IBM this morning published the first experimental demonstration of a sub-1 nanometer transistor — a 0.7nm (7 angstrom) node using a new "nanostack" architecture that vertically stacks and staggers nanosheets in three dimensions rather than scaling them laterally. The chip packs roughly 100 billion transistors into a fingernail-sized die, approximately twice the density of IBM's own 2nm node, and early results show up to 50% more performance or 70% greater energy efficiency versus that prior generation. The manufacturing partners named in the announcement are ASML (High-NA EUV lithography), Lam Research, and Tokyo Electron — the same equipment trio that enables every leading-edge node in production today. IBM is projecting "earliest adoption within approximately 5 years," which positions the nanostack as the technology that carries logic scaling through the 2030s. IBM shares jumped roughly 5% in premarket trading on the news.
RunPod closed a $100M growth round led by Summit Partners at a $1B valuation — reaching unicorn status while declining acquisition offers above $500M. The platform has 1M+ developers, has processed 20B+ inference requests, and crossed $120M ARR before this round with material growth since. Unlike competitors converging on inference-only services, RunPod covers the full AI development lifecycle — training, fine-tuning, inference, and multi-node scaling — at per-second pricing with no commitment minimums. The rejection of multiple buyout offers above $500M is the most telling signal in the round: the founders believe a horizontal AI development platform commands more standalone value than any current acquisition premium, a view that diverges from the direction most neocloud consolidation is running.
Q3 FY2026 beat expectations on every metric: $41.46B in revenue (vs. $35.7B consensus), $25.11 in EPS (vs. $20.49 consensus), and 84.9% gross margin — a company record, up 10 percentage points sequentially. Data center revenue crossed $25B in the quarter, a $100B annualized run rate. HBM4 generated over $1B in revenue and is ramping twice as fast as HBM3E. Micron disclosed 16 customer supply agreements locking in $100B of minimum contractual revenue at floor prices, with $22B in deposits already received. Q4 guidance: $50B revenue, ~86% gross margin, EPS ~$31. CEO Sanjay Mehrotra said supply constraints will persist beyond 2027. The print reverses a 13%+ intraday decline Micron suffered on June 23 during a broad semiconductor sector sell-off.
Broadcom and OpenAI jointly unveiled Jalapeño on June 24 — OpenAI's first custom inference chip, built around OpenAI's LLM architecture and delivered from design to production in nine months. Broadcom CEO Hock Tan says early results show ~50% cost savings versus GPU-based inference. The chip is targeting end-2026 deployment at gigawatt scale across Microsoft data centers, with OpenAI citing plans to expand "in the years ahead." For Broadcom, Jalapeño adds a flagship AI lab to its XPU customer base alongside Google and Meta, reinforcing the company's position as the default design partner for hyperscaler and frontier-lab custom silicon. The stock's +3.4% move on the day reflects growing market confidence that Broadcom's ASIC business is durable and expanding, not a one-customer story.
Apple this morning raised prices on MacBook and iPad — its first formal move to pass higher memory costs to consumers. The MacBook Air (512GB) goes from $1,099 to $1,299; MacBook Pro (1TB) from $1,699 to $1,999; iPad Air (128GB) from $599 to $749. The driver is DRAM: spot prices rose roughly 98% in Q1 2026 and are projected to jump another 58–63% in the current quarter as AI data center demand absorbs supply faster than capacity can be added. TSMC's 5–10% wafer price hikes compound the squeeze. Apple is the bellwether — the world's largest consumer electronics company absorbing and then passing through memory inflation is the signal that the AI-driven memory supercycle has moved from data center procurement into consumer product pricing. Expect Samsung, Lenovo, Dell, and HP to follow.
TSMC notified customers on June 24 of price increases of 5–10% across all advanced nodes — from N7 down through N3 and the upcoming N2 — representing approximately 74% of its total wafer revenue. Every major fabless company is affected: Nvidia, AMD, Qualcomm, Broadcom, Apple, and MediaTek are all cited. This is the third consecutive year of advanced-node price hikes, and the broadest in scope: prior rounds targeted only sub-3nm nodes; this one extends to N7, which remains widely used for mid-performance AI accelerators, networking SoCs, and mobile chipsets. With no credible volume alternative to TSMC for leading-edge logic, fabless companies must embed foundry cost inflation into long-run product margin assumptions. The semiconductor era in which smaller transistors automatically meant cheaper chips is over. (Tom's Hardware · Culpium · TrendForce)