Analog Devices Eyes $1.5B AI Power Deal; China's CXMT Bets on Memory Surge

Power delivery M&A meets AI memory nationalism as the chip infrastructure trade shifts upstream from GPUs.

💰 Analog Devices Eyes $1.5B Cash Deal for AI Power Chip Startup Empower Semiconductor

Decoded: Analog Devices is in advanced talks to acquire AI power management chip startup Empower Semiconductor for approximately $1.5 billion in cash, Bloomberg News reported on May 18, citing people familiar with the matter; Reuters confirmed the discussions. Empower Semiconductor designs power-delivery integrated circuits specifically for high-density AI data center workloads — chips that regulate voltage demands of GPU clusters running large-scale AI inference. The acquisition target addresses a constraint that has emerged as a central infrastructure bottleneck in 2026: as Big Tech data center capex commitments now exceed $650 billion across Microsoft, Alphabet, Amazon, Meta, and Apple, the binding constraint has shifted upstream from GPU supply (2024) to site availability (2025) to power-delivery architecture (2026). Analog Devices (ADI) has not publicly confirmed the deal. (Reuters, Bloomberg, May 18-19, 2026)

Why it matters: Power management chips are the critical layer that determines whether a data center can run at the compute density it was built for. Empower Semiconductor's $1.5 billion private valuation signals the market is pricing the power-delivery bottleneck as a structural constraint. For investors, ADI's move positions the company — historically known for industrial and automotive analog chips — directly in the AI infrastructure supply chain at a layer with limited prior public equity exposure. The deal also signals consolidation pressure on other power-delivery chip startups, as hyperscalers seek supply from financially stable partners.

🗄️ China's CXMT Forecasts Revenue Surge as AI Memory Demand Accelerates

Decoded: China's Changxin Memory Technologies (CXMT), the country's primary DRAM producer, said it expects revenue to surge as demand for memory chips used in AI workloads continues to accelerate, Reuters reported on May 19. CXMT has been scaling production of DDR5 and LPDDR5 memory — components used across AI training servers, edge AI devices, and data center infrastructure — and is targeting the supply gap created by Western export controls restricting Micron's China operations. CXMT does not currently produce high-bandwidth memory (HBM) at competitive volumes, leaving it outside the highest-margin AI memory segment dominated by SK Hynix and Samsung, but the company is expanding capacity in standard AI-grade DRAM serving a broader infrastructure market. (Reuters, May 19, 2026)

Why it matters: CXMT's revenue expectations reflect two dynamics running simultaneously: China is scaling domestic memory production as a strategic response to Western chip controls, and AI server memory demand is broad enough to sustain multiple supplier tiers below the HBM premium segment. For Micron (MU) investors, CXMT's expansion in standard AI-grade DRAM is the most direct competitive pressure on Micron's China-accessible product lines. For the broader chip trade, CXMT's ascent is evidence that export controls have accelerated China's domestic semiconductor investment — a policy outcome that will continue shaping the geopolitical chip competition through the next funding cycle.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team