging markets are going to close the digital divide. Quadric's programmable architecture fundamentally changes the economics: SMEs in emerging markets can now deploy AI on devices they own, without the per-token cloud bills that price them out," explained Mohamed Eissa, Chief Investment Officer at the International Finance Corporation.

Mohamed Eissa, Chief Investment Officer at the International Finance Corporation

This economic argument resonates beyond emerging markets. Enterprise customers in automotive, AI PCs, and other sectors are already adopting Quadric's technology. The company reported that product revenue more than tripled in the year leading up to this funding round, and it reached profitability, suggesting real market demand.

Steps to Understand On-Device AI Infrastructure for Your Organization

• Assess Your Workload Requirements: Determine which AI tasks benefit from local processing (privacy-sensitive operations, low-latency features, offline functionality) versus those that still need cloud resources for complex training or real-time updates.
• Evaluate Hardware Capabilities: Modern processors now include neural processing units (NPUs) designed for AI inference. Check whether your current or planned device fleet includes this specialized silicon, and understand what performance levels it can deliver for your specific models.
• Plan for Software Compatibility: Unlike fixed-function chips, programmable AI platforms allow models to be updated through software. Verify that your chosen hardware supports the AI frameworks and model formats your development teams use.
• Consider Total Cost of Ownership: Compare the upfront cost of on-device hardware against ongoing cloud API fees, bandwidth costs, and the operational overhead of managing cloud dependencies for mission-critical AI features.

The broader context matters here. Nvidia recently unveiled RTX Spark, a small AI computing platform designed to bring AI capabilities directly to personal computers, allowing users to run AI models locally instead of relying solely on cloud services. Microsoft, meanwhile, is positioning Windows 11 as a platform for local AI through its Copilot+ PC initiative, which uses neural processing units to run certain machine learning tasks directly on devices.

These developments suggest the industry is moving away from a purely cloud-centric AI model. Local AI processing reduces latency, preserves bandwidth, enables features to run without constant internet connectivity, and supports scenarios where privacy and data residency matter. For enterprises, this creates new planning questions: which features should run locally, which should remain cloud-backed, and which might use a hybrid approach depending on the task and available hardware.

Quadric's funding round included participation from existing investors Pear VC, Uncork Capital, and BEENEXT, along with a new investor, Offline Ventures, co-founded by Facebook Platform creator Dave Morin and former Apple executive James Higa. Pear VC led Quadric's seed round and doubled down in this round, signaling confidence in the company's execution and market opportunity.

The company is preparing for demand across humanoid robotics, wearables, and networking applications, suggesting that on-device AI is not just a near-term trend but a structural shift in how computing will work across multiple industries. As AI models continue to evolve and new applications emerge, the ability to update hardware capabilities through software rather than waiting for new silicon becomes increasingly valuable.