Yaaa News
Tech / Business / Security — the few worth reading each day
iPhone 18 Pro US units keep Qualcomm X80 — Apple's C2 modem ships everywhere else, stuck on mmWave
A 630GB leak from Tata Electronics shows iPhone 18 Pro will ship two logic boards: US units keep Qualcomm's Snapdragon X80; the rest of the world moves to Apple's in-house C2 modem — because C2 doesn't do mmWave.
Eight months after Trump's pardon, CZ hit by 1,692 UK retail investors in London
1,692 UK retail investors sued Binance and CZ in London's High Court on July 1, seeking £150M for unauthorized crypto derivatives sold from late 2019 until the FCA's 2021 ban.
Microsoft to cut ~5,500 jobs — Xbox first as hardware bleeds $500M over five years
Microsoft is set to cut ~5,500 jobs next week — under 2.5% of its 228,000 headcount — hitting Xbox, sales and consulting as FY2026 closes.
South Korea moves to compress the 9–10 year nuclear-plant build cycle to feed AI demand
South Korea's presidential chief of staff Kang Hoon-sik said on June 30 the government will study how to cut the 9–10-year nuclear build cycle, to feed AI data centers and chip fabs. Net AI-driven power demand hits 262 TWh in 2026, 3× the 2023 level.
Microsoft cuts M365 Developer storage 416× — from 125 TB ceiling to a 300 GB tenant cap
Microsoft quietly added a storage clause to the M365 Developer FAQ: 300 GB tenant cap, 10 GB/user OneDrive, 0 GB SharePoint in mixed tenants. Versus the old 25 × 5 TB layout, that's roughly a 416× cut.
Xiaohongshu hit with real-name whistleblower complaint on the eve of its $70B Hong Kong IPO filing
On the eve of Xiaohongshu's $70B Hong Kong IPO filing, a former South-China commercial chief filed a real-name complaint to HKEX and the SFC on June 28, citing labor and option-vesting issues echoed by some 50 ex-employees.
Google's new hand-gesture reCAPTCHA — 21 keypoints, defeated by a stock photo
Google launched a hand-gesture reCAPTCHA on June 19, using 21 MediaPipe hand keypoints for liveness. Neowin bypassed it on first try by holding a stock photo of a hand in front of the camera.
Microsoft IP-blocks free Teams in mainland China from July 28 — 21Vianet enterprise version untouched
Microsoft will IP-block personal Microsoft accounts in mainland China from Teams desktop and web on July 28, 2026 — paid 21Vianet enterprise Teams keeps running, free personal users get cut.
Samsung, SK hynix, Micron hit with US class action — accused of using HBM pivot to rig DRAM prices up 700% in 4 years
Samsung, SK hynix, and Micron — 89% of global DRAM — face a US class action filed June 25 alleging they used the HBM pivot as cover to throttle DDR3/DDR4 supply, driving prices up ~700% in four years.
California puts Claude on every state agency at half price — 4 months after Pentagon flagged Anthropic as supply-chain risk
California signs a first-of-its-kind deal with Anthropic giving every state, city, and county agency Claude at 50% off — four months after the Pentagon flagged Anthropic as a supply-chain risk.
OpenAI's first hardware is a 13-key macro pad for Codex — not Jony Ive's device
OpenAI's first own-branded hardware isn't Jony Ive's mystery device — it's a 13-key Codex macro pad co-made with boutique keyboard shop Work Louder, teased June 29, launching July 15.
China's first ten-figure domestic DRAM contract: CXMT locks Tencent for $2.94B
CXMT signs a 20-billion-yuan ($2.94B) long-term server-DRAM supply contract with Tencent, locking in its biggest domestic buyer days before its STAR Market IPO.
Korea's biggest industrial bet ever: $519B on memory fabs, DRAM output to double in five years
Lee Jae Myung commits 800 trillion won for four memory fabs in southwest Korea, two each for Samsung and SK hynix. DRAM output doubles in five years; Yongin completion pulled forward by 7–12 years.
Fable 5 is back. Mythos 5 came back too — only for US critical infrastructure
Claude Fable 5 is reappearing in the Claude App 17 days after the US pulled it worldwide; Mythos 5 came back first, only for US critical-infrastructure operators.
DeepSeek and Peking University open-source DSpark: 60%-85% per-user speedup at equal throughput, or 51%-661% more throughput at fixed latency
DeepSeek and Peking University co-release DSpark, an open-source inference framework: 60%-85% faster per-user generation at equal throughput, or 51%-661% higher throughput at fixed latency.