The Digital Tycoon

Bitcoin looks flat, but it took a round trip: it slid to about $81,400 Monday night as roughly $771 million in leveraged bets got wiped out, then climbed back. Tech did the slipping in stocks, led by the memory names.

China put a date on cheaper memory, and Korea's market took a 6% hit

On Monday, China's biggest memory maker, ChangXin Memory Technologies (CXMT), said it'll launch a new kind of server memory before the end of this year. By Tuesday afternoon in Seoul, South Korea's main index had fallen about 6% and closed below 6,000. A circuit breaker briefly paused selling, the first in about two months. SK hynix fell 8.5%, and Samsung fell 6.8%. In New York, SanDisk dropped more than 6% and Micron about 4%.

Here's how it works. DRAM is the short-term memory every AI chip reads from while it works. CXMT says its new design, called 4F², fits each memory cell into about a third less space than today's standard. More memory per chip means more memory per factory, which usually means a lower price for each gigabyte. CXMT's plan puts it about two years ahead of Samsung's own target. The catch: CXMT hasn't published specs, and Korean analysts say a working design and steady mass production are two different things.

Here's the bigger picture: nobody is worried about demand. The same Monday, Microsoft unveiled its next AI chip, Maia 300, built for memory-hungry work like long conversations and agents. One full rack carries 27.6 terabytes of high-speed memory. The fear is about supply, and an old one: a cheaper rival shows up, and prices fall for everyone.

That's the money angle. Memory works more like a commodity than most of tech. One maker's chip can stand in for another's, so supply and demand set the price, and profits swing hard both ways. Memory stocks rallied hard on AI demand. For the buyers, the clouds and the AI labs, more supply is good news. For anyone who rode the memory stocks up, it's the first serious question about how long those margins last.

Why it mattersThe most profitable layer of the AI build is also the easiest one to copy, and China just put a launch date on copying it.

Sources · Seoul Economic Daily · Kospi falls below 6,000 · Seoul Economic Daily · CXMT's server DRAM plan · Yahoo Finance · Samsung, SK Hynix, Micron, SanDisk · Microsoft · Azure Maia to multi-gigawatt deployments

1. AIMeta shipped its first frontier model, and it's built to remember you

   Meta's superintelligence lab released Muse Wonder on Monday, its first shot at the top tier of AI models. Builders can use it now through Meta's API, at prices TechCrunch puts in line with GPT-6 Sol and Opus 5.5. The pitch is memory: Meta says Wonder keeps track of what matters to you across days and weeks. Another top-tier option keeps the price war going, and that's good news for anyone buying tokens.

   Go deeper · TechCrunch

2. CHIPS + ENERGYMicrosoft says its next chip gets 40% more AI out of every megawatt

   Microsoft unveiled Maia 300, its third homemade AI chip, and said its Maia family will scale to multi-gigawatt deployments starting in 2027. It claims 40% more AI output per megawatt than leading GPU systems. The older Maia 200 already runs Copilot and serves OpenAI's models. When power is the bottleneck, output per megawatt is the number that matters, and every chip Microsoft makes itself is one it doesn't buy from Nvidia.

   Go deeper · Microsoft

3. DIGITAL BUSINESSX folded Grok, Cursor and X Premium into one subscription

   X launched The X Subscription on Monday, one plan for Grok, the Cursor coding tool and X Premium. The $30 tier bundles three products that used to cost $90 a month combined. Paid plans share one pool of Grok usage across Grok and Cursor, while other companies' models in Cursor are billed separately. It's a bundle with a nudge: the usage you've already paid for in Cursor runs on Grok.

   Go deeper · xAI

4. BITCOINA South Korean bank made a business loan backed by bitcoin

   Jeonbuk Bank has made at least one loan in Korean won with bitcoin as the collateral, which CoinDesk calls the country's first from a traditional bank. The bank won't hold the coins. A registered crypto custodian does, and a default means a court-ordered sale. New court rules for seizing digital assets made it possible on October 1. A bank lending against bitcoin treats it as collateral, not a curiosity.

   Go deeper · CoinDesk

5. CRYPTOBlur is closing its NFT marketplace website as trading dries up

   Blur will shut its website next month, though the onchain marketplace keeps running. Its lending product stops matching new loans on November 12. Monthly volume peaked around $1.43 billion in March 2023 and has fallen under $7 million. The BLUR token dropped more than 20% on the news. Blur paid traders in tokens to show up, and when the rewards faded, so did the trading.

   Go deeper · The Block

6. ROBOTS + FUTURE TECHXPeng put dates on its humanoid robot, and its first job is retail

   Chinese carmaker XPeng says it'll start mass-producing its IRON humanoid by the end of this year and deliver the first ones in the first half of 2027, with Europe later that year. The first jobs are in stores and malls, with factories and homes after that. The robot talks in three languages and remembers past conversations. Humanoids are being sold first as a storefront that greets your customers.

   Go deeper · TechNode

Q. Why do memory chip stocks swing so much harder than Nvidia?

Because they sell different kinds of things. Nvidia sells a design nobody else makes, so it has a lot of say over its own prices. A memory chip is closer to a commodity, like wheat or oil. One maker's chip can stand in for another's, so the market price is set by how much supply is out there compared to how much buyers want.

That makes memory a boom and bust business. Factories take years to build, so supply shows up in waves. When demand outruns supply, prices jump and profits balloon. When new supply arrives, prices fall for every maker at once, even the best ones.

That's why one Chinese announcement could knock SK hynix down more than 8% in a single session. Investors weren't doubting AI's appetite for memory. They were asking what happens to prices if a cheaper supplier shows up. With memory stocks, the question is always supply.

Sources · Seoul Economic Daily