☕️ Apple launches iPhone leasing

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

📱 Apple launches iPhone leasing

💵 X Money is here

🇨🇳 China starts making its own chip machines

🇹🇼 Taiwan detains Nvidia employee

🏠 Apple readies 3 new smart home devices

🤖 Anthropic won't back open-weight AI ban

Plus: 🎁 16 other news you might like, 🧰 6 tools, and 📚 5 papers.

📱 Apple launches iPhone leasing LINK

Apple has rolled out Apple Upgrade in the U.S., a leasing program for iPhone, iPad, Mac, and Apple Watch run with Klarna that replaces both the old iPhone Upgrade Program and iPhone Payments.

Leases start at $17.99 a month for iPhones, $11.99 for iPads and Apple Watch, and $24.99 for Macs, with 12 or 24-month terms for phones and watches and 24 or 36-month terms for iPads and Macs.

At the end of a lease, customers can return the device, keep it by paying the difference up to full retail with no Klarna fee, or trade up, though ending a lease early brings "substantial fees" equal to the remaining payments.

💵 X Money is here LINK

X has launched X Money, a financial services platform inside its app, now open to Premium and Premium+ subscribers in the US who are over 18, after an earlier invite-only beta test.

The service gives users a deposit account, peer-to-peer payments, and a debit card that comes digitally or as a personalized metal version, with links to both Visa and Apple Wallet.

Perks include up to 6% APY, 3% cashback bonuses, no foreign transaction fees, and early direct deposit, and while X Payments isn't an FDIC-insured bank, balances sit at Cross River Bank and are insured.

🇨🇳 China starts making its own chip machines LINK

A Chinese state-backed company has started mass producing deep ultraviolet (DUV) lithography machines, the tools needed to make modern chips, marking a step toward China building its own semiconductor supply without foreign gear.

The Chinese DUV machines are due to reach customers this year, including chipmakers SMIC, Hua Hong, and memory maker CXMT, though reports say they lag behind on performance and reliability compared with market leader ASML.

ASML shares fell 12% starting July 27 on the news, and the Dutch firm had expected to ship about 130 DUV immersion machines this year, even as US Congress recently proposed blocking DUV sales to China.

🇹🇼 Taiwan detains Nvidia employee LINK

Taiwan's prosecutors have detained an Nvidia employee over the alleged illegal export of Super Micro AI servers to China, servers that held Nvidia chips restricted under US export controls, according to Bloomberg and Reuters.

The Keelung District Prosecutors' Office said the suspect, surnamed Chang, was questioned on July 24 after raids on his home and office, then jailed over worries he might flee, destroy evidence, or sway witnesses.

The charges include forgery and breach of trust, and investigators also searched Nvidia's Taipei office in this third round of raids, following May and June operations that detained three people and two Super Micro workers.

🏠 Apple readies 3 new smart home devices LINK

Apple is preparing three new smart home products, according to a Bloomberg report, with a refreshed Apple TV box and an updated HomePod mini expected as soon as this fall, followed by a long-awaited smart home hub.

The new Apple TV and HomePod mini will look like today's models but pack faster chips to run Siri AI, while the hub is slated to arrive between October and early 2027 in two versions, one wall-mounted and one on a base.

The hub features a 7-inch display, a new operating system blending tvOS and watchOS, and facial recognition that identifies who's looking, gauges their distance, and adjusts the interface to show that person's calendar and notes.

🤖 Anthropic won't back open-weight AI ban LINK

Anthropic CEO Dario Amodei clarified that the company will not support a ban on open-weight AI models, distancing itself from a wider industry push to limit how freely these systems can be shared.

The stance follows an open letter shared by NVIDIA founder Jensen Huang on X, which was signed by Meta, Hugging Face, Microsoft, Mistral, and NVIDIA and asked policymakers to avoid "premature restrictions" on open-weight models.

Much of the debate centers on worries that Chinese AI labs are catching up by allegedly stealing intellectual property from American firms, using methods like distillation to copy the behavior of existing models.

Other news you might like

• Amazon’s new satellite network for mobile phones could turn up the heat on SpaceXLINK

• Microsoft rolls out Mythos-armor for enterprisesLINK

• Cursor makes its biggest India push yet ahead of SpaceX acquisition with localized pricingLINK

• YouTube "Hackquires" PeacockLINK

• Amazon winds down most flagship AI models in strategy overhaul, Business Insider reportsLINK

• Core Scientific lands AMD AI deal as bitcoin mining operation winds downLINK

• Hugging Face reportedly plagued with AI models generating adult deepfakesLINK

• Two arrested for checking into esports hotels and stealing RAM, making off with over $2,000 in profit in just one monthLINK

• AI companies are reportedly shredding millions of books after using them to train AI modelsLINK

• Apple ends day as world's most valuable company, passing NvidiaLINK

• ChatGPT starts blocking direct requests to copy an author's styleLINK

• Apple sued after alleged App Store crypto scam cost users $1.8MLINK

• Sam Altman deleted TikTok because he could not stop scrolling. OpenAI killed Sora for the same reason.LINK

• Google’s AI search is rapidly becoming the default, new data showsLINK

• The Robot Data PyramidLINK

🧰 Trending tools

Conduit: deploys AI agents to handle guest messaging and automate internal hospitality operations, streamlining communication across every channel property teams use.LINK

Prefactor: an evaluation layer that scores agent runs in real time, catching quality regressions and drift before they impact production customers.LINK

Cercle: a mood-sharing app that lets close friends see how you're feeling via a home-screen widget, prompting check-ins before someone has to ask.LINK

SUB/WAVE: converts your personal music library into a self-hosted radio station with an AI DJ that selects tracks, announces idents, and handles listener requests via natural language.LINK

qsa.sh: runs a one-command external security scan of your server's public IP, using naabu, nmap, and nuclei to detect open ports, service versions, and known CVEs in seconds.LINK

Cardzen: tracks credit card perks, credits, and points across your wallet, sends expiration reminders, and recommends which card to use per purchase.LINK

📚 Trending papers & reports

AI agent permissions get a new checkpoint that blocks tool actions, like exports or deletes, unless they match what the user actually asked for, even if the app's credentials technically allow it.LINK

Task-level model routing assigns one AI model per whole task instead of per step, lifting accuracy by up to 8 points while cutting response time by 36% versus using a single model.LINK

Chatbot problem-solving limits stem from training text only partly imitating human thinking, so basic chatbots can't match human cognitive flexibility, and further scaling won't fix this, despite massive real-world adoption.LINK

Multi-turn AI agents trained by blending fast copy-the-expert learning with trial-and-reward learning boost task success rates by 3.03 points over pure copying and 23.62 points over standard reward-only training, even beating the expert teacher by 2.16 points.LINK

Bias correction tools for hiring or lending algorithms work more reliably when tuned to match error rates across groups rather than just equalizing approval rates, a distinction that changes which "fair" model you'd actually deploy.LINK

See you tomorrow for a new dose of ☕️ Techpresso!