AI Is Bursting - WAN Show October 30, 2026

The show opens with the hosts reacting to a worldwide technology-stock selloff. After an earlier decline in China, the hosts joke that the hosts were waiting for the crash to reach America; Intel then fell roughly 25–30% in a day, while Nvidia and Micron also declined. They note that Nvidia remains up over six months, Micron remains substantially up, and Intel had previously experienced a spike, so a short-term fall does not automatically mean the entire sector has collapsed. The hosts say they do not want to celebrate people’s portfolios losing money, but argue that a decline could ultimately be healthy after the technology sector absorbed enormous amounts of money, land, water, electricity, supply, and talent. They are more concerned about companies trading on “vibes” than about reading charts as reliable predictions. The past six months, they say, have no predictive power over the next six months; treating chart patterns as dependable is comparable to the gambler’s fallacy. Both hosts describe themselves as mostly sitting out stocks and cryptocurrency, though they acknowledge that a responsible investment strategy can include some stocks. They discuss bonds and gold as possible safe havens and mention a video ranking which G7 countries might go bankrupt, with Canada apparently benefiting from owing money in favorable ways and having its own currency. They repeatedly stress that neither host is a financial adviser and that this is not financial advice.

The hosts examine OpenAI CEO Sam Altman’s reported claim that, from a global-allocation perspective, it is “quite terrible” for everyone to keep a large amount of the world’s limited DRAM supply in personal computers instead of putting it in data centers. Linus initially grants part of the underlying utilization argument: a personal computer is commonly used in three shifts—while its owner sleeps, works, and actually relaxes—so allowing someone else to use the hardware during idle periods could improve utilization. He compares the idea with Tesla’s proposed robotaxi model and publicly shared pools or skating rinks. He then rejects the analogy as applied to AI companies. Hyperscalers are consuming finite resources for shareholder profit, sometimes buying hardware they cannot power simply to prevent competitors from obtaining it. That is not like a public pool that everyone can use at a reasonable price. The hosts compare this behavior with the technology industry first hoarding skilled workers and then competing to lay them off, while preparing to hoard hardware as a replacement. They argue that companies such as Alphabet and Microsoft could simply make billions from good products without requiring stock prices to rise forever. Valve is offered as an example of a private company able to make long-term decisions, although Linus still criticizes Valve’s 30% Steam cut and lack of competition. They conclude that local AI systems are becoming surprisingly capable and that sovereignty over local models is valuable.

The hosts discuss Anduril’s Blue Chip laptop, announced by Palmer Luckey with pre-orders expected in 2027. The laptop is supposed to be assembled at Anduril’s Arsenal One facility in Ohio, use a Qualcomm Snapdragon X-series chip made at TSMC’s Arizona plant, and be “made in America.” The hosts point out that Anduril’s wording is more cautious: its press release says “designed and assembled in the United States,” not manufactured there. Luckey acknowledged that the display is one of several components currently made only in China and said 90% of the parts would come from America or allied countries, without naming all those allies. The Verge tried to establish a precise figure. Luckey initially implied the laptop would exceed the Buy American Act’s 65% threshold, but Anduril later said the prototype was only about 32% U.S.-origin by component cost and could not initially confirm whether the shipping model would clear 65%. The company then reversed that statement and said the launch version would exceed 65%, while Luckey was unavailable for follow-up. The hosts note that Luckey reportedly expects Blue Chip to cost about 20% more than a typical high-end PC, despite admitting that a normal consumer may be better served by a Dell. They explain why this is difficult: flat-panel displays have largely not been manufactured at consumer scale in the United States since the CRT era, and domestic supply chains for tooling, molding, PCB work, materials, and specialist labor have atrophied. Linus compares this with LMG’s own Canadian manufacturing: local molds can take months and cost nearly an order of magnitude more than overseas molds, while Shenzhen benefits from dense networks of suppliers and expertise. They conclude that restoring domestic electronics production is strategically valuable but cannot be achieved through branding alone.

The hosts discuss Apple’s newly announced MacBook Pro models, including a touchscreen, tandem OLED display reaching up to 2,000 nits, up to 120 Hz refresh, optional nanotexture glass, Thunderbolt 5, HDMI, a 3.5 mm headphone jack, upgraded speakers, and claimed battery life of 22–24 hours. Both 14-inch and 16-inch models are slightly thinner and lighter than their predecessors. Linus says the touchscreen is a good accommodation rather than a problem: when he wants to scroll something quickly, reaching forward can be easier than moving the cursor. Since the OLED panel would already account for much of the price premium, he argues that critics are effectively objecting to other people receiving an additional option that does not inconvenience them. Luke says he would almost never touch the screen because of fingerprints, but accepts that users can simply ignore the feature. Linus compares this with a previous laptop whose poor trackpad forced him to use its touchscreen. Apple has also designed touch sensors directly into the display rather than placing a separate layer over it, and macOS reportedly adjusts menu spacing depending on whether the user is touching or using a mouse. Linus acknowledges that Windows already performs similar adjustments, but says he had not noticed because the small, dense touchscreens he had used did not make the spacing useful. The discussion turns into broader criticism of Apple’s product naming and density. The hosts complain that iPads are difficult to date because generations are not clearly identified, unlike PlayStation numbering. They also praise Sony Xperia phones for retaining a headphone jack and for a new Android update that finally lets users swap the back and task-switcher buttons.

The hosts report that more than $100 million in freight intended for U.S. data centers has allegedly been stolen since June, with nearly two dozen similar incidents reportedly occurring since 2025. The stolen shipments have included servers, memory, copper wire, fiber-optic cables, custom cryptocurrency-mining machines, modular power equipment, cooling equipment, clothing, cosmetics, nicotine pouches, and consumer electronics. One shipment contained $300,000 in copper wire spools; another contained roughly $1 million in custom power equipment. The more sophisticated cases involved off-duty police hired as escorts. In two large thefts, the escorts were forced to stop or were struck by other vehicles, after which shipments worth a combined $23 million disappeared. The cargo could not be tracked because the thieves allegedly disrupted GPS or spoofed tracking coordinates. The hosts note that some of the basic methods—finding lucrative shipments through compromised email accounts, redirecting freight, and using fake driver credentials—are not especially high-tech, but the combination of social engineering, physical interception, and GPS interference is substantially more organized than ordinary theft. A cargo-security executive compared the shipments to “pure gold bars” that are easier to steal than a bank. Another expert argued that freight companies need to slow down and properly vet drivers. Linus tells a personal story about cleaning up electrical job sites as a child, stripping and selling leftover copper wire. Electricians eventually helped by stripping the wire for him with professional tools. The hosts joke about a copper bicycle before returning to the point that supply-chain security is becoming more important as data-center hardware grows more valuable.

The hosts warn Synology owners about ransomware attacks exploiting an approximately 18-month-old remote-code-execution flaw. Attackers have reportedly targeted internet-exposed NAS devices, encrypted files, and demanded payment. Separate newer vulnerabilities affect Synology software packages, meaning that an NAS operating system showing as current does not necessarily mean every installed application is patched. The hosts emphasize that many ordinary owners map a network drive and then never log into the NAS again. Because Synology reportedly lacks a direct channel to warn every customer of critical vulnerabilities, users should manually check both DiskStation Manager and the Package Center for updates. They also discuss the broader risk created by “vibe coders” or AI-assisted attackers who can exploit vulnerabilities more quickly and at larger scale. The security discussion leads to Linus disclosing that he is an investor in the company making HexOS, a competing NAS operating system. He says HexOS is planning integrated CrowdSec support and the ability to block entire countries from even reaching the login screen, along with custom-domain support for technically experienced users. The hosts argue that a more secure, easier-to-maintain NAS platform could make proprietary boxes look increasingly dated, especially as home-lab users repurpose powerful enterprise hardware.

The hosts introduce a new segment in which one person must defend a randomly assigned position. The topic is whether safety guardrails in Anthropic’s Claude Opus 5.5, which reportedly trigger when users ask it to reverse-engineer intellectual property, protect game developers or harm user freedom. Linus is assigned the position that the restrictions protect developers; Luke argues that they harm users. Luke says reverse engineering was possible before AI and that people should generally be able to modify and experiment with software they own. He distinguishes private experimentation from distributing a clone or selling someone else’s work, arguing that access to a tool should not be restricted merely because the user could employ it illegally. He also notes that knockoffs already exist without AI: games can imitate a concept or mechanic without copying source code, and Roblox has rapidly produced versions of successful games. Linus counters that scale changes the problem. AI can let almost anyone reproduce an intellectual property’s behavior quickly, whereas older reverse engineering required more expertise and time. He says indie developers are especially vulnerable because an AI-generated imitation could undermine a launch, not merely attack large corporations. He also argues that existing legal remedies will not move quickly enough to prevent large-scale copying. The live poll gives the pro-developer side roughly 74% support, but both hosts admit that the issue is morally and legally ambiguous. They plan to refine the format with timed opening arguments, rebuttals, and a final reveal of whether each host actually agreed with the position defended.

For the VIVERSE takeover, the hosts play Alpha Putt, a browser mini-golf game in which the course is generated from a chosen word. They choose a five-letter word and struggle with portals, sound traps, water, helicopters, and other obstacles. Luke finishes with the lower golf score, although the hosts joke about whether viewers understand that lower is better. They describe the game as a clever, genuinely fun use of typography and later recommend Heaven Dust 2, a browser/mobile survival-horror game inspired by PlayStation-era Resident Evil. The hosts promote LMG’s Pixel Wave apparel collection: an embroidered crewneck, a screen-printed long-sleeve shirt, a flight tag, shaker keychain, and hat. They also discuss a newsletter explaining how Linus’s “what if?” ideas become products, and a holiday promotion involving Floatplane subscriptions and LTT Store credit. A $130 U.S. Floatplane purchase is described as yielding $60 in LTT Store credit, with different Canadian pricing and details still being clarified live on air. They also mention an upcoming “David Week,” featuring a Q&A, office vlog, and Magic: The Gathering stream. Linus answers a question about a provocative shirtless True Spec cable advertisement. He says the concept came from Adam, who pitched a Hercules-like presentation; Linus trusted the creative team and did not expect the imagery to appear so widely. The backing track was recreated by Dan on a LinnDrum so accurately that it triggered a copyright claim. Finally, the hosts criticize Apple’s folding iPhone for awkward camera use, poor app density, and limited home-screen layout, while Linus says Sony’s Xperia phones may be more appealing despite their confusing names and high prices.

The hosts examine Alphabet’s third-quarter results. Revenue rose 25% year over year to a record $127.6 billion, but the stock fell about 10%, wiping approximately $500 billion from its market value. Net profit fell 25% to $26.2 billion, partly because Alphabet recorded an $8.7 billion loss on its stake in SpaceX. Search revenue growth slowed from 17% to 12% and missed estimates by roughly $1.6 billion. The hosts identify the central problem as the mismatch between search growth and AI monetization. Search queries grew at their fastest rate in five years, but AI-mode queries generate less revenue than ordinary searches and cost more to serve. Google says it will address this with “multi-turn ads” that adapt as users continue a conversation, although the hosts are unsure whether this means anything beyond placing more advertising impressions into an AI interaction. Linus observes that Google’s own AI products are now threatening the web pages that originally supplied its advertising ecosystem. Alphabet also spent $51.4 billion on data centers and chips, more than double the comparable period, while cloud revenue increased 103% to $30.7 billion. That growth was attributed partly to selling Google’s TPU AI chips to firms such as Anthropic. The hosts connect this spending to the wider AI bubble: a company can make tens of billions in profit and still lose enormous market value if investors think future AI spending will weaken margins. They compare this with Microsoft’s offer of voluntary buyouts to as many as 60,000 employees at level 70 or below in selected teams across Canada, India, the United Kingdom, and the United States. The hosts say the offer includes financial support and career planning, but note that a team’s participation may signal that its future is uncertain.

The hosts report that Binance may have suffered the largest crypto theft yet, with hackers allegedly taking an estimated $6.3 billion from wallets across ten blockchains on October 29. Most of the transfers occurred within about 20 minutes. TRM Labs attributed the incident with high confidence to TraderTraitor, an elite North Korean hacking group, while reports suggested a compromised Binance software engineer’s laptop or hot-wallet signing infrastructure. Binance says it has reserved more than $6.34 billion to cover possible losses and will use public proof-of-reserves reporting. The hosts question why exchanges do not have stronger circuit breakers for movements of that size. They then consider NVIDIA’s possible $2,000-per-GPU annual license for enterprise AI-agent software, noting that NVIDIA already charges $4,500 per GPU per year for its AI Enterprise suite. They do not expect such fees to affect gamers soon, because NVIDIA has generally backported software features and supported gaming hardware for a long time, but they acknowledge that recurring software revenue is increasingly attractive as product cycles lengthen. AMD’s history of abandoning products such as Vega and Radeon VII is cited as a contrast. Discord is making eligible public servers searchable on the open web. Servers with at least 1,000 members that meet activity and safety requirements will receive public pages, and individual messages may appear without usernames or avatars. The hosts see this as a possible remedy for useful forum knowledge disappearing into Discord, while also suspecting Discord may eventually license the indexed data to AI companies. Finally, they discuss a listing for a data-center installation containing 9,600 NVIDIA A100 accelerators, 4,200 AMD EPYC CPUs, 384 TB of accelerator memory, about 1.4 PB of system memory, and roughly a quarter-million Zen 2 cores for $59.7 million with free shipping. They expect increasingly creative “hardware hacking” as enterprise AI equipment is resold and repurposed.