Carbon News | 90% Cheaper! Chinese AI Captures the U.S. Market

Experts from the European Centre for Medium-Range Weather Forecasts stated on July 7 that the current El Niño event is expected to reach an unprecedented overall intensity, exposing regions worldwide to associated risks—including droughts, floods, and other extreme weather events. According to AFP’s report quoting Tim Stockdale, an expert at the center, this year’s El Niño differs markedly from observations over the past three decades. Forecast models indicate it will be an extreme event. He said, “We have never seen such strong and consistent model predictions for an El Niño event,” and barring unforeseen developments, it will break historical records. In early July, the World Meteorological Organization (WMO) issued a press release stating that El Niño conditions have already developed across the tropical Pacific and are expected to rapidly intensify into a strong El Niño event between July and September—thereby increasing the likelihood of heatwaves, droughts, heavy rainfall, and other extreme weather events across many parts of the world. (Source: Xinhua News Agency) 07 AI Token Expenditure Index Drops Nearly 20%; Marginal User Willingness to Pay May Cool According to Bloomberg, the LLM Token Expenditure Index—compiled by Silicon Data—nearly doubled since its launch in December last year but peaked in May and has since declined nearly 20%. Widely regarded as the most direct market indicator reflecting willingness to pay for AI services and shifts in marginal demand, this index’s weakening has triggered renewed reassessment of AI’s commercialization pace. Some investors believe this shift may signal AI firms’ entry into a more cost-sensitive competitive environment, where pricing power could be eroding. Veteran investor Louis Navellier noted emerging signs that some users are already limiting usage frequency due to cost pressures. Meanwhile, market rumors about OpenAI potentially delaying its IPO are also viewed as indirect evidence of ongoing profitability challenges. (Source: Wall Street Insight) 08 EIA: AI Boom Drives U.S. Electricity Consumption to Record Highs Year After Year On July 7 (U.S. Eastern Time), the U.S. Energy Information Administration (EIA) released its Short-Term Energy Outlook. In this report, the EIA revised upward its U.S. electricity demand forecast, projecting annual electricity consumption to hit record highs consecutively in both 2026 and 2027—driven primarily by rapid expansion of AI data centers, advancing electrification, and growth in commercial and residential electricity use. The EIA forecasts that U.S. annual electricity demand will rise from a record 4.195 trillion kilowatt-hours (kWh) in 2025 to 4.269 trillion kWh in 2026, and further to 4.399 trillion kWh in 2027. The report specifically highlights that the rapid proliferation of AI applications is prompting major tech companies to continuously expand their data centers; computing infrastructure has thus become one of the most significant sources of new U.S. electricity demand—and will continue supporting natural gas-fired power generation demand. (Source: EIA) 09 Amazon Plans to Issue at Least $25 Billion in Bonds to Accelerate AI Infrastructure Investment On July 7, Amazon launched a new round of bond issuance totaling at least $25 billion to bolster investment in AI infrastructure. Upon announcement, AI-related bonds suffered massive sell-offs: credit spreads on bonds issued by hyperscale cloud providers—including Meta and Amazon—widened sharply in a single day, with declines far exceeding typical fluctuations in the investment-grade market.

AI Computing Power Ignites a Revolution in Power Supply Architecture: Full Decoding of Data Center Power Delivery Transformation at PCIM Asia Shenzhen 2026

In 2026, the global AI computing power race has reached a fever pitch—but a fundamental physical constraint is surfacing: electricity is becoming the primary bottleneck for computing power expansion. The U.S. Energy Information Administration (EIA) forecasts that AI-driven data centers will push U.S. electricity demand from 40,950 billion kWh in 2025 to 43,810 billion kWh in 2027. The resulting ‘computing power without electricity’ gap is forcing unconventional solutions: Oracle has signed a 2.8 GW solid oxide fuel cell (SOFC) order, while China’s Jereh Group secured over $900 million in export contracts for gas turbine generator sets in Q1 alone. The true foundational transformation is unfolding deep within data center power delivery architecture: the traditional AC UPS + 48V bus system is showing strain under the impact of rack-level power densities exceeding 100 kW. A sweeping revolution—from AC to DC, from low voltage to high voltage, and from silicon to wide-bandgap semiconductors—is simultaneously taking place across all segments of the supply chain. In China, this transformation is receiving robust policy support. According to media reports, “coordinated planning of computing and electricity” has been explicitly incorporated into both the Government Work Report and the Outline of the 15th Five-Year Plan, elevating it to the status of a national-level new infrastructure project. In March this year, Shenzhen issued the “Shenzhen Action Plan for Accelerating High-Quality Development of the Artificial Intelligence Server Industry Chain (2026–2028),” which explicitly calls for “advancing iterative development of next-generation HVDC (high-voltage direct current) power supply solutions, and accelerating R&D and innovation of key components including high-voltage SiC MOSFETs, high-frequency GaN power devices, multiphase controllers, and intelligent power stages (DrMOS).” The clarion call for industrial transformation has already sounded. As the frontier of computing power becomes defined by electricity, the restructuring of AI data center power delivery architecture is accelerating. From August 26 to 28, 2026, PCIM Asia Shenzhen 2026 will open at Halls 14 & 16 of the Shenzhen World Exhibition & Convention Center (Bao’an New Venue). Its newly launched “AI Power Ecosystem Showcase Zone” will concentrate on displaying the complete power delivery chain—from grid interface to core chips. Strategic partners Mitsubishi Electric and Infineon Technologies, along with platinum sponsors Fuji Electric, Semikron Danfoss, and ROHM—as well as over 300 leading domestic and international power semiconductor enterprises—will jointly present full-chain technological breakthroughs spanning from grid to chip, collectively addressing the pivotal challenge of AI data center power delivery transformation. Mitsubishi Electric: Unifull™ Series SiC MOSFETs Target Solid-State Transformer Demand Responding to the surging demand for power semiconductors driven by AI computing infrastructure in 2026, Mitsubishi Electric (16E01) has positioned its Unifull™ series of high-voltage SiC MOSFETs and second-generation industrial SiC MOSFETs as core product offerings—precisely meeting data center customers’ stringent requirements for energy efficiency. SiC MOSFET The Unifull™ series is engineered for high power density, featuring an embedded Schottky Barrier Diode (SBD) structure. For example, a 3.3 kV / 800 A module reduces switching losses by 91% compared to conventional silicon modules. This series introduces new low-current variants—3.3 kV / 200 A and 400 A—with LV100-packaged modules achieving a power density of 50 kW/L. Operating across a wide temperature range of −40°C to 125°C, these modules meet the demanding requirements of solid-state transformers (SSTs) in AI data centers.

US Power Use Projected to Hit Record Highs in 2026–2027 Amid AI Surge, EIA Says

Oct 6, 2026

The US Energy Information Administration forecasts record power consumption in 2026–2027 driven by AI data centers and electrification, with wholesale electricity prices rising 11% and coal generation shares declining.

reuters

US Power Use Projected to Hit Record Highs in 2026–2027 Amid AI Surge, EIA Says

FAQ