1. Google & Blackstone Launch AI Neo-Cloud Joint Venture — Blackstone Commits $5B Equity, 500MW Target by ’27, Google’s Largest TPU Monetization Move Yet
• Core Source
“Alphabet and Blackstone plan to establish a cloud company to compete with neo-cloud players like CoreWeave using TPUs”
“Blackstone plans to invest $5bn in equity capital”
“This is the largest attempt by Google to sell and monetize TPUs to external customers, with a goal of operating 500MW of capacity by 2027”
“Approximately $25bn in computing investment is expected to be supported going forward, including leverage”
“One is a contract providing Anthropic with access to approximately one million chips, and the other is a contract with Meta Platforms. This collaboration with Blackstone is the first large-scale case of aggressively expanding beyond Anthropic/Meta”
• Expected Impact
The Google-Blackstone joint venture is significant not merely as a data center investment deal, but as the formalization of a business model that converts Google’s TPUs into an external revenue stream.
TPUs had long been used exclusively for Google’s internal AI workloads. However, following large-scale contracts with Anthropic (access to approximately one million chips) and Meta, the establishment of a separate entity with Blackstone signals that a cloud business model packaging TPUs + software + services for external customers has now gone mainstream. The appointment of Benjamin Treynor Sloss, a long-time Google executive, as CEO suggests this is a long-term strategy rather than a one-off pilot.
In terms of market impact, competitive pressure on existing neo-cloud players such as CoreWeave and Nebius intensifies considerably. The fact that Google, as a chip designer, is directly entering cloud infrastructure operations puts it in direct conflict with the ‘NVIDIA GPU rental’ business model that neo-clouds have relied upon. The immediate decline in CoreWeave and Nebius share prices following the WSJ report reflects this dynamic.
For Blackstone, this joint venture represents a strategic elevation from data center real estate investment to AI computing infrastructure operations, following QTS Realty Trust (acquired in ’21) and AirTrunk (acquisition agreed in ’24). The approximately $25 billion in investment including leverage is concrete evidence that the AI infrastructure CAPEX cycle continues to accelerate.
2. NextEra Acquires Dominion for $66.7B — Combined Enterprise Value of $400B, 51GW Data Center Alley Power Secured, World’s Largest Listed Utility Born
• Core Source
“NextEra acquires Dominion for approximately $66.7 billion (approximately ₩100 trillion, all-stock transaction). Combined enterprise value of $400 billion (= ₩600 trillion), creating the world’s largest listed utility”
“Data Center Alley handles approximately two-thirds of global internet traffic, making it the world’s largest data center concentration”
“Dominion is already connected to more than 450 data centers in Virginia, with 28% of Virginia’s power sales last year going to data centers”
“Contracted data center capacity of approximately 51GW secured. Customers include Google, Amazon, Microsoft, Meta, Equinix, CoreWeave, CyrusOne, and the full roster of Big Tech”
“NextEra CEO John Ketchum stated: ‘America needs to expand energy infrastructure faster, more efficiently, and more affordably than ever before'”
•Expect Impact
The crux of this deal is not simply a merger between utility companies, but a race to secure exclusive supply rights over the most critical region for AI data center power demand.
Dominion’s Northern Virginia ‘Data Center Alley’ is the world’s largest data center hub, through which approximately two-thirds of global internet traffic passes. Google, Amazon, Microsoft, Meta, CoreWeave, and virtually all of Big Tech are already customers, with contracted data center capacity of approximately 51GW already locked in. Last year, 28% of Virginia’s power sales were already directed to data centers. In an environment of exploding AI demand, the power supply rights to this region have become not merely a utility asset but the core bottleneck of AI infrastructure.
For NextEra, adding regulated utility assets concentrated with AI data center power demand to its existing position as a renewable energy powerhouse secures a stable, long-term revenue base. Upon deal completion, approximately 80% of the merged company’s revenue and profit will come from regulated utility operations, reducing exposure to more volatile unregulated businesses.
The bigger picture is that utility-scale economics are emerging as a new variable in AI infrastructure competition. As noted in the JP Morgan expert call, one of the core reasons approximately 60–70% of planned AI data center expansion is expected to face delays of one to two years is the power supply bottleneck. Gas turbine lead times exceed three to four years, and bringing new utility generation capacity online requires more than five to six years. In this environment, Dominion’s already-secured 51GW of contracted capacity and its network of over 450 connected data centers constitute a virtually irreproducible infrastructure asset.
3. Citi Raises Micron Target to $800 from $425 — DRAM +200% YoY, NAND +186% Forecast, HBM Capacity Expansion Incentive Fading, ’27 HBM Price Further Upside Signaled
• Core Source
“In aggregate, we estimate DRAM prices will increase +200% YoY and NAND prices will increase +186% YoY in ’26”
“Memory companies currently have little incentive to expand HBM capacity further. This is because the wafer conversion ratio for HBM production is 3–4x, and the profitability difference versus commodity DRAM has narrowed significantly”
“Given the tight HBM capacity situation, we forecast HBM prices will rise again further in ’27”
“Memory companies are expected to maintain CAPEX discipline on supply expansion to prevent HBM content per AI data center from declining next year”
“Raising Micron’s target price from $425 to $800”
• Expected Impact
This report delivers two core insights. First, the sharp rise in memory prices in 2026 is already materializing. Second, and more importantly, there are structural reasons why the upcycle will persist through 2027.
The primary driver of the 2026 DRAM price surge is a supply-demand imbalance in commodity DRAM. Even as semiconductor equipment makers such as AMAT raised their systems revenue outlook by approximately +30%, the resulting DRAM bit supply growth remains around +30%, failing to keep pace with AI demand growth.
More significant is the logic of fading HBM expansion incentives. HBM production requires a wafer conversion ratio of 3–4x, and with commodity DRAM prices surging, the profitability gap between the two products has narrowed. Memory makers have little reason to aggressively expand HBM capacity. As a result, HBM supply remains tight, and HBM prices are forecast to rise further in 2027.
4. JPM Conference Unveils Boston Dynamics Roadmap — 30,000 Units/Year Production by 2028, Hyundai Internal Demand of 25,000+, Atlas Equipped with Google DeepMind Reasoning AI
• Core Source
“Factory production capacity presented at 30,000 units per year”
“HMG internal captive demand is more than 25,000 units”
“Google DeepMind handles the Reasoning AI Layer; Boston Dynamics handles the Physical AI Layer. In other words, judgment and reasoning are handled by Google DeepMind, while physical control and execution are handled by Boston Dynamics”
“The Robot Metaplant Application Center is scheduled to launch in summer 2026 at Hyundai Motor’s Georgia Metaplant (AI training, data collection, and real-world verification hub)”
“Plans call for launching a U.S. robot production platform and actuator manufacturing facility in 2028 (actuator production capacity of more than 350,000 units per year)”
• Expected Impact
The roadmap unveiled by Boston Dynamics at the JP Morgan Conference represents a formal declaration of transition from the ‘demonstration phase’ to full-scale industrial mass production.
The most notable point is the role-sharing structure with Google DeepMind. The ‘dual AI brain’ architecture — where DeepMind handles reasoning and judgment while Boston Dynamics handles physical control and execution — goes beyond technical collaboration to formally establish the pathway through which Google’s AI capabilities expand into the physical AI market. This means Google DeepMind’s reasoning AI can be directly monetized in the robotics market.
For the Hyundai Motor Group, the figure of internal demand exceeding 25,000 units is the key number. The virtuous cycle of accumulating data through actual deployment across more than 130 factories and feeding it back into AI training creates a real-world data-driven moat that competitors will struggle to replicate in the short term. The launch of RMAC within Hyundai’s Georgia Metaplant in summer 2026 marks the official starting point of this data accumulation.
There is analysis suggesting that manufacturing an Optimus without Chinese supply chains would triple the cost, which implies that Chinese robotics companies with ready access to Chinese supply chains hold a structurally superior cost position. In this environment, the Boston Dynamics-Hyundai Motor Group combination is well-positioned to emerge as the leading counterpart from the U.S.-Korea camp, combining technological excellence with manufacturing capability.
5. Panasonic Pivots Portfolio to AI Data Center ESS — FY2027 ESS Revenue ¥550B (+71% YoY), EV Battery Lines Converted to Data Center Use
• Core Source
“The Industrial/Consumer battery segment posted revenue of ¥151.3B (YoY +44%, QoQ -1%) and operating profit of ¥23.3B (YoY +74%, QoQ -20%), driven by expanding data center ESS demand”
“Data center ESS revenue is forecast to expand from ¥322B in FY2026 to ¥550B in FY2027 (YoY +71%), and the ¥800B sales target previously set for FY2029 has been brought forward to FY2028”
“The existing EV battery production line at the Suminoe plant in Japan has been converted to data center use, with shipments beginning in April 2026, and the Tokushima plant plans to expand data center cell production capacity approximately threefold by FY2029 versus FY2026”
“Panasonic Industry plans to mass-produce a CBU (Capacitor Backup Unit) utilizing its proprietary supercapacitors in FY2027, building a differentiated data center power solution portfolio combining battery-based BBU and capacitor-based CBU”
• Expected Impact
Panasonic’s moves are a concrete example of how the slowdown in EV battery market growth and the explosion of AI data center power demand are forcing corporate strategy pivots.
While the EV battery business swung to a loss this quarter due to one-off costs (¥40B), the data center ESS business grew YoY +44%, becoming the pillar of the company’s overall profitability. More notable is the physical conversion of existing EV battery lines to data center use. The Suminoe plant has already begun converted shipments since April, with the U.S. Kansas plant and Mexico plant also scheduled for sequential conversion. This is a structural decision grounded in a conservative outlook on EV demand.
On the technology side, the battery + capacitor hybrid power solution is the key differentiator. By adding a CBU utilizing proprietary supercapacitors to the existing battery-based BBU, Panasonic aims to secure advantages over competitors in terms of instantaneous power response and longevity. AI servers’ power consumption patterns are characterized by frequent sharp fluctuations, making capacitor characteristics well-suited to the application.
Ultimately, what Panasonic is targeting is a repositioning from a mere battery supplier to an AI data center power infrastructure solutions provider. The guidance to achieve the ESS revenue target a full year ahead of schedule reflects real-world evidence that AI data center power demand is growing even faster than the company’s own projections.
6. Intel CEO Lip-Bu Tan, CNBC Interview — 14A 2029 Volume Production Target, Exploiting TSMC CoWoS Supply Shortage, Customer Demand 3x Requests Despite Tight Supply Chain
• Core Source
“14A is 1.4nm — the most advanced technology available. Honestly, we will enter risk production in 2028 and volume production in 2029. Same timing as TSMC. This is a very significant breakthrough and I’m very excited about it. Multiple customers are already working with us and we have PDK 0.5”
“TSMC’s CoWoS has insufficient production capacity. As a result, we are in a unique position to support this, and it is a very exciting opportunity”
“One customer wanted to triple their volume versus their forecast for this year, but I couldn’t respond in the short term and told them we’d catch up in a few quarters. This is not short-term demand — it’s demand that will persist for years”
“CPU demand is very strong right now. In the past, the GPU-to-CPU ratio in AI training was 8:1, but now, with the spread of inference and agentic AI, it’s shifting to 4:1, 1:1, and even 1:4”
“When you see me increasing capex and buying equipment, that means there are real customers. That is my principle”
• Expected Impact
Lip-Bu Tan’s interview formally articulated Intel’s turnaround narrative with specific numbers and timelines. There are two core messages.
First, TSMC’s CoWoS supply shortage is creating an opportunity for Intel on the foundry side. The argument is that Intel’s proprietary EMIB-T advanced packaging technology positions it to absorb demand waiting for CoWoS. The fact that some customers have already begun paying advance deposits for substrates to secure supply chain access lends credibility to this claim.
Second, the structural shift in CPU demand provides a tailwind for Intel. As the computing paradigm moves from AI training to inference and agentic AI, the CPU:GPU ratio shifting from 1:8 to 1:1 and even 4:1 is a direct structural benefit for Intel’s server CPU business.
That said, while 18A yields are improving, stabilization for volume production still requires time, and the 14A volume production target of 2029 means foundry market competitiveness will be proven or disproven in 2028–2029. The CEO’s logic that continued capex and equipment purchases signal real customer demand is a confidence-building message to the market, but investors will need to validate this against actual yield data and customer qualification milestones.
7. CTEE Analysis: NVIDIA Next-Gen AI Chip Supply Chain Shifts to ‘Long-Term Binding’ Structure — 3–5 Year TSMC Capacity Pre-emption for Feynman Generation, Fab 22 P7 Breaks Ground in Q2
• Core Source
“NVIDIA is pursuing a Lock-in strategy to pre-secure TSMC’s advanced process technology for the Feynman-generation chips, which are slated for mass production after 2028”
“Fab 22 P7 is breaking ground in Q2 of this year. It targets sub-2nm advanced processes and has secured space for up to two additional facility expansions”
“Global Big Tech companies are shifting away from simple supply cooperation to a structure in which they pre-pay to secure 3–5 years’ worth of capacity in advance”
“AI GPU and ASIC substrates have much larger area and more layers than standard CPU substrates, driving a 5–10x surge in ABF material consumption”
“With AI GPU, ASIC, and CPO demand converging simultaneously, there is a possibility that the high-end ABF substrate shortage will recur in 2027”
• Expected Impact
The core of this analysis is that the collaborative structure of the AI chip supply chain is undergoing a paradigm shift from simple orders to an advance reservation model.
Sub-2nm processes and advanced packaging such as CoWoS and SoIC require years just to build out fab lines. Big Tech players who experienced CoWoS and HBM supply shortages are no longer placing orders after demand materializes. Instead, they are shifting to a long-term binding structure in which 3–5 years’ worth of production capacity is secured upfront with prepayment. For TSMC, this means stable revenue visibility; for NVIDIA and other Big Tech players, it creates barriers to entry by making it difficult for competitors to secure the same process capacity.
The ripple effects across the broader supply chain are also significant. For ABF substrates, AI GPU applications require 5–10x more material than standard CPU substrates, and with AI GPU, ASIC, and CPO demand converging simultaneously, the possibility of a high-end ABF substrate shortage recurring in 2027 is being flagged. This is a direct demand driver for Taiwanese ABF substrate makers including Kinsus, Unimicron, and Nanya PCB.
TSMC’s Southern Taiwan Science Park (Nanke) is emerging as the core production hub for NVIDIA’s next-generation AI chip transitions, with Fab 18 (5nm/3nm), Fab 22 P7 (sub-2nm), and the AP 8 packaging facility all clustering in the region. Jensen Huang’s visit to Taiwan on May 27 — where he will host a large dinner, the ‘Trillion-Dollar Banquet (兆元宴)’, with the leaders of TSMC, Foxconn, Delta Electronics, MediaTek, and other key AI supply chain partners — carries the character of a formal ceremony cementing these long-term binding relationships.
Comment [0]