Micron posts the best quarter in its history, with $54.23 billion in revenue

The AI stack on 30 September 2026. 11 moves, 9 with a primary source.

Micron's own earnings call now puts the memory shortage it flagged this week at least through 2028

· L2 Silicon · Confirmed

On the same September 30 earnings call where Micron reported record revenue, already covered here, the company's own market outlook added a specific extension: supply and demand in calendar 2027 and 2028 will stay much tighter than in 2026, and Micron said it does not have line of sight to when that balance returns. This is Micron's own quarterly disclosure, not a third-party forecast, confirming in the company's own words what outside analysts had only predicted. Micron makes the memory chips used in AI servers as well as ordinary phones and laptops, so a longer shortage points toward memory prices, and potentially the price of phones, laptops, and other electronics, staying higher for longer.

Source

DeepSeek open-sources chip tools that let AI models run on Huawei hardware instead of Nvidia's

· L2 Silicon · 431 TFLOPS (BF16); 861 TFLOPS (FP8) · Confirmed

Chinese AI lab DeepSeek has released, for free, software tools that let AI models run efficiently on Huawei's Ascend chips instead of the Nvidia chips most AI software is built around. The tools, posted on DeepSeek's own code pages, include performance figures like 431 trillion calculations per second on one common type of math Ascend chips do. Nvidia's chips are hard for Chinese companies to buy because of US export rules, so software that makes a Chinese-made alternative easier to use is part of China's effort to build an AI industry that doesn't depend on American chips.

Source

Micron posts the best quarter in its history, with $54.23 billion in revenue

· L2 Silicon · $54.23bn, record Q4 · Confirmed

Micron, one of only three companies that mass-produce the memory chips used in phones, laptops and AI servers, made $54.23 billion in revenue in the three months to early September, the biggest quarter in its history and up from $41.46 billion the quarter before. Its profit for the quarter was $37.70 billion. 'Micron delivered record fiscal 2026 results, and we expect an even stronger fiscal 2027,' said the company's chairman and chief executive, Sanjay Mehrotra, in the earnings release. For anyone buying a phone or laptop in the next year, a company this central to the world's memory-chip supply calling its own year ahead 'even stronger' is a sign the high prices behind this quarter are not about to reverse.

Source

Google launches Gemini 4 Argon with an output window of up to 1 million tokens

· L4 Models · up to 1M tokens · Confirmed

Google DeepMind has released a new flagship AI model, Gemini 4 Argon, that can produce up to 1 million tokens (small pieces of text) in a single answer, up from 64,000 tokens before, what Google calls an 'industry-leading' increase. That lets it work through much longer problems in one pass instead of stopping partway. For now it is available only to a small group of trusted testers working on cyber-defence, with wider access to come later, and Google says the price will roughly double once an introductory period ends.

Source

California's 'No Robo Bosses Act' bars firing workers by AI decision alone

· Band Rules and society · SB 947, first in nation · Confirmed

California's governor has signed SB 947, which state Senator Jerry McNerney's own office calls the 'No Robo Bosses Act.' The law stops an employer from relying only on an automated system to discipline or fire a worker, requiring a human to review the decision too, and McNerney's office calls it the first law of its kind in the country. For most California workers, it means a person has to sign off before an AI system can get them disciplined or fired, not just an algorithm.

Source

HPE's first AMD Helios server-rack order lands at $1.2 billion, from GPU cloud Vultr

· L3 Iron and cloud · $1.2bn; 72 GPUs/rack · Confirmed

HPE will build $1.2 billion worth of AI computer racks for the cloud company Vultr using AMD's new Helios design, with each rack packed with 72 AMD Instinct MI455X chips. It's HPE's first order for this AMD rack. The computers will run in the United States, though neither company has said how many racks that is or when they'll be ready.

Source

SoftBank completes its roughly $3.1 billion buyout of data-centre investor DigitalBridge

· L1 Power and land · ~$3.1bn · Confirmed

SoftBank has finished buying all the shares of DigitalBridge, a company that owns and manages data centres, phone towers and fibre-optic networks, for about $3.1 billion. DigitalBridge will keep running as its own unit, still led by its current chief, Marc Ganzi. The deal gives SoftBank more of the physical buildings and cables the internet and AI computing run on.

Source

Regulators delay a plan to force 6,831 megawatts of new power onto PJM's grid until 2027

· L1 Power and land · 6,831 MW; to Feb 2027 · Reported

PJM, the grid operator for 13 US states and Washington DC, wanted power companies to build 6,831 megawatts of new backup electricity as data centres multiply. Per PJM's own account of the decision, federal regulators accepted the plan but pushed its start back five months to February 28, 2027, and gave any new plants built for it until June 2032 to start running. This desk read PJM's summary, not the regulator's order itself. If you live in one of those 13 states, this is one of the mechanisms meant to keep your power on as more data centres plug into the same grid.

Source

A federal appeals court sides with Thomson Reuters over AI startup Ross, but keeps its reasoning sealed

· Band Rules and society · 3rd Cir., sealed · Reported

The Third Circuit Court of Appeals has ruled for Thomson Reuters in its long-running copyright case against Ross Intelligence, a now-closed legal research startup, according to the court's own order and a legal-news account of it. The panel's written opinion is, for now, filed under seal while both sides propose redactions, so the court's actual reasoning isn't public yet, only the fact that Thomson Reuters won. One published account reports the tool predated today's generative AI chatbots and was not itself generative AI; this desk could confirm that detail only through that account, not the sealed opinion.

Source

A space-solar startup signs a 20-year deal to supply 100 megawatts to an undersea data-centre company

· L1 Power and land · 100 MW, 20 years; option for 250 MW; public demo March 2027 · Confirmed

Virtus Solis, which is developing solar power collected in orbit and beamed to Earth, announced on 30 September a 20-year power purchase agreement with Brae Systems, a company building data centres under the sea. Virtus Solis will supply 100 megawatts of around-the-clock power, 876,000 megawatt-hours a year, through a receiving station to be built in Illinois, and Brae can expand that to 250 megawatts within three years of commercial operation. The same release says Virtus Solis's first public technical demonstration will be held in March 2027. This is a contract for power from a system that has not yet been shown working in public, so nothing reaches a grid or a data centre soon.

Source

Chip-equipment maker ACM Research Shanghai says its order backlog reached 17.1 billion yuan, up 88 percent

· L2 Silicon · RMB 17.1bn (US$2.5bn) backlog, +88.2% year on year · Confirmed

ACM Research, which supplies machines that process silicon wafers and panels for chipmaking and advanced packaging, said on 30 September that its Shanghai operating subsidiary's order backlog was 17.1 billion yuan, about $2.5 billion, as of 29 September. That is 88.2 percent higher than the backlog it disclosed a year earlier. The figure counts tools already shipped but not yet booked as revenue, and tools still to be shipped, and the company notes that backlog is not an accounting measure and may not all turn into revenue. Full results for the first nine months follow on 28 October. For people outside the chip industry nothing changes directly; it is one measure of how many chipmaking machines are on order.

Source

Strata, the whole AI stack, explained simply. Every number carries a source and a confidence label.

Today · The Stack · Stories · Archive · Glossary · Break the Chain · Focus · About · RSS feed

Privacy · Terms · Corrections · Report an error

Also from Dheeco: Dheeco · AP Fact Check

Published by dheeco.com · © 2026 Dheeco