The M25 – the original “road to nowhere” – around London was completed 40 years ago. Is it a success? CC-licensed photo by Timo Newton-Syms on Flickr.
You can sign up to receive each day’s Start Up post by email. You’ll need to click a confirmation link, so no spam.
A selection of 9 links for you. Driving on. I’m @charlesarthur on Twitter. On Threads: charles_arthur. On Mastodon: https://newsie.social/@charlesarthur. On Bluesky: @charlesarthur.bsky.social. Observations and links welcome.
Nvidia chief Jensen Huang is seen as target of Chinese spy plot • The New York Times
Jonathan Swan, Maggie Haberman, Ana Swanson, Paul Mozur, Tripp Mickle and Julian E. Barnes:
»
American spy agencies had gathered striking intelligence.
A top executive at a Singapore-based data center company was under scrutiny for possibly steering advanced Nvidia AI chips to China. And now American officials had concluded that she was a Chinese intelligence asset, funneling information to Beijing.
But what came next was even more alarming. Surveillance of the executive, Alice Huang or Huang Le, a Chinese citizen, revealed that she had developed a close relationship with Jensen Huang, Nvidia’s chief executive.
For years, Washington has worked to deny China the technology needed to match or surpass the United States in the intense global competition for superiority in artificial intelligence. The discovery that Beijing was getting reports from someone with direct access to Mr. Huang, one of the most powerful figures in the AI revolution as well as an influential ally of President Trump’s, set off alarms at the White House.
Roughly a year ago, Mr. Trump was briefed about the intelligence findings, which have not been previously reported, and chose not to do anything with the information. Mr. Trump has been wary of disrupting Nvidia’s chip sales, which are helping to drive the economy and the stock market, and of further complicating the trading relationship with China.
This account is based on interviews with people briefed on the surveillance, who spoke on the condition of anonymity because the findings are drawn from classified intelligence, as well as other officials and industry executives and analysts.
American officials already suspected that Ms. Huang was helping China evade constraints on its ability to gain access to advanced AI chips.
«
Cast of thousands putting content into the story, and the details about “Alice Huang” – which might or might not be her real name – are fascinating. Absolutely classic honeypot operation, from the sound of it.
unique link to this extract
The neocloud shakeout is coming, and even the neoclouds seem to know it • Fortune
Beatrice Nolan:
»
Neoclouds are a new and increasingly popular type of business that has exploded alongside the AI boom. The business model is simple: Buy up in-demand AI chips, build or lease the data centers to house them, and rent out the computing power to AI companies. As the demand for data centers has grown, so has the number of neoclouds promising to provide them.
…I spent last week in Berlin at Nvidia’s GTC event where I met with several neocloud companies and, surprisingly, most agreed some kind of consolidation in the market is looming.
“I don’t think it will be like a light consolidation. It will be a massive wave,” Raimund Riedl, Volta’s chief financial officer, told me. “We’re going to start to see it in the relatively near future. There will start to be M&A activity, and we’ll consolidate.”
“We’re not worried about being one of those dead bodies, so to speak, because of the nature of this model,” he added.
Broadly, there are two types of neoclouds. The first is the established players, like CoreWeave, Nebius, and Iren, which have data centers and big customer contracts, often funded with big loans. Some, like Iren and CoreWeave, are former Bitcoin miners that already have at least some power and sites. The second is the newer startups, which often have little more than a business plan.
Volta is one of the new crowd. It came out of stealth in August and says it wants to lean less on venture capital and more on the huge pool of institutional money that funds roads, power grids, and cell towers, though it did raise $300m from investors including Andreessen Horowitz and Nvidia. It already has a $10bn contract, reportedly with Anthropic, that it has used to help secure a $5bn debt deal, partly underwritten by JPMorgan and Goldman Sachs.
“It’s just like every other new technology sort of hyped area. Everybody piles in,” Paul Miller, a principal analyst at Forrester, said of the neocloud market. “They will not all survive. There’s a lot of dead wood that needs to be cut out.” Miller added he expected to see the start of some consolidation over the next year, but then much more after that.
Gartner predicts that 80% of independent neoclouds will “cease independent operations” by 2030, while Bernstein analysts estimate that only 35% of the data center capacity that has been announced in North America will ever get built, according to a Bloomberg report.
«
Google admits AI search makes less money, but it’s fixing that with more ads • Ars Technica
Ryan Whitwam:
»
Google parent Alphabet reported its Q3 financial results on Wednesday afternoon, showing a healthy rise in revenue for the months of July to September. But the company’s stock price has taken a beating, marking one of the largest post-earnings drops in its history. It comes down to several troubling aspects of the new report, including an admission by the company that AI search isn’t earning as much as organic search, and despite that, Google expects to spend even more on AI next year.
The company took in almost $128bn in Q3 2026, earning more than $26bn in net income. These may sound like good numbers, but financial markets are not impressed. While revenue is up 25% versus last year, net income is almost $9bn lower. As of this story’s publication, the stock price is down almost 12%, and it was down about 13% earlier in the day.
…Google saw search grow by 12% versus last year. Good for the company, right? Except that’s a much slower rate than previous quarters. Growth in Q2 was 17 percent, and Q1 saw 19% growth over 2025. Search revenue was also flat in Q3.
So that’s the first problem, likely driven by more consumers using AI chatbots to find answers. The other issue is that Google’s take on AI search isn’t pulling in revenue like standard web search yet. According to Schindler, each AI Mode query earns less revenue compared to a similar web search—there are fewer ads, and they cost [advertisers] less to run. [Hence Google earns less money from them.]
«
GM to pay $220m to settle allegations it violated drivers’ privacy • Reuters
Jonathan Stempel:
»
General Motors will pay $220m to settle allegations that it illegally sold drivers’ private behavior and location data to brokers, who then sold the information to insurers that used it to set and sometimes increase rates.
The settlement, announced on Thursday, resolves claims by the attorneys general of 35 US states and the District of Columbia over the automaker’s Smart Driver program, which enrolled millions of drivers between 2016 and its discontinuation in 2024. Other regulators have also reviewed GM’s handling of driver data.
GM was accused of pitching Smart Driver to help people improve their personal driving safety, only to violate drivers’ privacy by reneging on its promise not to sell their data without permission. Many people who signed up were also enrolled in GM’s OnStar safety, navigation and diagnostic service.
“People bought a car, not a tracking device,” Oregon Attorney General Dan Rayfield said in a statement. Rayfield’s office and the office of Washington Attorney General Nick Brown led the bipartisan state coalition.
The settlement also requires GM to limit driving data it collects, and obtain drivers’ express consent before collecting or sharing that data.
A GM spokesperson said the settlement reinforces other steps the Detroit automaker has taken to strengthen privacy.
«
It’s just astonishing how companies can straight-out lie like this. GM had revenue of $167bn, operating profits of $11bn in 2025; after restructuring charges, still $3.1bn. So this is the tiniest of scratches. But equally, they couldn’t have made any appreciable money from selling the data. Just ruined the brand.
unique link to this extract
UK social media ban may lock out more than 100,000 teenagers without ID • The Guardian
Dan Milmo:
»
More than 100,000 British teenagers risk being locked out of social media if they do not have a passport or driving licence when an under-16 ban is introduced next year.
Ofcom, the UK’s online watchdog, said about 150,000 16- to 17-year-olds do not own a passport or driving licence, which are likely to be needed to access age-restricted platforms.
The Australia-style ban on under-16s accessing social media services, including Instagram, TikTok and YouTube, is due to come into force next spring.
The head of Ofcom, Melanie Dawes, said “motivated children” would seek to bypass the restrictions, as has happened in Australia.
On Thursday, Ofcom published an assessment of age verification measures that could underpin the ban, outlining four “highly effective” methods: checking a person’s official ID against a live selfie, a bank verifying an account holder’s age, an official digital ID service like the Young Scot card, or estimating someone’s age via a selfie.
However, the watchdog said a form of official ID would probably be needed as a “backstop” for other methods, such as facial age estimation, which could be inaccurate by several years. Ofcom said a cohort of 16- to 17-year-olds could miss out as a result.
“This means that people without a form of photo ID will likely need to take steps to obtain one to access age-restricted services,” the regulator said.
«
Does feel like we’re very slowly boiling the frog of IDs linked to accounts, whether anonymous or not.
unique link to this extract
Can LLMs supercharge geolocations? Our tests show big improvements but humans remain key • bellingcat
Foeke Postma and Miguel Ramalho:
»
We tasked not only closed-weight models from Claude, Grok, Gemini, ChatGPT, and Mistral, but also open-weight models from Moonshot AI’s Kimi, commercial provider GeoSeer and the non-generative geolocation model GeoClip. In total, we fed each model 52 photos that had not been published anywhere on the internet before.
The results showed that the current generation of models have markedly improved geolocation capabilities. Anthropic’s Opus and Fable, as well as Google’s recent Gemini 4 Argon model, performed the best on geolocation tasks, narrowing down locations from images faster than a human could under specific conditions. Preliminary tests of a set of agentic skills and tools being developed internally by Bellingcat also showed promising results in further improving the geolocation performance of generic-use AI agents.
But crucially, the tests also showed why humans remain necessary for accurate and trustworthy geolocation work, as no models were foolproof. This is particularly important given the increased use of AI in military contexts. The Anthropic blog post focused specifically on geolocating photos for military and intelligence purposes such as targeting, a historically labour-intensive process requiring significant human expertise.
While our tests showed some models performed extremely well at narrowing down results to the correct general area, none were close to being perfect at finding the exact spot a picture was taken. That accuracy is extremely important in geolocation, especially in military contexts where the cost of incorrect information can be life or death. It is also important for organisations like Bellingcat whose reputation relies on pinpoint factual accuracy.
«
OK, so they’re not perfect for bellingcat’s purposes, but they’ll probably serve for most of us who just want to figure out where something is. (For those who don’t know, or have forgotten it, the bellingcat origin story is extraordinary and worth a read: probably the most -only? – useful thing to come out of the Guardian’s comments.)
unique link to this extract
Where did the M25 come from? • Roads.org.uk Blog
Chris Marshall on the circumnavigating motorway around London which is now 40 years old:
»
How should we see the M25, 40 years on? This morning I was invited on to BBC Radio Essex to talk about the anniversary (hello Essex!) and one of the things I was asked was whether the M25 could be described as a success. I think if you asked most people to pick a word to describe the M25, “success” would not be the one they’d choose. It is slow, congested and frustrating. But does that make it a failure?
One way of looking at it is to ask what it was meant to do. It was meant to bypass London. Before it was built – and this can be hard to even imagine today – if you wanted to get from one side of London to the other you almost certainly had to go through it. You could spend hours churning through the suburbs on the North or South Circular, or you might even end up just going right through the centre. But for 40 years and counting, there’s no way you’d do that now. No matter how long the queues at Dartford or how slow it might be around Heathrow, you’d be extremely unlikely to attempt a trip through London.
If anything, the fact that it is so slow, so congested and so frustrating is a product of how huge is the demand for the M25. It has enabled journeys that weren’t viable before, opened up opportunities and possibilities. The fact that it can be such a nightmare is, in many ways, because it’s a victim of its own success.
It’s often said – and I said it at the start of this post – that Margaret Thatcher’s speech, when she opened the M25, included an announcement of widening works. It’s often stated as an example of just how early things started to go wrong, of how intrinsic its failings are. Well, maybe. But that’s not the point that Thatcher was making.
«
Thatcher insisted that the M25’s popularity indicated that it had succeeded, not failed. The problem with motorways is always all the other drivers. The roads themselves are great. Certainly given the choice of going around the M25 or driving through London, I’d take the former.
unique link to this extract
New prize launches for books co-written with AI • The Guardian
Ella Creamer:
»
A prize for AI-written books has launched this week.
The Prix Pascal de Littérature IA invites submissions of manuscripts co-written with AI tools, with the winner promised publication.
Entrants are asked to submit an unpublished manuscript of between 10,000 and 50,000 words, along with information about which AI model and prompt were used to generate the text. All major language models are allowed, including ChatGPT, Claude, Gemini and LLaMA. Entrants can be based anywhere, but entries must be submitted in French.
The Prix Pascal is named after Blaise Pascal, a French polymath who helped invent one of the first mechanical calculators and also wrote theological texts, including Pensées, which Pope Francis praised for its “philosophical depth and literary charm”.
While the terms and conditions ask for a work of fiction, the prize also states that submitted texts should help solve an ethical, social or ecological problem threatening humanity. The prize was founded by Philippe Rodriguez, Albin Serviant and Gilles Mentré, all of whom are technology entrepreneurs.
«
You can see that the problem for every book prize now is “was AI used to write this at all?”, which is a bad place to be. Perhaps the quality (of lack thereof) in these submissions will tell us how easy it is to distinguish.
unique link to this extract
The beleaguered chip sellers getting their revenge on Apple • WSJ
Rolfe Winkler:
»
During a ride in a Southern California desert, sales executive Scott Nelson stopped his dirt bike, pulled out a rifle and shot his old iPod.
Nelson sold flash storage to Apple and hated feeling like he always got the worst end of the deal. His Japan-based manufacturer, Kioxia, had little leverage. Apple was the world’s biggest buyer of flash—the chips in iPhones that store photos, music and apps—and used its position to squeeze every cent from suppliers. Apple bullied flash makers into opening their books so it could monitor sales, making sure no one else got a better deal.
Kioxia and other flash makers stewed as Apple paid them pennies for flash storage that the iPhone maker resold to customers for dollars. The disparity in bargaining power had galled Nelson, who used Android phones in protest.
Nelson kept the broken pieces of his iPod, the first Apple device to use flash. They sit on his desk and hold a special reverence now that the tables have turned.
In the past year, demand by AI hyperscalers has quintupled prices for flash, according to research firm TechInsights, opening a new era for Kioxia and other suppliers. Two years ago, Apple bought nearly twice the flash storage of Microsoft, Amazon, Meta and Google combined, according to TechInsights estimates. Hyperscalers are now buying double Apple’s share.
Flash memory suppliers and their sales teams privately expressed their glee at Apple’s comeuppance during chitchat at the industry’s largest conference this summer in Santa Clara, Calif. AI deals have given Kioxia sales executives and other flash suppliers new leverage to charge Apple higher prices.
«
Gift link to a story that tells fabulous stories about how Apple used to be able to squeeze memory makers’ prices down to its advantage when it was the bigger buyer. Now it’s not the bigger buyer; the hyperscalers are.
unique link to this extract
| • Why do social networks drive us a little mad? • Why does angry content seem to dominate what we see? • How much of a role do algorithms play in affecting what we see and do online? • What can we do about it? • Did Facebook have any inkling of what was coming in Myanmar in 2016? Read Social Warming, my latest book, and find answers – and more. |
Errata, corrigenda and ai no corrida: none notified
