1. Apple·Intel Foundry Preliminary Agreement, Intel Stock Surges +14% — Breaking TSMC’s Monopoly and the Dawn of U.S. Semiconductor Supply Chain Diversification
• Core Source
“Apple·Intel, preliminary agreement reached for Intel Foundry to manufacture chips for specific Apple devices”
“The two companies have been in intensive negotiations for over a year, reaching a formal agreement in recent months”
“Immediately after the news, Intel’s stock surged approximately 14% to an all-time high (up over 200% year-to-date), with Apple’s stock also rising 2%”
“Apple, virtually 100% dependent on TSMC — a structural vulnerability in its single-source supply chain”
“Intel expanding its big tech foundry customer base — NVIDIA: $5 billion investment and chip collaboration, Tesla: Terafab project collaboration, with Apple’s addition, three of America’s leading big tech companies secured”
• Expected Impact
The essence of this agreement has two dimensions: structural resolution of supply chain risk for Apple, and proof of foundry business viability for Intel.
Apple currently sources nearly all of its advanced chip production from TSMC. However, with TSMC’s advanced process capacity (CoWoS, N2, etc.) approaching saturation amid surging AI demand, a single-source supply structure has become both a procurement risk and a source of weakened negotiating leverage. By securing Intel Foundry as an alternative, Apple can restore leverage in pricing negotiations with TSMC, while also gaining a credible response to the Trump administration’s pressure for domestic production.
For Intel, this agreement represents securing the most powerful reference customer for its foundry business (IFS). Since Pat Gelsinger’s departure, the market has questioned the direction of IFS under new management. By attracting Apple — one of the world’s largest fabless customers — Intel gains external validation of its manufacturing capabilities. Intel’s 18A process, featuring backside power delivery (PowerVia) and RibbonFET gate-all-around technology, is currently assessed as competitive with TSMC’s N2 on spec, and this agreement provides a genuine opportunity to validate production yields. With NVIDIA ($5 billion investment), Tesla (Terafab), and now potentially Apple, three of America’s leading big tech companies could be secured as Intel Foundry customers.
On a macro level, this signals the emergence of a “Pax Silica” supply chain alliance under U.S. government leadership. The government has acquired approximately 10% of Intel’s shares, and Commerce Secretary Howard Lutnick has directly approached Tim Cook and Jensen Huang to encourage Intel partnerships. As TSMC dependence has emerged as a geopolitical risk, increasing the share of U.S.-made chip production has become a structural policy direction. The critical variable to watch is whether Intel’s 18A process can achieve stable mass production yields. While skeptics point to cost and yield competitiveness still lagging TSMC, if this agreement is finalized, it could mark the inflection point at which global foundry competition transitions from TSMC’s de facto monopoly to a duopoly structure.
2. xAI Colossus 1 (220,000 GPUs) Fully Leased to Anthropic, Converting to $5–6 Billion Annual Revenue — AI Computing Power Realignment and SpaceXAI IPO Strategy
• Core Source
“xAI deployed over 220,000 NVIDIA GPUs in the Memphis Colossus 1 data center. Approximately 150,000 are estimated to be H100s, 50,000 H200s, and 20,000 GB200s”
“The most important detail of the Colossus 1 lease is that it is for inference, not training. Unlike training, inference requires far less tight synchronization between GPUs”
“What was a ‘cluster from hell’ from a training perspective has been redeployed for inference, becoming a ‘golden goose’ generating $5–6 billion annually”
“The annual $5–6 billion in revenue from leasing Colossus 1 to Anthropic effectively hedges xAI’s losses almost perfectly”
“For the SpaceXAI IPO — currently cited as early as June 2026 at a $1.75 trillion valuation — this cloud business model will be the core of the pricing narrative”
• Expected Impact
The essence of this deal is reversing the profit and loss structure through asset redeployment. xAI’s Colossus 1 was a heterogeneous architecture mixing H100s, H200s, and GB200s. In distributed training, the entire cluster must synchronize to the slowest GPU — the “straggler effect” — resulting in a reported GPU utilization rate (MFU) of just 11%, starkly contrasting with Meta and Google’s 40%+. It was effectively an asset with extremely low efficiency for training purposes.
However, the situation reverses for inference workloads. Inference does not require tight synchronization between GPUs, nearly neutralizing the weaknesses of a heterogeneous architecture. With Anthropic occupying all 220,000 GPUs as a single tenant, network jitter issues from multi-tenancy also disappear. For xAI, leasing out the low-training-efficiency Colossus 1 rather than selling it — while converting it into rental income — and concentrating its own model training on the all-Blackwell homogeneous Colossus 2, represents optimal resource allocation.
Financially, annual rental income of $5–6 billion effectively offsets xAI’s annual net losses, transforming the IPO narrative from “an AI lab burning cash” to “an infrastructure leasing operator stably generating $6 billion per year.” Ahead of the SpaceXAI IPO cited as early as June 2026 at a $1.75 trillion valuation, this deal functions as a critical financial defense line that fundamentally changes how the company is valued.
On a macro level, this demonstrates that the AI computing market is evolving from model competition to competition over ownership and leasing rights of infrastructure assets. The market is now realigning into the OpenAI-MS-Oracle camp versus the Anthropic-AWS-SpaceXAI alliance, entering a phase where those who control the chokepoints of physical resources — GPUs, HBM, power, land — become the primary beneficiaries of AI revenue structures.
3. Akamai Pivots from CDN to Edge AI Inference Infrastructure with Anthropic’s 7-Year, $1.8 Billion Deal — A Signal of Structural Realignment in Edge Computing
• Core Source
“Anthropic committed $1.8 billion over 7 years in a Cloud Infrastructure Services agreement, the largest deal in Akamai’s 28-year history, with revenue recognition starting at $20–25 million per quarter from Q4 2026”
“Q1 cloud infrastructure revenue grew 40% year-over-year to $95 million, security revenue grew 11% to $590 million, while legacy CDN revenue declined 7%”
“At NVIDIA GTC, Akamai announced the deployment of thousands of RTX PRO 6000 GPUs and the global implementation of ‘NVIDIA AI Grid,’ formalizing its transition from a CDN company to an AI inference infrastructure provider”
“Akamai is actively pursuing expansion in the edge data center-based AI inference market”
• Expected Impact
Akamai is a CDN (Content Delivery Network) company with over 4,400 edge nodes worldwide. This contract with Anthropic is the starting gun for converting that infrastructure into dedicated AI inference computing resources. Notably, Akamai’s Q1 results showed cloud infrastructure revenue growing 40% year-over-year while legacy CDN revenue declined 7% — the center of gravity in its business model is already shifting.
The structural significance of this deal in the AI inference market is the emergence of a genuine alternative to centralized hyperscaler infrastructure. Major clouds like AWS, Azure, and GCP face physical limitations in simultaneously achieving broad bandwidth and low-latency processing. Performing inference at edge nodes geographically close to users dramatically reduces latency, making this the core infrastructure for applications requiring low latency — real-time AI agents, conversational AI, and autonomous systems.
Akamai has already deployed thousands of RTX PRO 6000 GPUs across its edge nodes and announced collaboration with NVIDIA AI Grid at GTC. Anthropic’s $1.8 billion commitment secures stable long-term demand for this infrastructure, with quarterly revenue recognition of $20–25 million beginning in Q4 2026.
From an investor perspective, this deal is a revaluation trigger for CDN companies broadly. Cloudflare also positions edge AI inference as a core growth strategy, and as AI inference demand grows, a division-of-labor structure between centralized cloud and distributed edge infrastructure is forming. Akamai’s stock surging 27% after the announcement signals that the market is beginning to recognize the business model transformation potential of CDN companies.
4. Cloudflare Declares Agentic AI Transformation with 1,100 (20%) Layoffs — The Competitive Landscape Shift from a Structural Pivot to an AI-Native Enterprise
• Core Source
“Cloudflare announced a reduction of approximately 20% of its workforce. Management stated that as AI agent usage has significantly increased internally and productivity has expanded, it will not hire more staff than necessary”
“CEO Matthew Prince: ‘Over the past three months, internal AI usage has surged over 600%, with AI workflows becoming the operating model across engineering, HR, finance, and marketing'”
“Q1 revenue of $639.8 million grew 34% year-over-year, beating market expectations, but after-hours stock plunged due to weak next-quarter guidance and $140–150 million in restructuring costs”
“This sweeping transformation appears to stem from management’s confidence in rapidly achieving sales model optimization through AI agents”
• Expected Impact
Cloudflare’s layoffs are not simple cost-cutting. As management explicitly stated, this is a declaration that AI agents have replaced human work, rendering those roles unnecessary. Internal AI usage surged over 600% in three months, with AI workflows becoming the actual operating model across engineering, HR, finance, and marketing. The restructuring costs accompanying the 1,100 (20%) layoffs total $140–150 million.
The critical point is that this is not a layoff driven by poor performance. Q1 revenue grew 34% year-over-year, beating market expectations. In other words, a growing company has voluntarily redesigned its workforce structure due to AI — one of the first major cases of its kind. While the stock plunged due to weak next-quarter guidance, RBC maintained an Outperform rating (target $240), and Truist maintained a Buy (target $225).
The key investor question is whether Cloudflare can demonstrate actual execution capability in its AI-native transition. Uncertainty exists given precedents of companies like Klarna declaring similar AI-native pivots only to face setbacks. However, if successful, two simultaneous paths to profitability improvement open: reduction in labor cost structure + capture of AI inference infrastructure demand. Alongside Akamai as a leading player in edge AI inference infrastructure competition, if Cloudflare successfully completes this transformation, it sets a precedent showing that the cost structure of software companies in the AI era can be fundamentally redesigned.
5. Structural Shift in Memory Supply — Big Tech Offers to Directly Fund SK Hynix Dedicated Lines and Negotiate 30–40% Upfront Payments, AI-Driven Memory Capacity War Emerges
• Core Source
“It has been reported that major big tech companies are proposing to fund SK Hynix’s dedicated memory production lines and provide financing for ASML EUV equipment purchases to secure memory supply priority”
“Available production capacity is reportedly at virtually zero, and in some long-term supply contracts, customers are discussing prepayment of 30–40% of total costs as upfront payments”
“Micron CEO admits the company can only supply 50–66% of demand”
“HBM sold out through 2027 — multi-year price and volume contracts completed”
“AI restructuring from ‘compute first’ to ‘memory first’ — HBM TAM forecast at approximately $100 billion by 2028”
• Expected Impact
Big tech companies directly funding dedicated production lines at memory manufacturers and prepaying 30–40% upfront represents the strongest evidence yet that the memory semiconductor market has broken out of its traditional demand-supply cycle and entered a structural shortage phase. This is unprecedented in the history of the semiconductor industry.
HBM is already sold out through 2027, with multi-year price and volume contracts completed. Micron’s CEO publicly admitted the company can only supply 50–66% of demand. The structural root of this shortage is the sharply increased memory dependency of AI accelerators. AI model computation is ultimately constrained by how quickly data can be read from memory, and this bottleneck has emerged as the key constraint on AI infrastructure expansion. The market now characterizes this as AI restructuring from “compute first” to “memory first”.
Big tech’s direct funding of dedicated lines represents the emergence of a new supply chain structure where memory companies share investment risk with their customers. For SK Hynix, this reduces the financial burden of capacity expansion while securing stable long-term demand. For customers, it is a trade-off of deploying liquidity upfront in exchange for supply certainty.
The HBM TAM (total addressable market) is projected to grow to approximately $100 billion by 2028. SK Hynix currently holds dominant market share with Micron in pursuit, while Samsung Electronics’ HBM4 customer certification status remains the key variable in future competitive dynamics. In the near term, structurally strong memory pricing is expected to persist, but Kiwoom Securities forecasts that commodity memory (DRAM/NAND) price growth will slow to single digits from Q3 onward, suggesting an intensifying bifurcation between HBM and commodity memory.
6. ByteDance CY26 AI Capex 200 Billion Yuan (+25%) · DeepSeek Pursues Up to 50 Billion Yuan Fundraise — China’s AI Infrastructure Investment Surge and Global GPU Supply Implications
• Core Source
“ByteDance is pursuing expansion of its CY26 CapEx to over 200 billion yuan (approximately $30 billion), at least 25% higher than the existing plan of 160 billion yuan”
“DeepSeek reportedly pursuing fundraising of up to 50 billion yuan ($7.35 billion)”
“ByteDance’s AI service ‘Doubao’ saw token consumption surge from 60 trillion per day at end of last year to 120 trillion per day in early April this year”
“The token price of Chinese AI models is only one-third to one-twentieth of international competitors”
“According to IDC data, the share of domestically produced chips in China’s cloud AI accelerator market reached 41% in 2025, with Huawei leading at 812,000 chips shipped”
• Expected Impact
ByteDance’s CY26 AI Capex expanding 25% above plan to 200 billion yuan (approximately $30 billion) represents the largest AI infrastructure investment by a single Chinese internet company on record. Two factors underpin this expansion. First, Doubao’s token consumption doubling in just half a year (60 trillion → 120 trillion per day) validates real demand for its own AI services. Second, Chinese AI models including DeepSeek have driven token prices to as low as one-twentieth of overseas competitors, accelerating actual market penetration.
From a global GPU supply perspective, the implications of this investment expansion are substantial. In an environment of tightening U.S. semiconductor export controls on China, ByteDance is simultaneously increasing the share of domestically produced AI chips. China’s domestically produced chip share in the cloud AI accelerator market already reached 41% in 2025, with Huawei leading at 812,000 chips shipped. Export controls are paradoxically accelerating the cultivation of China’s domestic AI chip ecosystem.
DeepSeek’s pursuit of up to 50 billion yuan in fundraising — which, if completed, would be the largest single fundraising in Chinese AI company history — signals that capital deployment in China’s AI model competition has reached a global scale. The core use of proceeds is expanding AI infrastructure resources. Chinese AI models are already rapidly spreading to Southeast Asia, the Middle East, and Latin America through an open-source strategy, recording 2.11 times the weekly token volume of U.S. AI models on OpenRouter for two consecutive weeks to claim the global top position.
From an investor perspective, this trend confirms that AI infrastructure demand is not a U.S.-only cycle, while simultaneously representing an additional demand factor that tightens global GPU supply. However, local media reports that ByteDance’s net profit declined over 70% in 2025 (the company claims growth on an options-cost-excluded basis), and the sustainability of the investment expansion warrants monitoring.
Comment [0]