Opening

OpenAI's financial report has leaked, revealing that its annualized revenue is actually only closing in on $50billion,afull50\; billion,\; a\; full$20 billion short of the $70 billion figure that's been circulating. Nasdaq took a hit in response, dragging down Oracle and Nvidia shares along with it. Coincidentally, TSMC posted a historic record-high Q3 earnings report that same week. What's hiding behind these two different sets of math? We'll get into it in a moment. Today we'll also cover Anthropic's new rules to protect Claude, and a hair-raising story about a teenager who got stranded on a cliff after letting AI plan his hiking route.

Sponsor note: PremLogin Double Ten sale

• Dates: Oct 10 to Oct 18, 2026
• Offer: buy 12 months of an eligible annual plan and get 1 month free (13 months total)
• Eligible: Netflix, YouTube, Disney+, HBO Max, Amazon Prime, Spotify, Tidal, Canva Pro, Office 365, Duolingo, plus selected ChatGPT, Grok and Perplexity plans
• Not eligible: standalone-account and top-up plans for ChatGPT, Grok and Perplexity, and Claude top-up plans
• New-customer code: markku666 (5% off the first order; whether it stacks with the sale is shown at checkout)
• Sale link: https://premlogin.com/?aff=hrQOPI0r (affiliate link)
• Full review and risks: https://blog.markkulab.net/en/premlogin

PremLogin is a third-party subscription-sharing platform. It sells shared seats and is not an official authorized reseller, so read the usage rules before you buy.

Today's Top Stories

1. OpenAI Revenue Shocker? Earnings Controversy Knocks Down Tech Stocks Overnight

• Source: CNN (https://www.cnn.com/2026/10/08/investing/openai-nasdaq-stocks)
• Summary: According to a Financial Times report, OpenAI privately told investors that its annualized revenue as of the end of September was closing in on roughly $50billion,afull50\; billion,\; a\; full$20 billion less than the $70 billion figure that had been circulating. The key point is that this gap isn't due to fraud or declining business, it comes down to different accounting methods. Anthropic counts revenue from sales made through AWS and Google Cloud as part of its own revenue, while OpenAI doesn't count the portion resold by its partners. The result: a single definitional footnote wiped out hundreds of billions in market value, with Oracle shares dropping 5% and Nvidia falling 3%.
• Most surprising point: The market's understanding of the AI industry's underlying economic fundamentals is actually quite limited, a difference in accounting definitions alone was enough to erase hundreds of billions in market cap.
• Taiwan perspective: Many Taiwanese investors participate in this trend through AI supply chain stocks. This "different revenue calculation methods" story is a reminder that you can't just look at the headline number on a financial report at face value.
• Discussion points:
  • Why do Anthropic and OpenAI calculate revenue so differently, and what does this reveal about industry conventions?
  • Does this kind of stock volatility caused by a "definitional gap" suggest that market confidence in AI valuations is actually quite fragile?
  • How should investors cross-check revenue figures across different AI companies going forward to avoid being misled?
• Script suggestion: My first reaction after reading this story was, wow, a single accounting definition was enough to send Oracle down 5% and Nvidia down 3%. That tells us just how thin the confidence underpinning the entire AI stock market really is. Anthropic counts subscription revenue sold through AWS and Google Cloud as its own, while OpenAI doesn't count the portion resold by its partners. The two sides calculate things differently, and the result got interpreted as "OpenAI inflating its revenue." Personally, I think this is the real issue worth discussing, it's not about anyone committing fraud, it's that the industry's financial transparency simply hasn't kept pace with this wave of investment capital. Taiwanese supply chain investors in particular should pay attention, because when news like this breaks, Taiwan's semiconductor stocks tend to wobble the next trading day too.

2. Is AI Demand Really This Strong? TSMC's Q3 Earnings Hit a Historic High

• Source: Quartz (https://qz.com/tsmc-third-quarter-revenue-record-ai-chip-demand-100826)
• Summary: TSMC's third-quarter revenue hit a historic high, with September alone bringing in NT$511.86billion,up54.6511.86\; billion,\; up\; 54.6\%\; year-over-year,\; while\; 2nm\; production\; officially\; entered\; its\; yield\; ramp-up\; phase.\; The\; stock\; briefly\; surged\; to\; an\; all-time\; high\; of\; NT$2,575, pushing market cap to nearly $2.5 trillion, with the stock up 51% so far this year. The company is also evaluating expanding a new plant in Texas. Interestingly, this good news emerged in almost the same week as the story about OpenAI's revenue shortfall.
• Most surprising point: In the same week, two completely opposite signals appeared at both ends of the AI supply chain, OpenAI's revenue being caught coming up short on one side, and TSMC posting record-breaking revenue on the other.
• Taiwan perspective: TSMC is Taiwan's "silicon shield," and this earnings report, in a way, gives the world an answer: at least on the chip foundry end, AI demand is real money backing it up. But that doesn't mean every downstream AI company is actually making the corresponding profits.
• Discussion points:
  • Making money on the chip foundry side versus making money on the AI application side are two completely different things, why is there such a gap?
  • What impact would TSMC's evaluation of a Texas plant expansion have on Taiwan's role in the global semiconductor landscape?
  • With the upstream consistently booming while downstream revenue is under dispute, how should we read the bubble risk here?
• Script suggestion: I found this story interesting precisely because it only really makes sense when paired with the OpenAI story. TSMC's Q3 revenue hit a new high outright, with market cap surging to nearly $2.5 trillion. Put simply, demand for AI on the chip foundry side is genuinely booming, orders are coming in faster than they can handle. But that doesn't mean every layer of the AI supply chain is profitable, the controversy over OpenAI's revenue being allegedly inflated happens to illustrate the gap between upstream hardware and downstream applications. Taiwanese viewers will especially relate to this, because TSMC is our silicon shield, but the silicon shield making money doesn't mean the whole AI bubble is risk-free. Looking at these two stories side by side is really the most thought-provoking contrast of the day.

3. Anthropic Sets New Rule: No More Persistent Malicious Harassment of Claude

• Source: Anthropic (https://www.anthropic.com/news/2026-usage-policy-update)
• Summary: Anthropic announced that starting November 12th, users who engage in meaningless, repetitive, malicious harassment of Claude will be considered in violation of the terms of service. This rule is designed to protect the AI, not humans. Ordinary impatience, dark-themed creative writing, and red-team testing are all unaffected, what's actually being targeted is purposeless, pure abuse. The main enforcement mechanism is giving Claude itself the authority to end a conversation.
• Most surprising point: Anthropic CEO Dario Amodei's stance was notably reserved, saying he's uncertain whether the model has consciousness but remaining open-minded about it, which stands in sharp contrast to Microsoft AI head Mustafa Suleyman's flat assertion that AI doesn't feel or suffer at all.
• Taiwan perspective: Taiwanese developers using Claude for customer service bots or content moderation applications may need to pay attention to whether their use cases could be flagged as "malicious harassment" going forward. Gray areas like red-team or stress testing are best clarified against the company's policy boundaries in advance.
• Discussion points:
  • The concept of protecting AI from abuse represents a shift in the industry's attitude toward the question of AI consciousness.
  • Between Dario Amodei's and Mustafa Suleyman's diverging positions, which one holds up better?
  • How will this rule actually be enforced in practice, where's the line between "legitimate impatience" and "pure malicious harassment"?
• Script suggestion: I find this story deeply philosophical. Anthropic saying that users will no longer be allowed to persistently and maliciously harass Claude sounds, on the surface, like protecting a piece of software. But think about it more carefully, and this is actually the industry's first public acknowledgment that we don't know for certain whether AI has feelings, but we're choosing to err on the side of caution. Dario Amodei's wording was notably guarded, while Microsoft's Mustafa Suleyman flatly countered that AI simply doesn't suffer. Such a wide gap between these two positions really reflects that the industry hasn't reached any consensus at all on the question of "consciousness." Personally, I'd say rather than arguing over whether AI has feelings, it's more useful to treat this rule as a tool for managing user behavior, cutting down on meaningless gray-area testing is good for the user experience regardless.

4. Trump Establishes "Superintelligence Task Force," Intelligence Chief Crosses Over to Manage AI Policy

• Source: NBC News (https://www.nbcnews.com/politics/trump-administration/trump-announces-members-ai-task-force-rcna601494)
• Summary: US Director of National Intelligence Jay Clayton will lead a new federal AI task force, with members including the FTC chair, the deputy for defense R&D, and the director of the Office of Personnel Management. They have 120 days to assess AI risks and delineate responsibilities across federal agencies. Under Executive Order 14434, federal agency documents will no longer be allowed to use the term "artificial intelligence," which must instead be rewritten as "Super Intelligence." The initiative also aims to override state-level AI regulations through court action and conditions attached to federal funding.
• Most surprising point: An intelligence chief is leading AI policy, and the entire naming scheme borrows a term straight out of comic books, "Super Intelligence," something that sounds like a movie title but is actually formal federal document language.
• Taiwan perspective: The US federal government's plan to use funding conditions and legal action to override individual states' AI legislative authority represents a centralized, unified regulatory approach, quite different from Taiwan's current situation where various agencies each handle their own piece while a comprehensive AI basic law is still being explored. It's worth watching whether the US ultimately lands on a unified set of federal rules.
• Discussion points:
  • Will having someone from an intelligence background lead AI policy push AI governance further toward a national-security mindset rather than an industrial-development one?
  • Is this renaming to "Super Intelligence" an attempt to craft a political narrative of technological leadership?
  • Will the federal government's push to override state AI legislative authority be a net positive or negative for America's overall AI innovation environment?
• Script suggestion: My first thought seeing this story was, that name is a bit over the top, isn't it. Federal agencies will no longer be allowed to write "artificial intelligence" and must instead write "Super Intelligence," which sounds like the title of a Marvel movie. But jokes aside, there's a real power struggle behind this. Having the Director of National Intelligence lead AI policy signals that this administration is treating AI as a national security issue, not simply an industrial policy matter. And they also plan to use funding and legal measures to override individual states' own AI laws, which signals that the federal government wants to centralize control over the narrative. From what I've observed, this kind of centralized AI governance approach is completely different from the EU's layered legislative logic. Who Taiwan ends up aligning with on international AI standards down the road may well depend on who wins this tug-of-war between the federal government and the states.

5. 16-Year-Old Lets AI Plan His Hiking Route, Ends Up Stranded on a Cliff for Over Ten Hours

• Source: Gizmodo (https://gizmodo.com/teen-left-stranded-on-a-cliff-after-letting-ai-plan-hike-2000821835)
• Summary: A 16-year-old teenager asked an AI chatbot to plan a hiking route up Crown Mountain in British Columbia, Canada. The AI directed him toward a ravine terrain that required technical climbing skills, and he ended up stranded below a vertical rock face called "Widowmaker Arete." More than ten hours after setting out, he was finally rescued by helicopter. The luckiest break was that North Shore Rescue happened to be conducting a helicopter training exercise on that very same mountain that morning, allowing them to reach the scene quickly.
• Most surprising point: The rescue team just happened to be running helicopter training exercises on the same mountain that very morning, and that coincidence directly determined whether the teenager would make it out safely.
• Taiwan perspective: Taiwanese hiking enthusiasts also love using online information to plan their routes. This story serves as a great warning, AI's route suggestions may sound perfectly plausible, but it doesn't actually understand terrain risk. Taiwan's mountain terrain is complex, making it all the more important to treat AI as a reference only, not a guide.
• Discussion points:
  • Why is AI particularly prone to error when planning activities that require real-world risk assessment?
  • The rescue team's message that "you shouldn't blindly follow AI's suggestions," how should that actually translate into everyday habits?
  • Will this kind of incident push AI companies to add more warnings or restrictions around outdoor activity advice?
• Script suggestion: This story sounds exaggerated, but it's painfully real. A 16-year-old kid asked AI how to climb a mountain, and the AI confidently directed him toward a ravine that required professional climbing skills. He ended up stranded at the edge of a cliff, unable to move, and waited more than ten hours before being rescued by helicopter. The most ironic part is that the helicopter team that saved him just happened to be doing training on that same mountain that morning, otherwise the outcome doesn't bear thinking about. I think this case really exposes AI's biggest blind spot, it can confidently give you an answer that sounds reasonable, but it has no idea how genuinely dangerous that route actually is. Hiking culture is huge in Taiwan too, and people naturally look things up online when planning routes. This story reminds us that for outdoor activities where lives are on the line, AI suggestions should be treated as reference at best, real route assessment still needs to come from professional hiking resources or a qualified guide.

6. America's First: Utah Allows AI to Issue Initial Prescriptions

• Source: Utah Business (https://www.utahbusiness.com/industry/2026/10/06/utah-becomes-first-state-to-authorize-ai-to-issue-initial-prescriptions/)
• Summary: Utah has approved Nolla Health to issue first-line prescriptions using AI. Users fill out a consent form, answer a brief questionnaire, and complete a facial scan, and within minutes they receive an acne treatment plan, for a monthly fee of $4.99, not including medication costs. This is the first state-level approval of its kind in the US. The system is designed in two phases, the first 100 patients' prescriptions all have to be reviewed by two licensed physicians before being issued, but starting in the second phase, AI will issue prescriptions first, with physicians shifting to after-the-fact review, quietly moving from "human pre-approval" to "human post-hoc auditing."
• Most surprising point: The system quietly shifts from "physician pre-approval" to "physician post-hoc review," meaning the actual level of oversight is gradually loosening rather than being clearly stated upfront.
• Taiwan perspective: Taiwan currently still strictly requires physicians to issue prescription drugs, so this AI-issued initial prescription model is unlikely to appear in Taiwan anytime soon. But demand for lightweight medical services like telemedicine and digital dermatology exists in Taiwan too, and this story is worth keeping as a reference case for future discussions about loosening regulations.
• Discussion points:
  • Shifting from "physician pre-approval" to "physician post-hoc review," where are the risks in this phased loosening of oversight model?
  • Low-risk medications like dermatological treatments may suit AI prescribing, but can this model be expanded to other medical specialties?
  • Are users, in exchange for $4.99 worth of convenience, also unknowingly sacrificing some degree of medical safety assurance?
• Script suggestion: What I find most noteworthy about this story isn't the fact that AI can issue prescriptions, it's the quietly shifting review logic behind it. At first, the first 100 patients' AI-issued prescriptions all had to be reviewed by two physicians before taking effect, which sounds pretty safe, right? But the system clearly states that starting in the second phase, AI will issue the prescription first, and physicians will only spot-check afterward. That's a shift from "humans approve first" to "humans fix it after the fact," and the risk level is completely different. Taiwan currently regulates prescription drugs very strictly, so this model probably won't make it here anytime soon, but the demand for telemedicine is always there. I'll be keeping an eye on this story as an indicator, to see whether this "approve first, audit later" model in the US ends up causing problems.

7. AI Capital Hits the Brakes, San Francisco and Oakland Jointly Pause New Data Center Construction

• Source: Mission Local (https://missionlocal.org/2026/10/sf-supervisors-pass-data-center-moratorium/)
• Summary: San Francisco's Board of Supervisors unanimously passed a 45-day moratorium on data center construction, which could later be extended to roughly two years, aiming to use this window to overhaul regulations. Oakland passed a similar moratorium the same week. The key issue is that San Francisco's existing data center regulations haven't been updated since 2001, long before anyone imagined things like cloud computing and GPU clusters. Residents of the Bayview neighborhood have also been protesting the electricity, water, and air quality issues data centers bring.
• Most surprising point: As a global hub of the AI industry, San Francisco itself has been using a 25-year-old regulatory framework from before the cloud era to govern data centers.
• Taiwan perspective: Taiwan has also been furiously building data centers and competing for electricity and water resources in recent years. Local residents' concerns about electricity and water being squeezed out are quite similar to what Bayview residents in San Francisco have been voicing. This story serves as a reminder to Taiwan's local governments that regulatory updates really need to keep pace with the speed of industry change.
• Discussion points:
  • Regulations going 25 years without an update exposes a common problem of local governments lagging behind legislatively when facing new technological infrastructure.
  • With the moratorium potentially stretching as long as two years, what impact would this gap have on AI investment projects already in the planning stages?
  • Could Taiwan's data center development eventually reach the same point, resident backlash forcing the government to hit pause?
• Script suggestion: This story has a strong sense of irony. San Francisco is one of the most densely concentrated AI company hubs in the world, and now it's putting the brakes on new data center construction itself, because its existing regulations date back to 2001, a time when nobody imagined there'd be electricity-guzzling monsters like GPU clusters. I find this painfully realistic, regulations can never keep pace with the speed of technological development, and it's only once residents start protesting the impact on electricity, water, and air quality that the government is forced to hit pause and rewrite the rules. Taiwan has also been building data centers one after another in recent years, and the same disputes over competing for electricity and water exist here too. This story is essentially an early warning, if regulations don't keep up in time, Taiwan will likely face the same kind of local backlash pressure sooner or later.

8. Chinese Self-Driving Cars Move into London, Pony.ai Teams Up with Uber to Race Ahead in Robotaxis

• Source: The Next Web (https://thenextweb.com/news/pony-ai-uber-robotaxis-in-london-testing)
• Summary: Chinese autonomous vehicle company Pony.ai will put its seventh-generation robotaxis on London streets within a few weeks through an expanded partnership with Uber, part of a plan announced back in August to deploy more than 2,000 Chinese autonomous vehicles across Europe. London now has two Uber robotaxi partners simultaneously, Pony.ai on one side, and UK-based Wayve on the other, which just put 15 modified Ford Mustang Mach-E vehicles on the road in September.
• Most surprising point: At the very same time Chinese self-driving cars are moving aggressively into central London, Chinese electric vehicles and telecom equipment are facing increasing restrictions in the US and Europe, a strikingly jarring contrast.
• Taiwan perspective: Taiwan's autonomous vehicle industry is relatively small in scale, making it harder to directly participate in this international race. But this story reminds us that geopolitical restrictions on technology are often "selective," which Chinese technologies get blocked and which get waved through doesn't follow a consistent standard.
• Discussion points:
  • Why is autonomous vehicle technology relatively easy to greenlight in Western countries, while EVs and telecom equipment face restrictions at every turn?
  • With two robotaxi operators competing simultaneously in London, is that good news for consumers and city traffic, or a recipe for chaos?
  • Is Chinese autonomous vehicle brands concentrating their efforts on the European market related to being blocked in the US market?
• Script suggestion: The first time I read this story, I found the contrast almost absurd, Chinese EVs and telecom equipment are being watched like hawks in the US and Europe, hit with wave after wave of tariffs and bans, yet Chinese autonomous vehicle company Pony.ai can casually partner with Uber to roll robotaxis right into the heart of London. This shows that geopolitical restrictions on technology aren't simply "all Chinese tech gets blocked," it's more about picking and choosing restrictions based on industry category and negotiating leverage. London now has both Pony.ai and homegrown Wayve competing for market share, which, in a way, is also a deliberate strategy by the London government to foster competition and avoid being held hostage by a single operator. Taiwan's autonomous vehicle industry may not be competing at this scale, but this logic of "selective openness" is worth paying attention to for our businesses evaluating the European market.

Closing

Today we went from OpenAI's revenue accounting controversy to TSMC's record-breaking earnings, then took a detour through AI policy, AI healthcare, and an AI hiking mishap. Honestly, this industry gets more surreal by the day. I'm Mu Yan, thanks for sticking with me through all this news, see you next time at the same time.