Nvidia's Q2 FY2027 results show surging revenue and massive memory procurement commitments as AI demand accelerates.

OpenAI revealed its first in-house AI accelerator, co-developed with Broadcom, claiming superior efficiency and latency for inference versus Nvidia's flagship.

A reported acquisition would strengthen Nvidia's open-model AI strategy and position against rivals at a price exceeding 80x Hugging Face's revenue.

Igor Arsovski detailed Groq 3 LPX rack architecture at Hot Chips, with LP30-based racks already in production.

Intel's next-gen AI accelerator targets maximum AI FLOPS per watt with larger caches and deeper XMX engines for inference workloads.

Nvidia's custom HBM implementation targets higher bandwidth and lower power for NVLink Fusion partner program chips.

Arm revealed details of its 136-core AGI server processor with 12-channel memory controller but did not disclose performance figures.

Nvidia emphasized power as the hard limit on data center capacity, showcasing Vera Rubin NVL72 compute within fixed facility power budgets.

Intel's budget laptop chip leverages UCIe interconnect and the company's latest 18A manufacturing node.

Samsung detailed the industry's first LPDDR5X-PIM at Hot Chips, adding logic directly to memory to accelerate data-intensive workloads.

Fujitsu detailed its 144-core Arm-based Monaka server CPU at Hot Chips, confirming narrower 256-bit SVE2 vector units versus its A64FX predecessor.

d-Matrix presented Raptor, a 3D DRAM accelerator for generative inference with a TSMC 4nm compute die bonded face-to-face on custom DRAM.

Hot Chips 2026 coverage highlights challenges with thinner dies, TSV area, thermal dissipation, and limited manufacturing capacity for HBM.

An environmental group is suing the EPA claiming two new chemicals approved for semiconductor manufacturing lack adequate safeguards.

A proposed EPA change would allow air pollution permits without public notice, potentially reducing pushback on data center construction.

Bipartisan resistance to data centers is driven by tangible local environmental impacts rather than beliefs about AI.

Backside power delivery and design-technology co-optimization enable new transistor structures for continued scaling.

OXMIQ presented High Bandwidth Flash use-case scenarios at Hot Chips, narrowing the circumstances where HBF makes sense.