XPU.news

TechInsights grew 17% in 2025 and is on track to grow by 20% this year. The 2025 growth came while holding costs constant, enabling top-line dollars to flow directly to the bottom line. In dollar terms, half of the growth came from custom projects, with the remainder coming from subscriptions. Both…

00:00 Introduction and episode overview 01:06 NVIDIA's NV-HBM and custom memory interface 03:16 Moving memory controllers to base die and implications 06:31 Astera's second-generation CXL memory controllers 09:17 Are CPUs made for agentic AI just hype? 13:55 Is local AI worthwhile? 15:06 What is th…

Arm has squeezed more performance from its Lumex Ultra CPU; however, Arm’s new GPU advances the company’s mobile platform further by implementing significant hardware changes. At the same time, Arm has introduced the Neoverse N4 for servers and high-end embedded chips. Because Arm’s smartphone-chip…

The Byrne-Wheeler Report needs to bump up its watch time, and you can help. Please check out the channel and play an episode all the way through. At least we don't ask you to like, comment, and subscribe. Not that we would stop you.

At Hot Chips 2026, companies unveiled new processors and discussed related topics, such as memory and networking. Large companies dominated the agenda, and several had multiple speaking slots. Below are observations from the conference, covering the majority of the program. Quick links: How Are Mem…

Join Byrne-Wheeler Report hosts Joe and Bob as they discuss key disclosures from Hot Chips 2026, including Intel’s Diamond Rapids Xeon, AMD’s Helios MI450 platform, NVIDIA’s LPX/LPU architecture, Cerebras’s CS-4 wafer-scale engine, and OpenAI’s Jalapeño ASIC. If you work in AI hardware, semiconduct…

I finally overcame the delays from BWR, regular doctor visits, distractions, consulting projects, and a side project to post a new XPU.pub story. More to come (e.g., Arm news).

In conjunction with supplying chips and services to Amazon, Qualcomm is granting Amazon warrants for 25 million shares @ $161. They vest through 2036 in tranches tied to actual purchases totaling $60 billion (max). QCOM opened at $170; the warrants were already in the money.

Qualcomm has scored an important data-center win. The company will collaborate with Amazon on AI inference and optical connectivity, but it provides no details. A consequence of QCOM’s Alphawave acquisition, the win could siphon AMZN biz that otherwise would’ve gone to Marvell.

Join Byrne-Wheeler Report hosts Joe and Bob as they discuss key disclosures from Hot Chips 2026, including Intel’s Diamond Rapids Xeon, AMD’s Helios MI450 platform, NVIDIA’s LPX/LPU architecture, Cerebras’s CS-4 wafer-scale engine, and OpenAI’s Jalapeño ASIC. If you work in AI hardware, semiconduct…

Back from Hot Chips 2026. Bob and I will discuss highlights in the next Byrne-Wheeler Report. 👈👉

An Arm-compatible IBM mainframe wasn’t on your Hot Chips bingo card, was it?* Kicking off the annual processor-nerd confab, IBM disclosed the next-gen Z-series microprocessor will natively run 64-bit Arm code concurrently with traditional IBM mainframe software. *Traditional* includes that accoun…

ARMONK, Apr 2, 2026: IBM is collaborating with Arm to develop dual-architecture hardware. IBM & Arm will combine IBM's enterprise systems leadership with Arm's software ecosystem to build flexible & scalable computing platforms.

Investors, if not also AI labs, remain hungry for alternative suppliers of and approaches to AI acceleration. British startup Fractile is raising $600 million at a $6.5 billion valuation, months after landing $220 million at a $1 billion valuation, according to Bloomberg. The company reportedly has…

Cerebras has updated its AI-compute rack. Due to ship later this quarter, the CS-4 rack roughly doubles the throughput of the earlier CS-3 to 4,465 tokens/sec. Performance gains come from an updated “chip,” the Wafer-Scale Engine 3 Turbo and lower-latency communication among wafers. More importantl…

A year ago, Nvidia announced it would invest up to $100 billion in OpenAI. By January, “up to” had deflated to $30 billion. Now, a plan for Nvidia to backstop OpenAI’s Ohio data center with a $250 billion guarantee has been whittled down to a $120 billion guarantee. As the company making the most …

High-end RISC-V developer Condor Computing has shut down, confirms David Harold at JPR. The company’s technology and some staff are folding into Andes, Condor’s main backer and a leading RISC-V licensor. The startup was one of several companies designing server-class RISC-V cores and, according to …

"While we don't share production volumes, the figures reported don't reflect the scale of our program." is a fancy way of saying Microsoft hasn't made a viable Maia NPU yet but is still trying. Reuters is reporting that The Information is reporting that Microsoft will announce the third-gen Maia 30…

AMD has acquired Taalas, an AI-acceleration startup that hardwires models to achieve eight times the token rate of its nearest competitor while requiring a fraction of the power. Located in Toronto and founded by Tenstorrent cofounder Ljubisa Bajic, Taalas has demonstrated a chip implementing the L…

Anthropic is going to build a chip to reduce costs, as if it didn't already have a capex problem. If it builds them somewhere other than TSMC, it may also gain access to more chip supply.

AMD has aligned its data-center GPU architecture with that of market leader Nvidia. However, AMD highlights the token-throughput and efficiency gains of the new MI455X over the previous-generation MI355X instead of touting how customers will find it easier to port their AI models to the GPU. Implem…

In Episode 16 of the Byrne-Wheeler Report, Joe and Bob discuss the latest developments in AI hardware, including disclosures at AMD’s Advancing AI day, NVIDIA Vera, and upcoming conferences Hot Chips and OCP APAC. They analyze new processor architectures, interconnect standards, and market implicat…

The Financial Times is reporting that NXP is evaluating Ambarella as an acquisition. Although the FT describes $AMBA as a designer of camera chips for self-driving cars, the automotive business accounts for only 22% of the company’s sales. The rest is IoT, meaning a mix of industrial and consumer v…

In Episode 16 of the Byrne-Wheeler Report, Joe and Bob discuss the latest developments in AI hardware, including disclosures at AMD's Advancing AI day, NVIDIA Vera, and upcoming conferences Hot Chips and OCP APAC. This clip excerpts part of the discussion about AMD Helios. Joe: https://xpu.pub Bob…

Joe and Bob discuss AMD Helios in this clip from the most recent Byrne-Wheeler Report, BWR-16, available on podcast platforms and YT.

In Episode 16 of the Byrne-Wheeler Report, Joe and Bob discuss the latest developments in AI hardware, including disclosures at AMD's Advancing AI day, NVIDIA Vera, and upcoming conferences Hot Chips and OCP APAC. They analyze new processor architectures, interconnect standards, and market implicat…

Reuters reports that Intel has licensed Atom to Rosaic, providing RTL to ex-Rivos founder and LBT pal. Presumably "Atom" means Intel's E-Core. Even if not, the Atom/E-Cores (excepting the first gen or two) delivered exceptional performance/mm2. Of course, a lot comes from Intel's back-end team beat…

NextSilicon, a startup focusing on processors for high-performance computing, is developing a server-class 64-core CPU chip called Arbel. A RISC-V design based on a core NextSilicon employed in its Maverick-2 data-flow compute accelerator, Arbel is due to ship in early 2028. A development chip is a…

A couple of years ago I posted here about how what was going on in AI looked like the vendor financing of the telecom boom, which is now popularized as "circular financing." It hasn't concerned me until now because a vendor such as Nvidia might put in $x billion, but the customer would lever that u…

AMD has revealed the first models of its sixth-generation Epyc product line, Venice. Based on the new Zen 6 and Zen 6c cores, the now-shipping processors offer two packaging options. Maximum core count increases from 192 in the previous Turin generation to 256. As before, the highest-frequency chip…

Joe and Bob discuss the latest advancements in semiconductor technology, including IBM’s sub-1 nanometer process, Intel’s new milestones, Qualcomm’s data center strategy, Etched, OpenAI, and Oxmiq, interconnect companies/technologies such as CXL and Upscale, and memory company Micron Technology.. T…

As data-center AI has come to dominate the market for computing accelerators such as GPUs, new vendors are targeting the previous main application of accelerators, high-performance computing (HPC), introducing unusual technologies. NextSilicon is one of these companies, supplying national labs and …

Nvidia is mounting a charm offensive for its Vera microprocessor, touting its sizable strengths in key areas and retconning it into a framework of how data-center processors have changed. If this framework is valid, the company is at least one generation ahead of other server-processor suppliers. A…

Nikkei reports that hidden debt at U.S. tech giants swelled eightfold in roughly four years to an estimated $1.65 trillion as artificial intelligence investments ballooned, a Nikkei study shows, exceeding the roughly $1.35 trillion in debt reflected on their balance sheets and making it tougher for…

Ahead of AIpalooza, AMD and Anthropic announced a strategic partnership to deploy less than [pinky to mouth] 2 GW of AMD Instinct MI450 Series GPUs in AMD Helios rack-scale solutions, with deployment of the first gigawatt beginning in the first half of 2027. This deployment builds on Anthropic’s ex…

Ahead of AMD's AI pep rally this week, Microsoft has announced three new Azure offerings based on AMD chips. HDv2 virtual-machine instances are bruisers, "Featuring nearly 500 physical 6th Gen AMD EPYC CPU cores, 4 terabytes of RAM, 32 terabytes of local NVMe storage and 400 Gb Azure Boost network…

Three interesting stories about selling US GPUs to China: * July 14 a Reuters headline proclaimed Nvidia had begun shipping "powerful" H200s to China but the story went on to state that the number was "very few." The H200 is the nerfed version of Hopper, the generation before Blackwell (which has …

Lightreading reports that Qualcomm appears to be in retreat from the current small-cell market. Late last year, however, it began notifying the market of plans to stop selling FSM100, FSM200 platforms to new customers. While it will continue to provide support for existing clients, there is to be o…

Nvidia has teased details of its forthcoming Rosa CPU, the successor to Vera, which is based on Olympus cores. Rosa will integrate the new Rigel core (named after the low-cost oscilloscope supplier). The company writes it will deliver “higher per-core performance than Olympus while keeping the same…

Just a quick note: The Byrne-Wheeler Report is now available as a podcast. Please give it a listen.

On its fifth-generation chip yet still failing to dent the AI processing market, SambaNova has closed a $1 billion round. This follows a $350 million round earlier this year. The company bases its chips on a CGRA architecture, an array of compute and memory units onto which an AI model’s graph maps…

Bloomberg reports that AI chip startup Positron is in talks to raise about $750 million in a two-phase round. The first phase is expected to value the company at $3.5 billion, according to people familiar with the matter. The second phase could carry a valuation of around $5 billion. I’m in the wro…

Byrne-Wheeler Report, Episode 15, discusses Micron Technology's take-or-pay contracts

Joe and Bob discuss the latest advancements in semiconductor technology, including IBM's sub-1 nanometer process, Intel's new milestones, and Qualcomm's data center strategy. They explore how these innovations are shaping the future of chip design and data center infrastructure. 01:11 IBM's Sub-1 …

Joe and Bob discuss the latest advancements in semiconductor technology and businesses, including AI accelerators from Etched, OpenAI, and Oxmiq, interconnect companies/technologies such as CXL and Upscale, and memory company Micron Technology. They explore how these innovations are shaping the fut…

Reuters is reporting Deepseek is looking to make a data-center AI processor. The company has been using Huawei silicon and had used Nvidia GPUs. It would still need to use such products for training, using the new inference-only chip for serving models. The company has proven clever on the model si…

Byrne-Wheeler Report, Episode 15, discusses the roadmap Qualcomm unveiled for its data center business.

A little too canny for my taste.