LIVE · FRI, JUN 26, 2026 --:--:-- ET
Issue Nº 66 COST TOTAL $14556.78 ARTICLES TODAY 9 TOKENS TOTAL 9.16B
aiexpert
Running the wire
Research Frontier models saturate GPQA-Diamond benchmark at 93–94%; SWE-bench Pro becomes key differentiator Policy Trump signs AI security executive order; 30-day voluntary prerelease model review Funding Autodesk acquires MaintainX for $3.6B, extending into industrial AI operations Market OpenAI leans toward 2027 IPO, holding firm on $1T valuation floor Funding ON Semiconductor acquires Synaptics for $7B in physical AI bet Market OpenAI shifts IPO target to 2027; Anthropic poised to list first in October at $965B Breaking OpenAI stalls GPT-5.6 public release; US govt to approve access customer-by-customer Funding Onsemi acquires Synaptics for $7B to consolidate Edge AI and Physical AI stack Chips Apple skips M6 Pro/Max, fast-tracks AI-focused M7 to late 2027 as bandwidth race heats up Market OpenAI leans toward delaying IPO to 2027 over market volatility, holding firm on $1 trillion valuation Chips Solidigm ships 122TB SSD with unlimited 5-year random-write endurance; 84% less NAS power than HDD+TLC Breaking Google launches Gemini 3.5 Flash: outperforms Pro tier on coding, 40% cheaper, 4x faster Funding SE3 Labs emerges from stealth with €5.5M for autonomous defense AI, backs German sovereignty Market Anthropic, OpenAI face enterprise cost-cutting as customers switch to cheaper models Research OpenAI Codex agents now primary tool across all departments; 80% of users complete 30+ minute tasks Chips IBM's 0.7nm Nanostack breaks sub-1nm barrier with 100B transistors on fingernail die Chips Nvidia triple-qualifies HBM4 suppliers; SK Hynix, Samsung, Micron all production-ready for Vera Rubin Q3 ship Market Micron Q3 earnings blowout: $41.5B revenue quad, $22B customer contracts lock HBM supply through 2027 Breaking Hugging Face Ships vLLM on HF Jobs: spin OpenAI-compatible LLM endpoint in one command Policy White House EO mandates federal PQC migration by 2030-2031; quantum-safe silicon demand accelerates Research Frontier models saturate GPQA-Diamond benchmark at 93–94%; SWE-bench Pro becomes key differentiator Policy Trump signs AI security executive order; 30-day voluntary prerelease model review Funding Autodesk acquires MaintainX for $3.6B, extending into industrial AI operations Market OpenAI leans toward 2027 IPO, holding firm on $1T valuation floor Funding ON Semiconductor acquires Synaptics for $7B in physical AI bet Market OpenAI shifts IPO target to 2027; Anthropic poised to list first in October at $965B Breaking OpenAI stalls GPT-5.6 public release; US govt to approve access customer-by-customer Funding Onsemi acquires Synaptics for $7B to consolidate Edge AI and Physical AI stack Chips Apple skips M6 Pro/Max, fast-tracks AI-focused M7 to late 2027 as bandwidth race heats up Market OpenAI leans toward delaying IPO to 2027 over market volatility, holding firm on $1 trillion valuation Chips Solidigm ships 122TB SSD with unlimited 5-year random-write endurance; 84% less NAS power than HDD+TLC Breaking Google launches Gemini 3.5 Flash: outperforms Pro tier on coding, 40% cheaper, 4x faster Funding SE3 Labs emerges from stealth with €5.5M for autonomous defense AI, backs German sovereignty Market Anthropic, OpenAI face enterprise cost-cutting as customers switch to cheaper models Research OpenAI Codex agents now primary tool across all departments; 80% of users complete 30+ minute tasks Chips IBM's 0.7nm Nanostack breaks sub-1nm barrier with 100B transistors on fingernail die Chips Nvidia triple-qualifies HBM4 suppliers; SK Hynix, Samsung, Micron all production-ready for Vera Rubin Q3 ship Market Micron Q3 earnings blowout: $41.5B revenue quad, $22B customer contracts lock HBM supply through 2027 Breaking Hugging Face Ships vLLM on HF Jobs: spin OpenAI-compatible LLM endpoint in one command Policy White House EO mandates federal PQC migration by 2030-2031; quantum-safe silicon demand accelerates
Chips

Qualcomm plans Dragonfly data center chips for China, targeting export-compliant AI accelerators

Qualcomm announced that it will bring all four Dragonfly data center product lines to China, including custom AI accelerators engineered to stay below U.S. export control thresholds. CEO Cristiano Amon told Nikkei Asia on Wednesday that Qualcomm is developing export-compliant versions of each product line and is "engaged in conversations" with Chinese customers. The first accelerator, the AI250, is due next year and uses Qualcomm's HBC near-memory design instead of HBM stacks.

Qualcomm told investors the data center unit is expected to generate $300 million in revenue this fiscal year and $5 billion in fiscal 2027, with a projected total addressable market exceeding $1 trillion by 2029. The company is banking on its existing relationships with Chinese phone makers and automakers to extend into data center sales. However, Qualcomm enters a market where export-compliant chip strategies have struggled: Nvidia's H20, built specifically for China, generated only $50 million by late last year, and CEO Jensen Huang has said Nvidia has "zero" China market share.

The venture faces headwinds from Chinese market protection policies. China's regulator opened an antitrust investigation into Qualcomm's Autotalks acquisition in October, and Chinese operators are being pressed by regulators to source at least 50% of chips locally while customers like Alibaba, ByteDance, and Tencent are being steered toward Huawei and Cambricon parts. Huawei's Ascend and Cambricon's accelerators are already scaling production well past current volumes.

For architects evaluating data center silicon outside the U.S.-China sphere, Qualcomm's Dragonfly entry signals the geopolitical stalling of foreign chip supply into China. The delay to fiscal 2027 availability, combined with entrenched domestic competition, suggests export-compliant silicon is unlikely to be a growth lever for any Western vendor in China. Operators should expect the bifurcation of the global data center chip market to deepen.

Sources