1. Fracture in the OpenAI–Apple Partnership — Apple Adopts Google Gemini and Anthropic Claude in Succession, Weakening OpenAI’s Exclusive Position as iOS 27 Explores Multi-Model Integration
• Core Source
“Apple has signed a contract to use Google’s Gemini model for Siri’s overhaul. Anthropic’s Claude is being actively utilized for internal coding and work support purposes”
“Apple is also considering expanding the integration of external AI models such as Google and Anthropic into Apple Intelligence based on iOS 27. The atmosphere of OpenAI’s exclusive position weakening”
“OpenAI determined that ChatGPT functions within iOS were restricted and that Apple was not sufficiently proactive in expanding the partnership”
“Conversely, OpenAI is actively recruiting Apple hardware engineers to pursue the development of AI device product lines. Inside Apple, there is growing dissatisfaction and concern over talent drain from these moves”
• Expected Impact
The ChatGPT integration agreement announced in 2024 was a strategic partnership through which OpenAI expected to drive large-scale subscriber inflows via the iOS ecosystem. However, the results fell far short of expectations. Apple has signed a formal agreement to use Google Gemini for Siri’s upgrade and is actively adopting Anthropic’s Claude for internal coding and business tools, meaning OpenAI’s exclusive position has effectively entered a process of dissolution. With iOS 27 now exploring the integration of multiple external AI models into Apple Intelligence, this could mark a pivotal moment in which the competitive landscape of the AI model market shifts from “OpenAI exclusivity” to a “multi-vendor” structure.
From an investor perspective, the core of this shift is the intensification of distribution channel competition. Hundreds of millions of iPhones worldwide represent the most powerful deployment platform for AI model providers. If OpenAI is excluded from this channel or loses market share, Google and Anthropic will fill that space, accelerating the reshuffling of market share in consumer AI. Google secures a dual-benefit structure by supplying Gemini as the core engine of Siri while connecting it to its advertising and cloud ecosystems, while Anthropic gains a foothold to expand beyond the enterprise (B2B) market into Apple’s consumer (B2C) channel.
OpenAI’s counterstrategy also warrants attention. OpenAI is recruiting large numbers of Apple hardware engineers and moving toward developing its own AI devices. This is a response reflecting an awareness of platform dependency risk, and if OpenAI succeeds in building its own hardware-software ecosystem, the competitive relationship with Apple could completely replace any cooperation. The evolution of the relationship between AI model providers and device manufacturers from cooperation to competition is a signal that the race toward vertical integration across the entire AI infrastructure value chain is intensifying.
2. Applied Materials (AMAT) — Record Quarterly Revenue, Margins, and EPS Signal That the AI Semiconductor Equipment Cycle Is Just Beginning, with Customer 8-Quarter Rolling Demand Forecasts Securing Unprecedented Visibility
• Core Source
“Q2 revenue of $7.91 billion exceeded consensus ($7.65 billion), and semiconductor systems segment revenue reached a record high of $5.97 billion, up 16% quarter-over-quarter. Gross margin reached 50.0%, the highest level in 25 years, and EPS of $2.86 exceeded consensus ($2.68)”
“The July quarter guidance of $8.95 billion (midpoint) far exceeded consensus of $8.15 billion”
“Semiconductor systems revenue growth is expected to exceed 30% in CY26 (raised from prior 20%+ guidance), and management explicitly referred to CY27 as ‘another powerful record year.’ Advanced packaging revenue is estimated to surpass $2 billion in CY26, growing more than 50%”
“As major customers have begun providing 8-quarter rolling demand forecasts, AMAT has secured the longest and clearest demand visibility in its history”
• Expected Impact
In the semiconductor equipment sector, demand visibility is the most critical variable for investment decisions. The shift from the prior norm — where customers provided only 1–2 quarters of demand forecasts — to 8-quarter (2-year) rolling demand forecasts represents a structural change. It directly demonstrates that TSMC, Samsung, Micron, and other leading semiconductor manufacturers view AI demand not as a temporary cycle but as long-term structural growth.
The performance figures corroborate this. A gross margin of 50.0% in the April quarter — the highest in 25 years — shows that AMAT is enhancing pricing power in the high-value-added process equipment market, beyond mere volume increases. Particularly significant is the estimate that advanced packaging revenue will grow more than 50% in CY26 to exceed $2 billion. Demand for CoWoS, HBM, and 2.5D/3D packaging processes — all essential for AI accelerators — is surging independently of traditional logic and memory equipment, and AMAT is simultaneously expanding share across GAA transitions, advanced DRAM, and HBM processes.
Semiconductor equipment sits at the very top of the AI infrastructure investment value chain. With big tech data center CAPEX projected to exceed $918 billion in 2027 and $1 trillion in 2028, AMAT is capturing that capital as it translates into actual fab equipment. Given that equipment stocks tend to lead semiconductor cycle recoveries, AMAT’s dramatic guidance upgrade should be read as a positive signal for the broader semiconductor supply cycle ahead.
3. TSMC Declares the Era of AI Optical Communications — COUPE Technology Targets 200Gbps Optical Modulator Mass Production This Year, Delivering up to 4x Power Efficiency and 90–95% Latency Reduction vs. Copper
• Core Source
“TSMC plans to mass-produce the world’s first 200Gbps optical modulator (MRM) using its COUPE technology this year. COUPE is an ultra-high-speed, low-power optical communications technology for AI data centers, and TSMC emphasized that COUPE will become the next most prominent semiconductor keyword after CoWoS. This represents up to 4x improvement in power efficiency and up to 90–95% reduction in latency compared to conventional copper wiring”
“TSMC has already received approximately 25 confirmed 2nm product designs, with more than 70 customer designs currently in planning or in progress”
“TSMC projects that AI accelerator demand in 2026 will be 11 times higher than in 2022. The global semiconductor market size forecast for 2030 has also been raised from $1 trillion to $1.5 trillion”
• Expected Impact
The way data moves within AI data centers is changing. Until now, copper wiring has been the standard means of connecting chip to chip. However, as AI model scale grows and inference speed requirements intensify, the power loss and latency generated by copper interconnects have emerged as performance bottlenecks. TSMC’s COUPE — entering mass production this year — addresses this problem with optical signals. The figures of up to 4x power efficiency improvement and up to 90–95% latency reduction compared to copper represent not merely a performance upgrade but a fundamental shift in data center design paradigm.
What matters is that TSMC is positioning this technology not as a simple component supply but as an integrated platform combining packaging, photonics, and logic. Just as CoWoS became the standard process for AI semiconductors by combining HBM and GPUs in a single package, COUPE has the potential to become the standard platform for next-generation AI accelerators by integrating optical communications at the semiconductor process level. TSMC’s own statement that “COUPE will become the next most prominent semiconductor keyword after CoWoS” is a declaration that it intends to make this technology the core of its medium-to-long-term business strategy, not a short-term product.
On the demand side, TSMC’s own projection that AI accelerator demand in 2026 is 11 times higher than in 2022 provides the backdrop. The upgrade of TSMC’s 2030 global semiconductor market forecast from $1 trillion to $1.5 trillion reflects the same conviction. The disclosure that 25 confirmed 2nm process designs are in place with more than 70 additional designs under review demonstrates that demand for TSMC’s most advanced processes is already locked in for the long term. As COUPE photonics, 2nm logic, and CoWoS packaging converge, TSMC’s process monopoly is set to deepen further.
4. NAND Market Enters Supply Deficit Cycle Driven by AI Server Demand — Kioxia Operating Profit up 93% YoY, NAND Prices Double in the March Quarter
• Core Source
“The company explained that NAND prices denominated in U.S. dollars doubled during the March quarter, and projected that NAND memory supply shortages will continue through fiscal year 2027”
“Annual operating profit is projected at ¥870.3 billion, up 93% year-over-year. Net profit is expected to reach ¥554.4 billion, significantly higher than the prior year’s ¥272.3 billion, with annual revenue guided at ¥2.34 trillion, up 37% year-over-year”
“Next quarter guidance: revenue ¥1.75 trillion (+74.5% QoQ), operating profit ¥1.3 trillion (+117.0%), OPM 74.3%”
“AI DC/Enterprise NAND demand strongly driven by: 1) continued general server replacement demand, 2) surge in inference cache demand, 3) increase in high-capacity/high-spec QLC SSD demand due to Nearline HDD shortages”
“The supply-demand imbalance is expected to continue through 2027”
• Expected Impact
The NAND market has passed through a prolonged oversupply and price decline cycle and has now entered the opposite phase. Kioxia’s results quantify the speed and magnitude of that transition. Dollar-denominated NAND prices more than doubled quarter-over-quarter in the March quarter, and the next-quarter operating margin is guided at 74.3%. This is not a simple demand recovery — it is the result of structural and sustained excess demand created by AI data centers.
AI is affecting the NAND market simultaneously through three channels. First, large language model (LLM) inference servers consume large volumes of high-spec NAND as KV cache storage to deliver fast response times. Second, the explosive growth of AI data centers is simultaneously stimulating general server replacement demand. Third, Nearline HDD supply shortages are driving additional surges in high-capacity QLC SSD demand as a substitute. While all three channels operate simultaneously, new capacity additions across the industry remain limited, meaning the supply-demand imbalance is projected to persist at least through 2027.
NAND memory manufacturers are prioritizing profitability maximization over supply expansion. With Kioxia holding its annual CAPEX to approximately ¥450 billion while posting a 93% year-over-year surge in operating profit purely on price increases, this Capex discipline suppresses supply growth and prolongs the price strength phase. The trend of significantly upward-revised earnings estimates for memory companies including SK Hynix and Samsung Electronics reflects this structural shift.
5. Mobile DRAM Price Surge Disrupts the Smartphone and PC Supply Chain — LPDDR5X ASP Forecast to Rise 78–83% Quarter-over-Quarter, Smartphone Production Cuts Now a Reality
• Core Source
“LPDDR4X ASP is expected to rise 70–75% quarter-over-quarter”
“LPDDR5X ASP is expected to rise 78–83% quarter-over-quarter”
“Smartphone brands are facing increasing cost pressure from DRAM price surges. The possibility of smartphone production volume reductions in 2026 has been raised”
“High-end models are being reconfigured around 12GB. 16GB adoption is declining. Mid-range devices are shifting toward 8GB, and entry-level toward 4GB”
“(PC) Overall shipment volumes are forecast to decline due to rising BOM cost pressure from memory price increases”
• Expected Impact
The memory supply-demand imbalance created by AI servers is now rippling through the consumer electronics market. TrendForce’s Q2 mobile DRAM contract price increase of 78–83% quarter-over-quarter for LPDDR5X and 70–75% for LPDDR4X is sufficient to raise the DRAM component cost share in a smartphone’s BOM by tens of percent in a short period.
Smartphone manufacturers have already begun responding. Flagship model memory configurations are being scaled back from 16GB to 12GB, while mid-range and entry-level devices are shifting to 8GB and 4GB respectively. This directly conflicts with the long-term direction of the smartphone industry — increasing memory capacity to support expanded on-device AI features. In other words, until the price pressure eases, the mass-market deployment of AI smartphone capabilities is likely to be delayed.
The PC market faces the same dynamics. Rising BOM costs from memory price surges are driving forecasts for overall PC shipment declines, with ODMs and brand manufacturers conservatively revising their second-half shipment plans downward. This demand deferral paradoxically reinforces the cycle of supply being concentrated toward AI server DRAM. For memory manufacturers, the optimal strategy is to concentrate shipments on high-margin AI server applications while managing supply to consumer segments to maintain pricing power — and this is precisely the approach being pursued. As a result, mobile DRAM price strength should be understood not as a short-term phenomenon but as a structural reality driven by AI demand fundamentally reshuffling memory production priorities.
6. AI Data Center Power Demand Surge Drives U.S. Largest Grid Electricity Prices Up 76% — PJM’s 13-State Wholesale Power Cost Jumps from $77.78 to $136.53 per MWh
• Core Source
“In Q1 2026, driven by surging AI data center power demand, electricity prices in the PJM Interconnection region — the largest power grid in the United States — soared 76% year-over-year”
“Monitoring Analytics, the independent market monitor for the power market, reported that the average wholesale electricity cost across PJM’s 13-state grid in Q1 2026 was $136.53 per MWh — a significant increase from $77.78 per MWh in the same period of 2025”
“As AI data center expansion generates massive power consumption, the burden on the aging U.S. power grid is rapidly increasing, and upward pressure on electricity rates and the challenge of easing the burden on consumers have emerged as core issues for grid operators”
• Expected Impact
PJM Interconnection is the world’s largest grid operator, serving approximately 65 million people across 13 eastern U.S. states and Washington D.C. The fact that wholesale electricity prices in this region surged 76% in a single year — from $77.78 to $136.53 per MWh — represents the first official statistical evidence that AI data center power consumption has begun materially reshaping the pricing structure of the U.S. electricity market as a whole.
For investors, the core significance of this issue is the emergence of a new variable in the AI CAPEX discussion. Until now, the debate around AI infrastructure investment has centered primarily on the semiconductor value chain — GPU supply, data center construction costs, and memory demand. However, as electricity prices begin rising structurally, data center operating costs (OPEX) rise in tandem, directly pressuring the per-unit profitability of big tech’s AI services. Hyperscalers are already making power availability the top criterion for data center site selection, and power infrastructure is emerging as the new bottleneck in AI scaling.
The beneficiary structure is clear. Rising electricity prices and demand for grid upgrades translate into direct tailwinds for power generation equipment (gas turbines, nuclear, solar), transmission and distribution infrastructure, large-scale energy storage systems (ESS), and power equipment manufacturers including transformer and switchgear companies. In particular, companies that have secured dedicated power purchase agreements (PPAs) for data centers or that own their own generation assets can avoid price volatility exposure while strengthening competitive advantage. Conversely, data center operators that depend on the public grid without secured power face structural cost pressure from rising operating expenses. This data point is the first to quantitatively demonstrate that power procurement capability is emerging as a new competitive moat in the AI infrastructure investment debate.
Comment [0]