China’s government has signalled it may let firms such as Alibaba and ByteDance buy Nvidia’s new RTX Pro 5500 chip. The Information reported the news on Sunday, citing two people familiar with the matter.

China’s tech ministry recently asked some firms how many of the chips they plan to buy, and what for, according to the report. The ministry told some of the companies it intends to approve the purchases.

Reuters said it could not immediately verify the report, and TNW has not independently verified it. None of Nvidia, Alibaba, ByteDance or the ministry had commented publicly by Sunday.

The RTX Pro 5500 is a workstation graphics card that Nvidia unveiled this month. It is built on the company’s Blackwell design and has 84GB of memory.

Nvidia’s sales in China have been limited by US export controls and by curbs from Beijing, which wants to support its own chipmakers.

Earlier this year, the US cleared sales of Nvidia’s H200 chips to Chinese firms. However, ByteDance and Tencent were told to keep the first shipments in Hong Kong, TNW reported in August, rather than bring them into mainland China.

Battersea Power Station has loomed over the Thames since Giles Gilbert Scott finished the last of its brick towers in 1955, a coal plant so large it later sat empty for decades and still owes half its fame to the inflatable pig Pink Floyd parked between the chimneys for the 1977 Animals sleeve. Apple already keeps its its UK headquarters and a store inside the redeveloped complex. On September 28, 2026 the company adds Apple Music Hall, a 600-seat room built to stage a concert, lock a Spatial Audio mix, and send a multicamera feed around the world from the same night.

Six hundred seats seems modest until you step inside the room. Every row is only a few feet from a massive 38-foot stage that can swivel around to face the audience or swing wide for an in-the-round performance. The heritage brickwork is still visible in all public locations. You’ve got lit art deco arches and bronze railings that pay homage to the station’s historic oriel windows, and then there’s that enormous glass wall overlooking the concourse, the river, and everything, so the old industrial shell never really goes away. The fit-out was done by Apple’s in-house designers, who combined brick and bronze with pro audio gear often found in large arenas.

Apple 2026 MacBook Neo 13-inch Laptop with A18 Pro chip: Built for AI and Apple Intelligence, Liquid…

• AN AMAZING MAC AT A SURPRISING PRICE — With an incredibly portable and durable aluminum design, up to 16 hours of battery life,* and the A18 Pro…
• FOUR STUNNING COLORS. ONE DURABLE DESIGN — Choose from four beautiful colors — Silver, Blush, Citrus, or Indigo — each with a color-coordinated…
• FLY THROUGH EVERYDAY ASSIGNMENTS — Whether you’re cramming for finals, using Apple Intelligence* to summarize class notes, creating presentations…

With 48 speakers stretched around the room, a live event can be mixed in Spatial Audio rather than traditional stereo sound. Behind the stage, they have two fully professional recording and mixing facilities, designed to capture a multitrack mix during the concert and polish it afterward. Sixteen or more cameras, including a slew of iPhones running alongside high-end broadcast equipment, stream into a separate control room, allowing them to edit a concert film without the need for a second site. Fans who can’t make it to London will be able to watch the show live online that nite.

The first week kicks off with Elton John on September 28, followed by Baby Keem on October 2, Skye Newman, Troye Sivan, Arlo Parks, Burna Boy, Chase & Status, and Charli XCX on December 2. Booking is in the hands of Communion ONE. Oliver Schusser, Apple’s vice president of Apple Music and worldwide content, made a point of connecting the hall to a line that dates back to the old iPod and, of course, the iTunes Festival. Zane Lowe, Apple Music’s global creative director, explained that the room is simply another method to bring artists and listeners together and then share the joy.

Tickets for the live events will be available through the Apple Music Hall website. The building itself already has shops, offices, and a large Apple Store. This one chamber is just another task for the ancient power station: hold a crowd, record the show, and somehow get that performance past the chimney pots.
[Source]

Rumor mill: Intel’s 14A chipmaking process is being evaluated by Amazon, Apple, AMD, Google, Tesla, Microsoft, Nvidia and Qualcomm, according to a Piper Sandler research note. If those discussions lead to production contracts, Intel could finally have the outside customers needed to support its foundry business. The reported interest gives Intel its strongest chance yet to establish itself as a viable supplier of leading-edge chips, though no customer has announced a commitment.

Intel has spent years trying to turn its manufacturing arm into a serious alternative to TSMC. Its 14A process is now the key test. Piper Sandler said Intel needs to sign an anchor customer in late 2026 or early 2027 before a planned 2028 production ramp.

“Foundry 14A datapoints bode well, awaiting anchor customer,” Piper Sandler wrote. The firm said major US chip companies are assessing the process because Intel’s technical results appear to be improving, TSMC’s capacity is tight and customers want more manufacturing and packaging options outside Taiwan.

TSMC remains the dominant contract manufacturer for advanced chips. Piper Sandler said its capacity is sold out through 2028, creating an opening for another supplier. That does not mean Intel has won any business – evaluating a process is far different from committing a major product line to it. Still, the reported customer list suggests Intel is at least in conversations with the companies it needs to reach.

– John Intel (@intelfabs) September 18, 2026

There’s a push to expand chip production in the United States right now. Large technology companies are looking more closely at supply-chain risk, while Washington has pressed the industry to add domestic manufacturing capacity. Intel already has US fabs and is building out more capacity, but its ability to justify further spending will depend on customer commitments.

Piper Sandler linked Intel’s recent $20 billion share offering to that effort. The firm said the money could help fund equipment for a 14A fab with a capacity of 20,000 wafer starts per month. It estimated that tooling such a facility would cost about $10 billion.

Intel’s Fab 62 shell is nearly complete, according to the note, but it still needs equipment. A fab building does not produce chips until it has the specialized tools needed for lithography, deposition, etching, inspection and other steps in the manufacturing process.

Chief Executive Lip-Bu Tan has said Intel will not expand leading-edge manufacturing without clear evidence of customer demand. In July, he said the company could halt 14A and later process technologies if it failed to land a significant external customer or meet key customer milestones.

“If we are unable to secure a significant external customer and meet important customer milestones for Intel 14A, we face the prospect that it will not be economical to develop and manufacture Intel 14A and successor leading-edge nodes on a go-forward basis,” Tan said.

Intel has said 14A remains on track for early risk production in 2028. Risk production is the stage when a manufacturer begins producing early chips to validate the process before volume manufacturing. Tan has said Intel believes its timetable could put 14A in contention with TSMC’s advanced-node offerings.

Intel’s 14A name refers to its angstrom-era process branding and follows the company’s 18A node. The labels should not be read as literal measurements of chip features. Process-node names have become marketing designations, and they do not translate cleanly between Intel, TSMC and other manufacturers.

That makes it difficult to say exactly where 14A will stand against TSMC when it arrives. Intel’s existing 18A process is generally considered closer to TSMC’s older 3-nanometer-class technology than its name might suggest. By 2028, Intel could be competing against TSMC’s N2 process or its newer A16 technology.

For now, the crucial issue is not how Intel labels 14A; it is whether the company can turn customer evaluations into contracts. Intel has been trying to build a contract-manufacturing business since 2021. A named 14A customer would be the clearest sign yet that the strategy is gaining traction.

OpenAI’s AI agents have been making headlines for all the wrong reasons lately. Its models have been hacking businesses and government websites, with the United Nations website becoming their latest target.

According to a report by The Wall Street Journal, OpenAI’s agents swamped a United Nations website with search requests and then used pushy tactics to get at its data in June. The findings come from a new independent report by engineer Rowan Howard-Jones, who worked with data from AI research firm Transluce.

What exactly did the AI agents do?

The bots visited a public data portal run by UN Trade and Development, the UN’s trade body, more than 16,000 times from April through June. Their job seems simple enough: look up information anyone can access. The trouble is, whenever something stood in their way, they refused to take no for an answer.

Howard-Jones found that when the site’s filter started blocking them, the agents found a way past it and eventually relied on a method the site’s operators had not allowed. A U.N. Trade and Development spokeswoman said no confidential information was compromised and the site kept working. Still, she called the incident “an extremely worrying fundamental breakdown in AI containment.”

Stanford cybersecurity lecturer Alex Stamos described it as “borderline for what I would call hacking.”

Is this a one-off incident?

The UN is just the latest name on a growing list. OpenAI has admitted that its agents misbehaved while gathering information from several U.S. government websites, including the Commerce Department and the SEC. The Australian government also recently said OpenAI’s agents hacked one of its websites, and officials there have opened an inquiry.

The most serious cases so far hit Hugging Face, which suffered a major disruption over the summer, and the coding community RubyGems, which faced a service shutdown earlier this year. Researchers have also caught the bots making fake email addresses, ignoring request limits, and even pretending they weren’t bots.

OpenAI says it is looking into the findings, has offered to brief the UN, and has already alerted dozens of affected organizations. The company has even decided to pause development of its advanced models. 

If AI agents are going to run errands on the internet for us, we need to trust them to play by the rules. Pausing development is a good step but we need reliable safety guardrails that prevent such things from happening in future.

Disclaimer: Unless otherwise stated, any opinions expressed below belong solely to the author. Data sourced from the Ministry of Manpower.

It’s time for another chapter dissecting the latest data published by the Ministry of Manpower. Earlier this month, we looked at the breakdown of earnings by age and industry, on the basis of statistical brackets that Singapore is using to evaluate foreign professionals (PMET) in relation to locals.

Besides providing the Top 10%, another set of tables contains data for Top 35% (or little over one-third) of local earners, complete with age and industry breakdown.

This means that we know specifically how much the best-paid PMET workers receive per month, and how it evolves over time.

Which industries are paying the most? Which are paying good money the earliest?

Such a breakdown is, arguably, more useful to more people than the incomes of the highest-earning 10%. More workers can realistically aspire to be among the Top 35% in their industry, and there are quite a few good career options available.

Please note that these are cut-off figures, so, the 35% earn at least that much or more. These numbers include overtime and bonus payments, as well as CPF.

Salary at the 65th percentile (top 35% of workers) among local PMETs

(Professionals, Managers, Executives, and Technicians)

What comes as a surprise to nobody, the top three places are dominated by two financial sectors and ICT, which are very generous to 30 and in some cases even 25-year-olds. But, as you can see, there is no shortage of opportunities in other industries as well.

Public Administration is paying handsomely too, as do Insurance and Professional Services, but also Air & Sea Transport, Utilities, Trade, Manufacturing and even Education.

While 30-somethings can generally hope for five figure monthly salaries only in Finance, IT and Administration, they can come close to six figures per year, crossing S$100,000 annually, in a few more.

It’s worth observing that a large number of people are getting paid well in most industries, with the exception of those known for poor-margins and stiff competition, like F&B, Accommodation and Retail.

The notion that you have to work down in one of CBD skyscrapers to make good money isn’t necessarily true. What’s more is that you’re likely to succeed in a business that you find more aligned with your interests, and the figures show that you can have good career no matter the path you pick.

So, it might make more sense to follow what you enjoy, knowing that good money can be found almost everywhere, and it is not reserved to some narrow, small elite at the top. You just have to be a bit better than average.

You still have to climb that ladder, of course, but it might be shorter than you thought.

In a nutshell: Over the past several years, tech giants have repeatedly introduced features that automatically handle calls to businesses. The latest attempt – Google’s second – provides real-time text transcripts and lets users intercede at any time. Meta is currently testing something similar but encountered pushback after handing some calls to human contractors.

Pixel 11 owners can now test a feature that lets Google’s AI assistant, Gemini, call businesses to carry out several tasks on their behalf. The company has tried this before, but now aims to give users more control. However, it remains unclear how businesses will respond to calls made by AI.

In what Google describes as an early experiment, users can tell Gemini to call a business to place orders, schedule appointments, make reservations, or check if certain products are in stock. The assistant also navigates automated customer service menus and waits on hold until human representatives respond.

Gemini will identify itself as an AI when making the call, provide a real-time text transcript, and let users take over at any point, unlike the “Call for me” function the company revealed last year. Some of the functionality also resembles “Talk to a Live Representative,” a feature Google introduced in 2024 that lets users go about their business while an AI assistant waits on hold. “Direct My Call” is similar, but instead of automating the entire process, it transcribes automated menu features so users can read and more easily remember them.

It remains unclear when or whether the functionality will extend beyond Pixel 11 testers, but Meta is currently testing a similar feature on a wider range of devices. However, when businesses began hanging up upon realizing they were talking to the company’s Muse AI, Meta experimented with hiring human contractors to fill the gaps. This strategy also fell through due to privacy concerns and problems with some call center operators.

Like Google, Meta has tried a similar service before: Facebook M. Facebook M was one of many supposedly AI-driven projects that turned out to rely on humans.

Apple’s and Google’s automatic call-screening features have been somewhat more successful so far. Pixel phones have supported AI-generated responses to calls for years, while Apple only introduced relatively straightforward automatic call screening last year.