Korean neural processing unit (NPU) startups are on the verge of a major inflection point, with companies like FuriosaAI, Rebellions, and DeepX targeting combined revenues that could exceed 200 billion won this year, up from just tens of billions in 2025. This represents a dramatic acceleration as chips that spent months in proof-of-concept testing with customers are now entering full-scale production and deployment.

What's Driving This Sudden Revenue Surge?

The growth stems from a fundamental shift in how companies are buying NPUs. Rather than experimenting with small batches, organizations are now placing serious orders based on successful pilot projects. FuriosaAI, for example, set a revenue target of 80 billion to 100 billion won for this year, a roughly 17-fold increase from its 5.7 billion won revenue in 2025. The company plans to produce 20,000 units of its second-generation chip, called "Renegade," with about 7,000 units destined for actual customer deployments.

Rebellions is following a similar trajectory. Market observers project the company will generate more than 80 billion won in revenue this year, compared to 32 billion won last year, driven by deliveries of its "Rebel100" chip in the second half of the year. DeepX is being watched even more closely, with some analysts expecting it to hit 200 billion won in revenue this year. The company already had 14 billion won in preliminary orders through last month, four times its entire 2025 revenue of 3.3 billion won.

Where Are These Chips Actually Going?

The geographic diversity of demand is striking. While Korean startups are naturally strong in domestic markets, they're aggressively expanding internationally. DeepX recently signed a strategic partnership with Koshidaka Tech, a Japanese technology distributor, and built a cooperation framework with Avnet Asia, a major global electronic components distributor, to expand its reach across Asia and beyond.

Perhaps most intriguingly, industry observers are paying close attention to the Middle East market. Governments and sovereign wealth funds in that region are building comprehensive AI ecosystems spanning data centers, cloud infrastructure, and AI models and applications. This could translate into enormous NPU orders. FuriosaAI has already positioned itself in this market by signing an AI full-stack cooperation agreement with Saudi Aramco Digital, alongside partners LG AI Research and Upstage.

How Are These Companies Scaling Production?

The transition from prototype to mass production is the critical test. Each company is pursuing slightly different strategies to ensure they can meet demand without overcommitting resources:

• FuriosaAI's Approach: The company is ramping production of its Renegade chip to 20,000 units annually, with plans to allocate roughly 35% to direct customer sales and the remainder to building out the ecosystem of applications and services that make the chip valuable.
• Rebellions' Timeline: The company is focusing on second-half deliveries of its Rebel100 chip, suggesting a staggered production schedule designed to manage supply chain risks and ensure quality control.
• DeepX's Distribution Strategy: Rather than building its own manufacturing capacity, DeepX is leveraging partnerships with major global distributors to handle supply chain logistics, allowing the company to focus on chip design and customer relationships.

An industry official noted the importance of execution at this stage: "Depending on the mass production and performance of second-generation chips, development of third-generation chips could also proceed smoothly. Each company's grade will be determined by how much is actually sold to customers and how much is used in practical services".

Why Should You Care About Korean NPU Startups?

Neural processing units are specialized chips designed to run artificial intelligence workloads efficiently. Unlike general-purpose processors, NPUs are optimized for the specific mathematical operations that power machine learning models. As AI becomes embedded in everything from smartphones to data centers, the demand for efficient NPU hardware is exploding globally.

Korean startups entering this market represent a significant challenge to established players like NVIDIA and AMD, which have dominated AI chip design for years. If these Korean companies can deliver reliable, cost-effective NPUs at scale, they could reshape the global AI infrastructure market. The revenue targets they're chasing this year are relatively modest in absolute terms, but they signal confidence that the proof-of-concept phase is ending and real-world deployment is beginning.

The success or failure of these companies over the next 12 to 18 months will likely determine whether Korea becomes a lasting player in the AI chip industry or whether these startups remain niche suppliers. The stakes are high, and the market is watching closely.