The AI wallet is tightening. A day after Alphabet’s earnings report, its stock plunged 13%—the worst single-day drop in company history. The Nasdaq 100 fell 6.7% in October, wiping out $2.7 trillion in market value. Even bonds tied to SpaceX and Meta are trading at steep discounts. Meta has even planned to issue stock, aiming to raise up to $97.5 billion. But the first to run out of cash are the companies that specialize in renting out computing power for AI. These firms are called neoclouds: they buy NVIDIA GPUs, build data centers, and lease computing capacity to large tech firms and AI startups. They’ve borrowed heavily—and when interest rates shift, they feel the pain first. CoreWeave issued a bond earlier this year with a coupon rate of 9.75%. Now, on the secondary market, buyers demand yields of over 15% to take it on. In other words, if it wants to borrow new money, it’ll have to pay interest rates of 15% or higher. The door to issuing high-yield debt for these companies is essentially shut. Nebius considered its first bond offering but ultimately abandoned it. Its CEO publicly stated last week that some funding channels are “temporarily closed,” and the company is now relying on customer prepayments for cash. The market doesn’t doubt AI—it doubts these companies can generate enough revenue to cover their interest payments. The first to have their necks squeezed in this AI wave are the landlords who serve AI.

You buy a cold wallet precisely because it’s offline. But an Indonesian buyer opened his Trezor and found an extra circuit board inside—capable of connecting to the internet. In August, he purchased a Safe 5 from CryptoBilis, a Southeast Asian distributor, on Indonesia’s e-commerce platform Tokopedia. Upon receiving it, he noticed the weight felt off. When he opened it, he discovered a board sandwiched between the mainboard and the screen, equipped with a cellular modem and antenna. According to him, this board can read the mnemonic phrase displayed on the screen and transmit it via mobile network as soon as it’s powered on. The device hasn’t yet been sent to the manufacturer for inspection, and Trezor hasn’t confirmed anything—but it has already issued a warning: anyone who bought a Trezor from CryptoBilis since March, regardless of model, should immediately transfer their funds to a new wallet created with a newly generated mnemonic phrase. Trezor will ship replacement units free of charge. No losses have been reported yet. But since wallets are self-custodial, the manufacturer cannot see users’ balances—so no reports don’t mean no compromises. Last month, Ledger faced a nearly identical scam: a group bought genuine devices, installed hidden components to steal mnemonic phrases, then resold them back to the market through Southeast Asian distributors. According to Arkham, an estimated $95 million was stolen in just three days. In response, Ledger has suspended all sales in six Southeast Asian countries plus Hong Kong.

More terrifying than hackers is someone standing in your living room with a gun. In September last year, a family in Minnesota, USA, was invaded by two armed men and held hostage for a full nine hours. The intruders didn’t want cash or jewelry—they wanted their cryptocurrency assets, totaling approximately $8 million. The intruders were brothers from Texas, Raymond Garcia and Isiah Garcia. Isiah forced the father to log into his cryptocurrency accounts and transfer the coins to a wallet provided by Raymond. But it didn’t end there. The brothers later discovered that the father had stored a hardware wallet in a cabin three hours away—and set their sights on that money too. Before they could get their hands on it, things took a turn: the family’s son called 911, forcing the brothers to flee. They didn’t get far. Police traced them using a Wendy’s receipt, surveillance footage from the motel where Raymond checked in, and a rented vehicle—leading them straight to the brothers’ home in Texas. In June this year, both pleaded guilty. The sentencing is now final: Raymond received 15 years, Isiah 15 and a half years. No matter how securely you store your private key, when a gun is pressed to your head, you’ll hand it over yourself.

Intel just delivered its best financial results in years, yet its stock plunged over 20% after hours. Intel reported third-quarter earnings: revenue of $16.51 billion, up 21% year-over-year. Non-GAAP earnings per share came in at $0.52, surpassing Wall Street’s expectation of $0.40. The biggest growth came from data centers. AI servers require not only GPUs but also CPUs to manage tasks and process data—and Intel’s Xeon server processors are selling strongly. Revenue from this segment reached $7.318 billion, a 78% year-over-year increase. The CFO called it the strongest data center growth in 15 years. The problem lies in the next quarter. Intel’s guidance for fourth-quarter revenue is $15–16 billion, with EPS of $0.25—well below market expectations of approximately $17.07 billion and $0.43. Two areas are dragging performance: PC chip revenue fell 5% year-over-year, and the CFO expects weak PC demand to persist at least through the first half of next year. While demand for server CPUs is surging, Intel can’t produce enough—further hampered by a facility accident in Israel that reduced capacity. Before the earnings report, Intel’s stock had already dropped 3.6% during the day to close at $92.07; after the report, it tumbled another 20% in after-hours trading. AI orders are real—but Intel simply doesn’t have enough inventory to fulfill them right now.

Placed orders totaling $170 million, yet ended up profiting only $119,000. The amount is small, but the entire strategy was clearly recorded on-chain. Hyperliquid is an on-chain derivatives trading platform that, in addition to cryptocurrencies, also offers crude oil futures contracts. Between March 8 and 10 this year, international oil prices experienced sharp volatility. The monitoring firm Solidus Labs detected that a trader placed over $170 million in unusual orders on Brent crude futures contracts using two accounts during this period. This tactic is known as "spoofing": placing large orders on one side to create the illusion of strong buying or selling pressure, then canceling them as soon as the price moves, while executing real trades on the opposite side. In the most typical instance, he canceled approximately $17 million in sell orders and almost simultaneously bought around $425,000. Who this trader is and whether the activity was intentionally manipulative have not been confirmed. Hyperliquid’s response was that all orders, cancellations, and executions are recorded on-chain and publicly accessible in real time. That’s true—Solidus precisely used these public records to identify the trader. But during the time that $17 million in sell orders remained visible on the order book, other market participants saw what appeared to be an impenetrable wall of supply.

Well, just a few days ago, Sumit Gupta, CEO of Indian exchange CoinDCX, said on X that reports of a chain-based theft sounded like a false alarm and urged people not to spread rumors. Today, the company officially confirmed that 15.99 million USDT were indeed stolen—and the company will cover the loss. The official explanation is that these addresses were old wallets disabled after being hacked last year, but some users continued depositing funds into them. Meanwhile, the automated script designed to transfer funds out of these wallets was temporarily malfunctioning, causing funds to accumulate in a vulnerable address until they were fully drained. Keep in mind that CoinDCX was previously hacked in July last year, losing approximately $44 million.

Elon Musk, while expressing concern that humans may not be able to control superintelligence, continues to actively participate in AI development. In the inaugural episode of “The Builders,” hosted by his biographer Walter Isaacson, he explained that he believes it’s impossible to stop AI’s advancement, so rather than stand by, he prefers to get involved and try to steer it in a positive direction. Yet he remains pessimistic about the endgame. If digital intelligence becomes a million times more capable than biological intelligence, how can humans possibly control it? Musk admits he struggles to even imagine such a scenario—“so I hope AI treats us kindly.” Coming from someone who continues to build AI, this statement carries a sobering weight. He proposed one concrete step that could be taken now: companies developing cutting-edge AI should communicate weekly, and allow competitors to test new models about a week before public release, helping to identify potential issues. He also hopes AI will pursue truth and maintain curiosity. I’m willing to listen to these practical suggestions, but I can’t help wondering: how far is a one-week head start for competitors from the kind of “million-fold” superintelligence he fears? When discussing U.S. national debt, he again places his hope in AI and robotics. Musk says that after engaging in government work, he became convinced that reducing deficits through internal budget cuts alone is impossible—only technology can boost productivity. Reflecting on his own political and governmental involvement, he admits he stayed too long and should have remained behind the scenes, participating only for one or two months. When Isaacson asked if he’d become a bit addicted, Musk replied: “Yes, I got a little carried away.” He estimates the economic impact of technology even more boldly: pure digital superintelligence could increase global output of goods and services by roughly 20% to 30%. If robots gain the ability to manipulate objects and enter space, the economy could potentially expand to a million times the size of today’s Earth-based economy. These are his own rough estimates. Yet when the conversation turned to his own platform, X, he acknowledged that it hosts too much content stoking political anger, and that the information feed sometimes feels dark. He warned people not to be misled by the “optimism bubbles” of real life or the “pessimism bubbles” online. Throughout the entire conversation, what I found most appealing was when he brought the discussion back down to planting vegetables and cooking meals for friends: even if machines eventually do these things better than humans, people can still choose to do them themselves. After hearing all those grand predictions, this small, human choice felt deeply relatable—it doesn’t have to outperform machines to be worth doing. Full video: https://t.co/CCSfTDDMAZ

I thought he’d made $8.99 million and would take a breather, but instead, he immediately opened a long position in crude oil. In early June, he borrowed 30 million USDS from Spark through circular collateralization and bought 18,212 $ETH at an average price of $1,647. On October 19, he sold 12,468 $ETH at an average price of $2,368, receiving 29.52 million USDC. According to @embercn’s calculations, this trade generated a profit of $8.99 million. Shortly after, he transferred 20 million USDC to Hyperliquid and opened a crude oil long position worth approximately $22.4 million. Both the CL and BRENTOIL positions carry 3x full-collateral leverage. Just after locking in profits on crypto, he shifted his bets to a different market. https://t.co/dnedRNVcpE

300 bags, 28 watches, and 6 cold wallets—the range of items seized by Turkish police left me momentarily stunned. The individual detained is Richard Heart, founder of the Ethereum token project Hex, launched in 2019. He and his wife were detained in Istanbul on Wednesday, followed by three others identified by police as his bodyguards. Authorities searched 15 locations locally, reporting the seizure of approximately $7.15 million along with the aforementioned items, as well as jewelry, graphics cards, and firearms and ammunition; Turkey’s state news agency also reported that nine vehicles registered in Heart’s wife’s name were confiscated. I initially assumed that in this crypto-related case, all attention would center on the six cold wallets. But when placed alongside the findings from the 15 search locations, the 300 bags stood out as harder to overlook: what was laid before the police wasn’t just on-chain assets. The seizure of cold wallets does not equate to control over the funds within them. Finland previously issued a Red Notice for Heart over allegations of serious tax evasion and assault. The U.S. SEC’s lawsuit against him was dismissed in 2025 due to jurisdictional issues and has not been refiled; these two matters must not be conflated. Legal proceedings and the examination of seized items are still ongoing—I won’t speculate on the value of those six cold wallets yet.

Toronto Police are investigating a kidnapping targeting the owner of a cryptocurrency exchange: On July 25, the victim was abducted at gunpoint while approaching his car, with suspects attempting to gain access to his crypto assets and company security systems. Authorities have not yet disclosed the victim’s identity or the exchange’s name. What troubles me isn’t just that someone committed a kidnapping for crypto assets—but that police revealed three of the seven suspects arrested are members of the Canadian Armed Forces. Searches also uncovered firearms, GPS tracking devices, and over 2,700 rounds of ammunition, approximately 2,500 of which have been identified as stolen from military stockpiles. When the target shifts from a wallet to a person who controls access to assets and systems, the risk extends far beyond the screen. The involvement of military personnel and the infiltration of military-grade ammunition into this case suggest this is not merely a personal extortion attempt—it’s a warning sign that demands urgent attention to weapons management.