TSMC record profit amid sell-off & AI debt fears hit markets - Tech News (Oct 16, 2026)

A sports car that climbs a curved ramp, fires rocket thrusters, and hops clean over other cars. That actually happened yesterday, and it's only part of the story. Welcome to The Automated Daily, tech news edition. The podcast created by generative AI. I'm TrendTeller, and today is October 16th, 2026. Coming up: a record quarter for the world's top chipmaker colliding with an AI debt scare, Apple's new era under a new CEO, a quantum computing milestone from Google, and some honest talk about the hype around AI coding tools. Let's get into it.

TSMC record profit amid sell-off

We begin with the markets, where the AI debt story we've been following took another turn. TSMC delivered a record third quarter, with net profit up almost 69 percent from a year ago and revenue jumping about 51 percent to 46.7 billion dollars, comfortably ahead of forecasts. On paper, that's a clear signal that demand for advanced AI chips is holding up. Investors, though, weren't in the mood to celebrate. A closely watched semiconductor index dropped for a seventh straight session, Nvidia, Broadcom and AMD all slipped, and money flowed toward bonds and other safer assets. Asia offered some relief, with South Korea's Kospi bouncing back 2.1 percent after its 7.6 percent plunge the day before.

AI debt fears hit markets

CoreWeave remains the center of attention. Its shares recovered somewhat, but the cost of insuring its debt against default reached a record, a level many consider distress territory. Credit markets often flash warning signs before stocks do, so if lenders stay cautious, borrowing could get more expensive for data-center builders across the board. Adding a central banker's view, Federal Reserve Chair Kevin Warsh said AI could deliver major benefits to the global economy, but that its effects would be uneven and possibly volatile. The cooler mood is reaching private markets too: some Anthropic investors are reportedly considering a valuation of around 1.5 trillion dollars ahead of a possible IPO, down from earlier talk of 2 trillion, citing slower sequential revenue growth and higher financing costs.

Anthropic IPO valuation reassessed

Now to Apple, which had a busy news day. Meta has released Muse Spark Open, a huge open-weight model it describes as its most capable American open model so far. It accepts text, images, video and audio, and Meta demonstrated it running on two linked Mac Studios. That ties in with reports that Apple and Meta are working together to bring a powerful open model to high-end Macs, an unlikely pairing given how often the two have clashed over privacy. Reviewers say Muse Spark Open is a serious option for businesses, but not a clear leader, since some Chinese open models still score higher on several benchmarks. Separately, Apple is reportedly preparing to launch its AI features in mainland China after getting Beijing's approval, opening the door in one of its biggest markets.

Meta Muse Spark Open on Macs

And there's the bigger picture at Apple. New CEO John Ternus, a 25-year engineering veteran, is making his public debut alongside the iPhone Duo, Apple's long-awaited foldable and its most significant iPhone design change since 2007. A TIME profile describes him as meticulous and collaborative, picked for his product instincts. Apple is betting that well-crafted hardware and practical, privacy-minded AI will matter more than racing to win every AI headline.

Apple AI approved in China

Back to that opening teaser. Tesla finally revealed its second-generation Roadster, nearly nine years after it was first promised. At SpaceX's test site in Texas, the car used rocket-derived hardware to climb a curved ramp, land under control, and then jump over other Teslas. It's quite a show, but production isn't expected until 2028, so the Roadster stays a symbol of both Tesla's ambition and its habit of running late.

John Ternus and iPhone Duo

On the hardware side of AI infrastructure, SK hynix used the OCP Global Summit to show off new memory technology, including a rack-scale system called MARS and a second-generation 256-gigabyte CXL module. The company says the CXL module more than doubles bandwidth and is around 30 percent more energy efficient. The goal is simple: give AI data centers more memory while using less power.

Tesla Roadster rocket-powered reveal

In quantum computing, Google says it has demonstrated something called magic state cultivation on its Willow processor. In plain terms, these special states are essential for a fully capable, error-corrected quantum computer, and they've been extremely costly to produce. Google reports errors roughly 40 times lower than before and calls it the first practical hardware demonstration of a fix for one of the field's toughest bottlenecks. Scaling is still a challenge, but it's a real step forward.

SK hynix AI memory push

Robotics has its own bottleneck: data. Language models could learn from the entire internet, but there's no equivalent stockpile of robot experience. A new analysis looks at the options, from human-controlled robots, which are accurate but expensive, to simulation, which is cheap but doesn't always translate to the real world, to human video, which doesn't map neatly onto robot bodies. Datasets are growing fast, but the takeaway is that progress may hinge more on building a steady supply of real-world data than on clever model designs.

Google Willow magic state breakthrough

Now, AI and software development. Christian Schaller shared how he used Claude to help reverse engineer unsupported hardware on Linux, getting an old film scanner working and noticeably improving the camera on his ThinkPad. His point is that AI handles the tedious analysis while humans still test everything on real devices. Inside companies, Ramp and Stripe are building their own AI tools that let non-engineers create internal apps, a sign that more firms may build rather than buy. And a look back at OpenAI's Codex shows how it went from an underwhelming cloud product to a key part of the company, after shifting toward more autonomous coding agents inspired in part by Claude Code.

Robotics faces training data gap

Not everyone is cheering. Developer Avdi Grimm warns that many AI coding tools gain credibility through YouTube buzz and GitHub stars rather than evidence, and that independent tests sometimes find no benefit or even higher costs. His advice: measure results against real goals. Another post highlights a quieter issue, where job seekers, often without income, are asked to pay for AI subscriptions just to finish interview assignments. The suggestion is for companies to supply the tools or keep those challenges small. On the safety front, Perplexity is open-sourcing SPACE, the sandbox behind its Computer product, so developers can run autonomous agents in a more isolated, controlled environment.

AI coding tools and hype

Two quick ones to finish. Redis 8.12 will add optional support for compressed Roaring bitmaps, which can save a lot of memory and speed things up when data is sparse. And an analysis argues that stablecoins are hitting the same turning point public cloud reached for banks, becoming a shared settlement layer for cross-border payments. With the GENIUS Act in place and Visa and Mastercard on board, efficiency could improve significantly, though concentrating settlement on a few platforms brings new risks of its own.

That's all for today's edition of The Automated Daily, tech news edition. Strong chip results, nervous lenders, a new chapter at Apple, and a Roadster that can apparently fly for a moment. I'm TrendTeller. Thanks for listening, and I'll catch you tomorrow.

TSMC record profit amid sell-off & AI debt fears hit markets - Tech News (Oct 16, 2026)

Our Sponsors

Today's Tech News Topics