devtake.dev

#ai-chips

RSS
Macro photograph of a Samsung-branded DRAM chip soldered to a green circuit board, its part number K4T51163QI-HCE6 clearly legible
Hardware·

Samsung won more than $200 billion of Broadcom chip work through 2030

Samsung signed an MOU with Broadcom worth more than $200 billion through 2030 across memory, 2nm foundry and advanced packaging. Nvidia locked up SK Hynix the same day.

Macro photo of a Micron DDR5 memory module: eight DRAM packages marked 2NA45 D8BNH on a green circuit board above the gold contact pins
Hardware·

Google blamed the Pixel 11 price hike on RAM. A gigabyte now costs $12, up from $2.80.

HBM stacks for AI accelerators outbid the DRAM in phones and laptops. Qualcomm, Roku and Google have all repriced, and IDC sees PC prices up 18.3% this year.

AMD launch slide reading 'Launching AMD Instinct MI455X GPUs' with a data-center photo and specs: CDNA 5, 2nm process node, HBM4, FP4/FP6/FP8, 320B transistors
Hardware·

AMD's Helios rack packs 72 GPUs and 31TB of memory to take on Nvidia

AMD launched the Instinct MI455X and its Helios rack-scale system, tying 72 GPUs and 31TB of HBM4 into one machine to challenge Nvidia's data-center grip.

A technician crouches with a laptop to service a black server rack in a data center, colored network cables running down the side
Hardware·

Nvidia wants to own every chip inside the AI data center

Nvidia already dominates AI GPUs. Now it's pushing to supply the CPU, the network, and the whole rack. Here's who that squeezes and whether buyers benefit.

Rows of supercomputer cabinets in a data hall, a technician working at an open rack
Hardware·

China built the world's fastest supercomputer without a single US chip

A Chinese system called LineShine debuted at No. 1 on the TOP500 at ISC'26, running entirely on domestic CPUs co-designed with Huawei. Here's what that means.

Sam Altman and Broadcom CEO Hock Tan holding a Jalapeño Intelligence Processor wafer mounted in an acrylic display
AI·

OpenAI built its own AI chip with Broadcom. The target is Nvidia's inference margins.

Jalapeño is OpenAI's first custom inference processor, co-designed with Broadcom. Here's what a purpose-built inference ASIC actually buys you, and who else is doing it.

A silicon wafer patterned with IBM's sub-1-nanometer node test chips.
Hardware·

IBM's 'sub-1nm' chip is a lab milestone, not a product you'll buy soon

IBM showed a 0.7nm 'nanostack' transistor with ~100 billion transistors on a fingernail. It's real research, but the name is marketing and shipping silicon is years out.

Benchmark comparison card for GLM-5.2 showing it as the leading open weights model
AI·

GLM-5.2 was trained on Huawei chips, not Nvidia. The open weights beat GPT-5.5 on coding.

Zhipu AI's GLM-5.2 is a free-to-download model trained without Nvidia silicon. Here's what the benchmarks claim and why developers should care.

Close-up of a Tenstorrent Galaxy AI compute unit with a hexagonal vented faceplate
Hardware·

Qualcomm is chasing a $10 billion AI-chip deal. The target is Jim Keller's RISC-V startup.

Qualcomm is in talks to buy Tenstorrent for up to $10 billion, per Reuters and The Information. The bet: a RISC-V challenger to Nvidia, not just phones.

Close-up of a semiconductor wafer, the kind Taiwan's foundries produce for AI servers
Policy·

Taiwan weighs making AI chip exports to all of China a crime, not just to Huawei

Taiwan is considering US-aligned export controls that would treat unauthorized AI chip shipments to any Chinese customer as a crime. It's under discussion, not law yet.

Nvidia RTX Spark laptop shown at the Windows AI PC announcement
Hardware·

Nvidia's RTX Spark laptops run 120-billion-parameter models locally, no cloud needed

Nvidia's RTX Spark laptops put a Grace Blackwell superchip and 128GB of unified memory in a notebook, aimed at running 120B-parameter models offline.

DeepSeek social card with the company's wordmark on a navy background
AI·

DeepSeek locked in the 75% V4-Pro cut. The API now undercuts every Western frontier model.

On May 23 DeepSeek told customers the V4-Pro discount becomes its standard price after May 31. Output drops from $3.48 to $0.87 per million tokens.

An Alibaba booth at a Chinese technology trade expo, with the company's logo above a display floor.
Hardware·

Alibaba's new Zhenwu M890 chip is 3x faster and aimed straight at agent workloads

Alibaba showed the Zhenwu M890 at its Cloud Summit on May 19. 144 GB of memory, 800 GB/s interchip bandwidth, and Qwen3.7-Max riding on top.

Cisco CEO Chuck Robbins speaking on stage.
Hardware·

Cisco cut 4,000 jobs on its record revenue day. AI infrastructure orders jumped to $9 billion.

Cisco posted $15.8B in Q3 FY26 revenue on May 14, up 12% YoY, then sent layoff notices to 4,000 staff the same day. AI orders are now tracking to $9B for the fiscal year.

Cerebras corporate site Open Graph card.
Hardware·

Cerebras priced its IPO at $185 and closed at $311. Andrew Feldman and Sean Lie became billionaires.

Cerebras raised $5.55 billion on May 14 and closed its first day at a $95 billion market cap. The wafer-scale AI chip maker shipped the year's biggest tech IPO.

Elon Musk speaking at an event, with Tesla and SpaceX branding behind him
Hardware·

SpaceX filed for a $55B Texas chip fab. Total Terafab spend could hit $119 billion.

SpaceX filed plans for a $55B Terafab semiconductor facility in Grimes County, Texas. Total spend could reach $119 billion across phases.

AMD CEO Lisa Su on stage at CES 2026, presenting AI accelerator products
Hardware·

AMD's data center revenue jumped 57% to $5.8 billion in Q1. Its stock closed up 18%.

AMD reported Q1 revenue of $10.25B, up 38% YoY. Data center hit $5.8B (+57%) and Q2 guidance midpoint of $11.2B beat consensus by roughly $700M.

John Ternus, Apple's incoming CEO, photographed at Apple Park.
Apple·

Apple dropped its 7-year 'net cash neutral' policy. Ternus is freeing up the balance sheet for AI.

Mark Gurman's May 3 Power On reads Apple's quiet capital-allocation shift as cover for John Ternus to spend more on AI infrastructure and acquisitions, less on buybacks.