Weekly Industry Insights

Jul 13 - Jul 19 Weekly Insight

AI chip supply chain restructuring accelerates with TSMC price hikes and advanced packaging expansion, while NVIDIA and Huawei vie for AI compute dominance, and cybersecurity faces new threats.

Read More →

Jun 22 - Jun 28 Weekly Insight

This week saw diversified developments in AI infrastructure, with vendors competing for control through custom chips, liquid cooling, and vertical integration, while ARM server market share exceeded 45%, accelerating the shift to AI-native architectures.

Read More →

Jun 15 - Jun 21 Weekly Insight

This week shows dual trends of hardware lock-in and software ecosystem restructuring in AI infrastructure, alongside rising global AI model export control risks.

Read More →

All Insights

127articles
TSMC's Additional $100 Billion Arizona Bet: The 'Made in USA' Gamble for AI Infrastructure

TSMC's Additional $100 Billion Arizona Bet: The 'Made in USA' Gamble for AI Infrastructure

TSMC reported Q2 results on July 16: $40.2B revenue, 77.4% net income growth, 67.7% gross margin, and announced an additional $100B Arizona investment, bringing total committed investment to $265B. The new fabs will focus on 2nm and advanced packaging. The company raised 2026 capex to $60-64B. However, shares plunged 7.29% on July 17 as investors questioned AI capex sustainability. This investment will reshape the global semiconductor supply chain geography but faces talent shortages, rising costs, and geopolitical challenges.

Apple vs OpenAI Trade Secret Lawsuit: Silicon Valley's AI Hardware Talent War Heats Up

Apple vs OpenAI Trade Secret Lawsuit: Silicon Valley's AI Hardware Talent War Heats Up

Apple sued OpenAI on July 10 for systematic poaching and trade secret theft, sending legal warnings to approximately 40 former employees on July 17. The lawsuit targets OpenAI Chief Hardware Officer Tang Tan (former Apple iPhone design lead) and engineer Chang Liu, alleging acquisition of confidential information on unreleased products, components, and supplier relationships. Over 400 former Apple employees currently work at OpenAI. This case marks the expansion of Silicon Valley AI competition from software to hardware entry points and will profoundly impact industry talent mobility rules and competitive dynamics.

AMD-OpenAI 6GW Historic AI Chip Agreement: Birth of the AI Compute Closed-Loop Economy and Systemic Risks

AMD-OpenAI 6GW Historic AI Chip Agreement: Birth of the AI Compute Closed-Loop Economy and Systemic Risks

On July 16, 2026, AMD and OpenAI signed a historic 6GW compute supply agreement (equivalent to 450,000 households' electricity consumption), with OpenAI eligible to receive 160 million AMD warrants (approximately 10% equity) at a strike price of only $0.01. AMD shares surged 35% pre-market. The first 1GW will launch in 2026 H2 using MI450 GPUs. This agreement comes shortly after OpenAI's $100B investment with NVIDIA, combined with Broadcom's $10B and Cerebras' $20B, forming OpenAI's 'four-supplier' compute layout, marking the official birth of the 'AI Compute Closed-Loop Economy' (capital-equity-compute three-layer cycle).

Intel Ships World's First High-NA EUV Mass-Produced Chips: The 18A Process Comeback and Foundry Supremacy Battle

Intel Ships World's First High-NA EUV Mass-Produced Chips: The 18A Process Comeback and Foundry Supremacy Battle

On July 15, 2026, Intel announced it became the world's first chipmaker to ship high-volume logic products using ASML High NA EUV lithography, producing Core Ultra Series 3 (Panther Lake) processors on its 18A process node. Each High NA EUV machine costs approximately $400 million, achieving 8nm resolution. Meanwhile, Intel 18A process yield has jumped from 65% to 85%, with Q2 revenue guidance of $13.8-14.8 billion. Although TSMC's Q2 net profit surged 77.4% to NT$706.6 billion, its High NA EUV mass production trails Intel by at least 3 years.

Cloudflare AI Content Paid Gateway: The Technical and Business Logic Behind Infrastructure's First Systematic Charge on AI Crawlers

Cloudflare AI Content Paid Gateway: The Technical and Business Logic Behind Infrastructure's First Systematic Charge on AI Crawlers

In July 2026, Cloudflare officially launched its AI Content Paid Gateway (Pay Per Crawl), charging AI crawlers a minimum of $0.01 per successful content retrieval. As an infrastructure giant handling approximately 20% of global web traffic, Cloudflare's move marks a historic shift in internet content distribution from 'free crawling' to 'paid usage.' The company reported Q1 2026 revenue of $639.8 million, up 34% YoY, with AI-related workloads becoming the core growth engine.

TSMC Q2 2026 Earnings and AI Compute Supply-Side Confirmation: Five Consecutive Quarterly Highs Reshape Semiconductor Pricing Power

TSMC Q2 2026 Earnings and AI Compute Supply-Side Confirmation: Five Consecutive Quarterly Highs Reshape Semiconductor Pricing Power

On July 13, 2026, TSMC announced June single-month revenue of NT$442.68B (~$13.8B), a new historical high, +67.9% YoY; Q2 cumulative $39.6B +36% YoY, 5 consecutive quarterly highs. N3 2026 capacity 100% sold out, CoWoS sold out through year-end. TSMC has notified NVIDIA/Apple/AMD of 5-10% price increases for 3/5/7nm processes, covering 70%+ of revenue. Same day, PSMC announced 45% memory foundry price increase, DRAM shortage to 2027. Hyperscaler 2026 Capex totals $725B +77% YoY, confirming AI compute 'sellers' market' is now established.

Nokia-NVIDIA AI-RAN Commercial Platform: The AI Generational Shift in Telecom Infrastructure and the 6G Prelude

Nokia-NVIDIA AI-RAN Commercial Platform: The AI Generational Shift in Telecom Infrastructure and the 6G Prelude

Nokia and NVIDIA launched the industry's first commercial AI-RAN platform, replacing dedicated communication chips with general-purpose GPUs to achieve 20%-100% spectrum efficiency gains. Commercial launch is scheduled for 2027, marking the historic shift of communication networks from dedicated hardware to AI compute-driven architectures and laying the foundation for the 6G era.

Intel 18A Yield Surges to 85%: The Turning Point from Chip Giant to Foundry Powerhouse

Intel 18A Yield Surges to 85%: The Turning Point from Chip Giant to Foundry Powerhouse

Intel's 18A process yield surged from 65% to 85%, securing major orders from NVIDIA and OpenAI, prompting KeyBanc to raise its price target by 41% to $155. This marks the critical inflection point for Intel Foundry transitioning from strategic losses to scalable profitability, as the global advanced foundry market evolves from TSMC dominance toward a tri-polar structure.

New York 1-Year AI Data Center Moratorium: From Meta 5GW to Hochul's Pause - The Regulatory Turn in AI Capex

New York 1-Year AI Data Center Moratorium: From Meta 5GW to Hochul's Pause - The Regulatory Turn in AI Capex

On July 14, 2026, New York State Governor Hochul signed an executive order imposing a 1-year construction moratorium on all new hyperscale AI data centers with power capacity exceeding 50MW, effective immediately. This is the first state-level AI data center ban in the US, marking the divergence between 'AI Capex Nationalization' and 'AI Capex Regulation.' Background: NYISO interconnection queue grew 11x in 3 years (1,045MW to 12,000MW); at least 11 states are considering similar legislation. Direct impact: $15-25B GDP loss in NY within 1 year + 5,000-15,000 construction jobs delayed.

NVIDIA Halves Asian AI Chip Customers: Export Control Whitelist Technical Implementation and the Competitive Window for AMD, Intel

NVIDIA Halves Asian AI Chip Customers: Export Control Whitelist Technical Implementation and the Competitive Window for AMD, Intel

On July 14, 2026, NVIDIA announced reducing authorized AI chip customers in Asia by more than half, establishing a whitelist system in Singapore, Malaysia, and Japan. Only buyers passing strict compliance reviews can qualify for purchases. This move directly responds to Washington's policy to prevent advanced US technology from illegally flowing to China. NVIDIA's review process now includes thorough examination of customer ownership structures and end-use purposes, with monitoring of data center deployments to ensure compliance.

TSMC Raises Advanced Process Prices by 5-10%: The Pricing Power Behind Foundry Tech Monopoly and the Survival Race of Samsung, Intel

TSMC Raises Advanced Process Prices by 5-10%: The Pricing Power Behind Foundry Tech Monopoly and the Survival Race of Samsung, Intel

TSMC's June 2026 revenue reached NT$442.68 billion, up 67.9% YoY to a record monthly high. The company notified core customers including NVIDIA, Apple, and AMD of planned 5-10% price increases for 3nm, 5nm, and 7nm processes, covering over 70% of foundry revenue. This price hike reflects TSMC's irreplaceable tech monopoly in advanced processes and the capacity scarcity driven by AI chip demand explosion. Samsung and Intel Foundry still face significant gaps in yield and capacity at 2nm/3nm nodes.

Indirect Prompt Injection Across 26 Models: AI Agent "Obedience Deficit" and Enterprise Defense Reconstruction

Indirect Prompt Injection Across 26 Models: AI Agent "Obedience Deficit" and Enterprise Defense Reconstruction

Three independent research groups from ETH Zurich, Warwick, and AutoDojo conducted empirical IPI testing across 26 mainstream LLMs, revealing: model scale ≠ safety. GPT-5 achieves ~95% resistance, Claude Sonnet 4.6 reaches 98.8%, but Gemini-2.5-Pro was deceived by a $3 payment attack, and Qwen3-4B showed 67% ASR in Slack scenarios. Static '0% ASR' defenses were breached to 28% against adaptive attackers. ARGUS causal tracing reduced ASR from 28.8% to 3.8%, while DPO on-policy training improved Llama resistance from 22.2% to 84.4%.