Opening

The White House signed an executive order today ordering federal agencies to rename AI as "Super Intelligence", but the rest of the world is still calling it AI. The same week, a Qualcomm-powered robot fell flat on its face during a Computex keynote, and Anthropic quietly made Claude's bills a little lighter. Coming up: who AMD just spent $8.2 billion to acquire, and why OpenAI would rather delay its IPO by six months.

Today's Top Stories

1. Trump Signs Executive Order Rebranding AI as "Super Intelligence"

• Source: Forbes (https://www.forbes.com/sites/siladityaray/2026/09/30/trump-signs-order-renaming-ai-to-super-intelligence--heres-what-it-says/)
• Summary: The White House signed an executive order requiring all federal agencies to refer to AI going forward as "Super Intelligence," or SI for short. According to Forbes, the order also directs tech advisors to draft an official definition of SI. But the order only governs the government's own documents, industry, academia, and everyday usage worldwide remain completely untouched.
• Most surprising bit: The order can only change the wording in government documents, the rest of the world, including industry and academia, is still calling it AI. It's essentially the administration playing a renaming game with itself.
• Taiwan angle: For Taiwan's AI and semiconductor industries, this kind of rebranding order has essentially zero real-world impact. What actually matters is chip orders and regulatory developments, no need to follow the U.S. government's lead on terminology.
• Discussion points:
  • Can an executive order really change an industry's established vocabulary?
  • Is "inventing an official definition" an attempt to use language to defuse public concern about AI risk?
  • Should Taiwanese companies and media adopt the term "SI"? (Probably not.)
• Script suggestion: The White House just issued an order saying federal agencies can no longer say "AI", they all have to say "Super Intelligence" instead. I genuinely laughed when I saw this. But the order only covers the government's own press releases and websites. Google, OpenAI, TSMC, and pretty much everyone scrolling their phones and talking about AI, nobody's paying attention to this new term. This feels less like policy and more like a language PR stunt, because renaming something doesn't make a single chip faster or a single model smarter.

2. Trump Calls Tech Industry AI Agreement "Morally Binding"

• Source: CNBC (https://www.cnbc.com/2026/09/29/tech-white-house-ai-lunch-trump.html)
• Summary: According to CNBC, tech titans including Musk, Zuckerberg, and Jensen Huang dined at the White House this week and signed an AI safety agreement covering internal controls and external audits. Trump described it as "morally binding", which, in plain terms, means it carries zero legal weight. At the same event, he reiterated that the administration won't hit the brakes on AI and would rather let industry police itself.
• Most surprising bit: An agreement meant to govern AI safety, where the signatories themselves openly admit it has no legal force, essentially a handshake deal dressed up as policy.
• Taiwan angle: If Taiwanese AI companies are considering this kind of self-regulation model, they should think carefully, an agreement with no legal teeth won't protect users, or the company itself, when something actually goes wrong.
• Discussion points:
  • Is the phrase "morally binding" essentially rhetorical sleight of hand?
  • Has self-regulation ever historically worked in a high-risk field like AI?
  • If the agreement has no legal force, what are the signatory companies actually getting out of it? (Most likely: a PR gesture.)
• Script suggestion: This one struck me as pretty ironic. A group of the world's wealthiest and most powerful tech CEOs sit down and sign a safety agreement with the President, and the President himself immediately clarifies it's only morally binding, not legally binding. In plain English: everyone shook hands and promised to behave, but nobody signed anything with actual teeth. Whether this kind of agreement does anything for people who genuinely care about AI risk, I have my doubts. But for stock prices and public image? Absolutely useful, because once your name is on a signed document, it's harder to walk it back later.

3. This Week in Tech Policy: Bipartisan Bill Aims to Pause AI Ahead of New Regulatory Agency

• Source: Nextgov/FCW (https://www.nextgov.com/policy/2026/09/tech-bills-week-creating-ai-focused-agency-reviewing-ai-assisted-cyber-attacks-and-more/416253/)
• Summary: In the same week the White House was championing industry self-regulation, a bipartisan group of lawmakers in the House and Senate jointly introduced a bill on September 25 arguing that development of more powerful AI capabilities should be paused until a new dedicated AI regulatory agency is established. That's two completely opposite scripts coming out of the same federal government at the same time.
• Most surprising bit: The White House says let industry regulate itself; Congress wants to hit pause and wait for a new agency to be created first. It's rare to see this level of internal disagreement within a single administration's approach to AI oversight.
• Taiwan angle: This kind of internal U.S. policy tug-of-war has limited short-term impact on Taiwanese companies, but if Congress eventually passes legislation creating a regulatory body, chip export rules and model evaluation standards could bring a fresh round of compliance costs, something the supply chain should watch closely.
• Discussion points:
  • Will the White House's approach and Congress's approach end up blocking each other?
  • Is "pause development until the agency exists" actually enforceable in practice?
  • Should Taiwan's tech industry be watching executive orders or congressional legislation more closely? (Usually the latter has more lasting impact.)
• Script suggestion: My first reaction reading this was, does the U.S. government even talk to itself internally? The White House just hosted tech CEOs for dinner and got them to sign a legally toothless agreement emphasizing a hands-off approach, and now Congress immediately fires back with a bill demanding a pause on development until a new agency is stood up. Completely out of sync. This kind of internal contradiction is actually a headache for industry, companies don't know which set of rules to prepare for. If you're an AI startup trying to build out operations in the U.S. right now, you're probably stuck watching both sides until an actual piece of legislation lands.

4. AMD Acquires Fei-Fei Li's World Labs for $8.2 Billion

• Source: TechCrunch (https://techcrunch.com/2026/09/28/amd-will-acquire-fei-fei-lis-world-labs-for-8-2-billion/)
• Summary: AMD announced an all-stock acquisition of World Labs, the spatial intelligence startup founded by Fei-Fei Li, for $8.2billion\mathrm{.}ThisisAM{D}^{\mathrm{\prime }}ssecond-largestacquisitionever,trailingonlytheroughly8.2\; billion.\; This\; is\; AMD\text{'}s\; second-largest\; acquisition\; ever,\; trailing\; only\; the\; roughly$50 billion Xilinx deal in 2022. Once the deal closes, Li will join as Executive Vice President and Chief Scientist, reporting directly to AMD CEO Lisa Su.
• Most surprising bit: The real headline isn't the $8.2 billion price tag, it's that Fei-Fei Li, who built ImageNet and essentially laid the groundwork for modern deep learning, will now report directly to Lisa Su. Two of AI's most influential women, now under the same roof.
• Taiwan angle: This move signals AMD's ambition to compete with Nvidia in physical AI, robotics and autonomous vehicles. For Taiwan's server and robotics component supply chain, that opens up new order opportunities worth tracking as AMD's hardware strategy unfolds.
• Discussion points:
  • How does "spatial intelligence" differ technically from the language-model-centric AI dominating headlines today?
  • Is AMD's big spend on talent and technology an attempt to close the gap with Nvidia's AI ecosystem?
  • What does putting a top-tier talent like Li in a direct reporting line to Lisa Su signal about the organization's priorities?
• Script suggestion: What strikes me most about this story isn't even the dollar figure, it's the human element. Fei-Fei Li built ImageNet, which is basically one of the foundations of the entire deep learning wave we're living through. Now she's joining AMD, reporting directly to Lisa Su, which means two of the most respected leaders in AI are now working at the same company. AMD isn't just buying World Labs' spatial intelligence technology, they're buying a figure who can rally talent and help define strategic direction. For a company trying to catch up to Nvidia in physical AI, that's a genuinely smart move.

5. Anthropic Launches Claude Sonnet 5.5, Positioned as a Faster, Cheaper Work Partner

• Source: TechCrunch (https://techcrunch.com/2026/09/28/anthropic-releases-sonnet-5-5-which-it-calls-a-significantly-cheaper-faster-work-partner/)
• Summary: Anthropic released its next-generation model, Claude Sonnet 5.5, with pricing unchanged, $2permillioninputtokens,2\; per\; million\; input\; tokens,$10 per million output tokens, but output speed is up more than 30%, and per-task costs can drop by up to 30%. The savings don't come from a discount; they come from the model itself learning to cut unnecessary chatter and skip unnecessary tool calls, getting the job done with fewer tokens.
• Most surprising bit: The cost savings here aren't a pricing strategy, they're the model getting smarter and that improvement showing up directly on the bill. Efficiency itself has become the product.
• Taiwan angle: For the many Taiwanese dev teams and content production pipelines already using Claude in their workflows, this same-price-but-fewer-tokens upgrade is close to a painless swap, just switch models and start saving. Worth testing in your own pipeline soon to see how much it actually saves.
• Discussion points:
  • Could "the model getting smarter shows up on your bill" become the next benchmark AI vendors compete on?
  • For companies making heavy use of the API, token efficiency gains are effectively a hidden price cut.
  • How does Sonnet's practical, affordable positioning differentiate it from flagship models?
• Script suggestion: This one genuinely resonates with me, because our content pipeline relies heavily on Claude for large-scale text processing. Anthropic didn't cut prices this time, the price sheet is unchanged, but by getting the model to cut the chatter and stop calling tools it doesn't need, they shaved up to 30% off the tokens needed to complete the same task. For teams making heavy use of the API, that's effectively a hidden price cut. Compared to a flashy press release announcing a price drop, I think baking efficiency directly into model behavior is both more practical and much harder for competitors to copy overnight.

6. OpenAI Seeks $30 Billion Funding Round While Shelving IPO Plans

• Source: Yahoo Finance (https://finance.yahoo.com/technology/ai/articles/openai-targets-30-billion-mega-203449336.html)
• Summary: According to Bloomberg, OpenAI is seeking a new funding round of at least $30billion,atavaluationofroughly30\; billion,\; at\; a\; valuation\; of\; roughly$1.4 trillion, while pushing back its previously planned IPO. Altman's stated reason: going public right now would be "ill-advised" because the company is still working through AI safety-related issues. Back in March, the valuation was $852 billion, meaning it's climbed roughly 64% in six months.
• Most surprising bit: A company about to ask the market for $30 billion is citing "not ready to be scrutinized by public markets" as its reason for delaying an IPO, which sounds a bit contradictory on its face.
• Taiwan angle: OpenAI's continued fundraising binge is, in a way, pushing up the global ceiling for AI infrastructure demand, chips, servers, power, which remains a sustained, long-term tailwind for TSMC and Taiwan's server supply chain.
• Discussion points:
  • Is a 64% valuation jump in six months a bubble warning sign?
  • Could "not ready for public market scrutiny" be masking financial or safety issues more complicated than they appear?
  • How does delaying the IPO affect employee stock liquidity and corporate governance transparency?
• Script suggestion: My first thought reading this was, $30billion,tha{t}^{\mathrm{\prime }}snotasmallnumber,andthevaluationjumpedfrom30\; billion,\; that\text{'}s\; not\; a\; small\; number,\; and\; the\; valuation\; jumped\; from$852 billion to $1.4trillioninjustsixmonths,a641.4\; trillion\; in\; just\; six\; months,\; a\; 64\%\; increase\; that\; most\; companies\; wouldn\text{'}t\; dare\; dream\; of.\; But\; I\; find\; Altman\text{'}s\; reasoning\; for\; delaying\; the\; IPO\; a\; bit\; curious,\; saying\; going\; public\; would\; be\; unwise\; because\; they\text{'}re\; still\; dealing\; with\; AI\; safety\; issues\; sounds\; like\; "we\text{'}re\; not\; ready\; to\; be\; put\; under\; a\; microscope."\; Except\; they\text{'}re\; about\; to\; ask\; private\; markets\; for$30 billion, which means in some sense they're already being scrutinized, just through private deals, without the disclosure pressure that comes with public financial reporting.

7. Qualcomm-Powered Humanoid Robot Falls Over During Computex Keynote

• Source: Tom's Hardware (https://www.tomshardware.com/tech-industry/robotics/qualcomm-powered-robot-collapses-spectacularly-on-stage-during-presentation-prepared-stagehands-rush-to-cloak-and-then-carry-off-stricken-humanoid)
• Summary: At the Computex 2026 keynote held in Taiwan, a Neura Robotics humanoid robot powered by Qualcomm's Dragonwing IQ10 chip walked onto the stage, threw a symbolic fist pump, and then collapsed face-first, unable to get back up. Staff rushed over, covered it with a cloth, and carried it off stage. Qualcomm's official explanation afterward: this was actually the robot's built-in safety fall-down procedure functioning normally, and it even provided valuable real-world data.
• Most surprising bit: The face-plant on stage was already awkward enough, but the official spin, framing it as valuable real-world data, ended up being the funniest part of the whole story.
• Taiwan angle: This mishap happened right on Taiwan's home turf at Computex, making it especially relatable for local audiences, and it's a good reminder that despite all the hype around humanoid robots right now, they're still a fair distance from truly stable, commercial-grade reliability.
• Discussion points:
  • Is the "safety fall-down procedure" a genuine engineering feature, or just PR damage control?
  • Why is there often such a gap between polished promotional videos and live demo reality for humanoid robots?
  • How much reputational damage does a live on-stage failure like this do to Qualcomm's push into the robotics chip market?
• Script suggestion: I couldn't stop laughing watching this one, a robot takes a few steps and then face-plants right on stage, and staff rush over to cover it with a cloth and carry it away. That image speaks louder than any bad review. What really seals it is Qualcomm's official response afterward, calling it a built-in safety fall-down procedure and even claiming they collected valuable real-world data, which sounds like textbook crisis PR. I think this is a good reminder that a lot of humanoid robot promo videos are polished and slick, but a live demo with no chance for a retake is really the true test of whether these robots are actually ready.

8. TSMC's AI Business Hits a Surprise Milestone

• Source: Yahoo Finance (https://finance.yahoo.com/technology/ai/articles/taiwan-semiconductors-ai-push-hits-202554417.html)
• Summary: TSMC's September revenue grew 53.3% year-over-year, driven by AI demand so strong that supply can't keep up. Monthly capacity for its 3nm process could hit 180,000 wafers as early as the beginning of Q4, two to three months ahead of prior expectations. The surge is driven primarily by follow-on orders from Nvidia, AMD, and Broadcom.
• Most surprising bit: The comparison is brutal, Nvidia's gross margin for this fiscal year sits at a staggering 71.1%, yet it doesn't manufacture a single chip itself. The company actually pouring blood, sweat, and tears into production is TSMC.
• Taiwan angle: This is the story that hits closest to home for us, TSMC running at full capacity is a direct reflection of Taiwan's irreplaceable position in the global AI supply chain. But it also means sustained pressure on local power, talent, and expansion capacity, a sweet but heavy burden.
• Discussion points:
  • What does capacity for 3nm filling up two to three months ahead of schedule mean for TSMC's upcoming capex plans?
  • Is it fair that Nvidia's 71.1% margin so vastly outstrips TSMC's foundry margins? What does that say about how value is distributed in the supply chain?
  • As AI chip demand keeps outpacing expectations, can Taiwan's power grid and water infrastructure keep up?
• Script suggestion: This is a story I think everyone in Taiwan should be paying attention to. TSMC's September revenue grew over 50% year-over-year, and 3nm capacity is set to hit full utilization two to three months ahead of schedule, a clear sign that AI chip demand shows no signs of cooling down. But what I keep coming back to is the margin comparison: Nvidia, purely through chip design, pulls a 71.1% gross margin, while the company actually etching the transistors and burning through massive amounts of power and water is TSMC. The value split clearly doesn't tilt toward the manufacturing side. Taiwan gets to enjoy this wave of orders, but at the same time, we need to start asking whether our power grid and talent pool can keep up with an even more aggressive pace of expansion ahead.

Closing

Today we went from the White House's absurd order to rename AI "Super Intelligence," to a Qualcomm robot taking a spectacular tumble on the Computex stage, all the way to the real industry-shaping stories, AMD's big-money hire of Fei-Fei Li and TSMC's capacity running ahead of schedule. The story of AI is always this mix of policy farce and technical seriousness happening side by side. Thanks for listening to today's episode of Mark's Tech Insights. I'm Muyan, see you next time.