K-AI Chip Makers Show Divergent R&D Spending Amid Production Ramp

South Korean artificial intelligence semiconductor fabless companies displayed sharply divergent research and development spending patterns in 2025 financial reports filed with the Financial Supervisory Service (FSS) on April 16, 2026, reflecting their distinct positions in the product development lifecycle. Rebellion and HyperAccel concentrated on next-generation chip design and prototyping with over 10 billion Korean won (KRW) annually, while Furiosa AI and DeepX transitioned to mass production, causing their reported R&D expenses to decline as development costs were reclassified as cost of goods sold, according to the FSS electronic disclosure system.

Heavy R&D Investment in Next-Generation Chip Development

Rebellion executed 119.8 billion KRW in ordinary research and development expenses in 2025, representing a 46.6% increase from 81.7 billion KRW in 2024, according to the company’s consolidated audit report filed with the FSS. This expenditure exceeded the company’s total revenue of 32 billion KRW by 3.7 times, reflecting intensive investment in next-generation AI semiconductor design and prototype manufacturing. The spending reflects Rebellion’s focus on securing advanced-node intellectual property (IP) from Samsung Electronics’ foundry division and covering design team payroll for global engineering talent, per the FSS filing and company disclosures.

HyperAccel similarly prioritized R&D investment, executing 4.22 billion KRW in ordinary research and development expenses in 2025, according to the company’s consolidated audit report. The company expanded initial chipset design personnel and increased related payroll costs to capture the large language model (LLM) accelerator market, per HyperAccel’s FSS filing. HyperAccel Chief Technology Officer Jin-won Lee stated in an official company communication that “as the company continues to expand personnel, research and development expenses are expected to increase continuously.”

Rebellion CEO Park Sung-hyun leads the company’s strategy to prioritize next-generation chip technology development despite growing operational losses.

Production Transition and Cost Reclassification

Furiosa AI’s ordinary research and development expenses declined to 36.2 billion KRW in 2025 from 56.3 billion KRW in 2024, a 35.7% decrease, according to the company’s consolidated audit report filed with the FSS. However, this reduction does not reflect diminished development activity. Rather, the second-generation product “Renegade” (RNGD) completed its development cycle and entered mass production (MP) phase, causing prototype manufacturing and testing costs to shift from the research and development category to cost of goods sold, per Furiosa AI’s official statement to the FSS. A Furiosa AI representative explained: “As development concluded and mass production began, related costs are now classified as cost of goods sold, which is why research and development expenses decreased compared to last year. Starting this year, we expect substantial research and development expenses to be recorded again as we begin third-generation chip development.”

DeepX, which specializes in edge artificial intelligence applications, reported 8.2 billion KRW in ordinary research and development expenses in 2025, according to the company’s consolidated audit report. The decline reflects the completion of first-generation chip development (DX-M1) and the transition to global distribution supply channels, with substantial prior research and development costs now reclassified as cost of goods sold due to commercialization, per DeepX’s FSS filing. The company’s accounting treatment shifted from development-phase expense recognition to manufacturing and initial supply cost classification as the product entered commercial deployment.

Profitability Verification as the Critical Benchmark

While the four companies’ reported R&D figures diverged, industry participants identified profitability validation following chip commercialization as the key observation point, according to industry commentary filed with regulatory oversight bodies. The reclassification of research and development expenses to cost of goods sold signals that fabless companies now face tangible inventory and cost-of-goods-sold financial risks as chips transition from design to production, per industry analyst statements. Industry participants emphasized the urgency of demonstrating mass production efficiency capable of generating actual market margins, according to statements provided to regulatory authorities.

An industry representative stated: “Until now, companies could secure investment based on market expectations alone. Now they must prove themselves through mass production. Once chips are actually manufactured, management capabilities in inventory management and cost reduction will become critical,” and noted that “these factors will also influence initial public offerings (IPOs),” per industry commentary documented in regulatory filings.

Frequently Asked Questions

Q: Why did Furiosa AI and DeepX report lower R&D expenses in 2025 compared to 2024?

As their flagship products transitioned from development to mass production, the accounting classification of related costs changed from research and development expenses to cost of goods sold. This reclassification is a standard accounting practice reflecting the product lifecycle stage, not a reduction in development activity. According to Furiosa AI’s official statement to the FSS, the company plans to resume substantial R&D spending in 2026 as it begins third-generation chip development.

Q: How much did Rebellion increase its R&D spending year-over-year?

Rebellion increased ordinary research and development expenses by 46.6%, from 81.7 billion KRW in 2024 to 119.8 billion KRW in 2025, according to the company’s consolidated audit report filed with the Financial Supervisory Service on April 16, 2026. This spending exceeded the company’s total revenue of 32 billion KRW by 3.7 times.

Q: What is the significance of the cost reclassification from R&D to cost of goods sold?

The shift indicates that fabless companies now carry tangible inventory and manufacturing cost risks as their chips move into commercial production. According to industry commentary, this marks the transition from technology validation to operational efficiency testing, where profitability depends on managing production yields, inventory levels, and unit economics rather than R&D intensity alone.