TheChek News Flow

Jensen Huang unveils AI models for humanoids and self-driving car partnership with Toyota at CES keynote

Nvidia is on the cusp of revolutionising robotics through artificial intelligence, chief executive Jensen Huang said on Monday, as he outlined his vision for the next stage of the company’s staggering growth and predicted a “multitrillion-dollar” opportunity.

Huang announced a range of new products and partnerships in the “physical AI” space, including AI models for humanoid robots and a major deal with Toyota to use Nvidia’s self-driving car technology, during his keynote speech at the annual Consumer Electronics Show in Las Vegas.

Nvidia has flown past a $3tn market capitalisation off the back of demand for its AI chips to become one of the world’s most valuable companies. Huang in turn has become a household name, more than 30 years after he founded Nvidia as a video game graphics chip company.

Massive queues had formed outside the Mandalay Bay convention centre long before the keynote started, with some people still lining up when Huang emerged onstage in a sparkly version of his trademark leather jacket, quipping: “I’m in Las Vegas after all.”

Outside semiconductors, Nvidia has been building the software that allows companies to train and deploy robots, from those used in smart factories and warehouses to self-driving cars and humanoids, pushing to expand the use cases for AI running on its chips.

Cracking the technological challenges involved in deploying robots at scale will pave the way to “the largest technology industry the world has ever seen”, said Huang.

Nvidia said the field of robotics had reached a technological tipping point, as AI accelerates and fine-tunes the process of simulating the physical world and generating the vast amounts of data needed to train robots. In the next two decades, the market for humanoid robots alone is expected to reach $38bn, according to the company.

On Monday, Nvidia announced a suite of foundational AI models on its new Cosmos platform, which developers can use for free to generate data and build their own models.

Nvidia said the foundation models, which it said were trained on 20mn hours of video data, were as fundamental a technological development as the large language models that underpin apps such as OpenAI’s ChatGPT. It pairs with Nvidia’s Omniverse platform, which is used to run simulations of the physical world.

“What [those models] are doing for language, we can now do for understanding the physical world,” Rev Lebaredian, Nvidia’s vice-president for Omniverse and simulation technology, told the Financial Times. While data on the physical world is much harder to gather and process than text, Lebaredian said “it’s a necessary part” of the company’s mission.

“The big takeaway [from Huang’s CES speech] is that this moment is going to be a special one,” he added. “I think this year is an inflection point where we’re going to see this acceleration of physical AI and robotics.”

The Omniverse platform and robotics currently represent a small share of the company’s overall revenue. For Nvidia’s quarter to the end of October, “professional virtualisation” accounted for $486mn of revenue, while automotive and robotics totalled $449mn.

This is a sliver of overall sales, as the company raked in $30.8bn in revenue from selling chips for the data centres that power AI models in the same period.

Nvidia’s search for new markets comes as it faces growing pressure from its biggest customers, including Amazon and Microsoft, which are rushing to build their own in-house AI data centre chips.

Bank of America analysts said Nvidia’s decision to double down on “physical AI” was the “next logical step”. The challenge would be in “making the products reliable enough, cheap enough and pervasive enough to spawn credible business models”, they added.

At CES, Nvidia also unveiled a collection of foundation models for humanoid robots, called the “GR00T Blueprint”, which it said would “supercharge” the development of robots, as well as new tools for developing and testing fleets of factory and warehousing robots and training autonomous vehicles.

Toyota announced it would build its next generation of autonomous vehicles on Nvidia’s hardware and software, known as Drive AGX. Self-driving car group Aurora and automotive parts maker Continental will use Nvidia’s hardware and software to power thousands of driverless trucks under their long-term strategic partnerships with the chipmaker.

Nvidia said it expected its automotive business to grow to $6bn in the 2026 fiscal year. Autonomous vehicles “will be the first multitrillion-dollar robotics industry”, Huang told the CES audience.

Separately, Nvidia said it would release a “personal AI supercomputer” with its latest and most powerful AI chip, Blackwell, allowing researchers and students to run multibillion-parameter AI models locally rather than through the cloud. It will be available in May at an initial price tag of $3,000.

Google is forming a new team to work on AI models that can simulate the physical world.

Tim Brooks — one of the co-leads on OpenAI’s video generator, Sora, who left for Google’s AI research lab, Google DeepMind, in October — will lead the new team, he announced in a post on X. It’ll be a part of Google DeepMind.

“DeepMind has ambitious plans to make massive generative models that simulate the world,” Brooks wrote Monday morning. “I’m hiring for a new team with this mission.”

According to job listings Brooks linked to in his post, the new modeling team will collaborate with and build on work from Google’s Gemini, Veo, and Genie teams to tackle “critical new problems” and scale models “to the highest levels of compute.” Gemini is Google’s flagship series of AI models for tasks like analyzing images and generating text, while Veo is Google’s own video generation model.

As for Genie, it’s Google’s take on a world model — AI that can simulate games and 3D environments in real time. Google’s latest Genie model, previewed in December, can generate a massive variety of playable 3D worlds.

“We believe scaling [AI training] on video and multimodal data is on the critical path to artificial general intelligence,” reads one of the job descriptions. Artificial general intelligence, or AGI, generally refers to AI that can accomplish any task a human can. “World models will power numerous domains, such as visual reasoning and simulation, planning for embodied agents, and real-time interactive entertainment.”

Per the description, Brooks’ new team will look to develop “real-time interactive generation” tools on top of the models they build, and study how to integrate their models with existing multimodal models such as Gemini.

A number of startups and big tech companies are chasing after world models, including influential AI researcher Fei-Fei Lee’s World Labs, Israeli upstart Decart, and Odyssey. They believe that world models could one day be used to create interactive media, like video games and movies, and run realistic simulations like training environments for robots.

But creatives have mixed feelings about the tech.

A recent Wired investigation found that game studios like Activision Blizzard, which has laid off scores of workers, are using AI to cut corners, ramp up productivity, and compensate for attrition. And a 2024 study commissioned by the Animation Guild, a union representing Hollywood animators and cartoonists, estimated that over 100,000 U.S.-based film, television, and animation jobs will be disrupted by AI by 2026.

Some startups in the nascent world modeling space, like Odyssey, have pledged to collaborate with creative professionals — not replace them. We’ll have to see if Google follows suit.

There’s also the unresolved matter of copyright. Some world models appear to be trained on clips of video game playthroughs, which could make the companies developing those models the target of lawsuits in cases where the videos were unlicensed.

Google, which owns YouTube, asserts that it has permission to train its models on YouTube videos in accordance with the platform’s terms of service. But the company hasn’t said which specific videos it is sourcing for training.

Of the cybersecurity risks facing the United States today, few loom larger than the potential sabotage capabilities posed by China-backed hackers, which senior U.S. national security officials have described as an “epoch-defining threat.”

The U.S. says Chinese government-backed hackers have — in some cases for years — been burrowing deep into the networks of U.S. critical infrastructure, including water, energy, and transportation providers. The goal, officials say, is to lay the groundwork for potentially destructive cyberattacks in the event of a future conflict between China and the United States, such as over a possible Chinese invasion of Taiwan.

“China’s hackers are positioning on American infrastructure in preparation to wreak havoc and cause real-world harm to American citizens and communities, if or when China decides the time has come to strike,” then-outgoing FBI Director Christopher Wray told lawmakers last year.

The U.S. government and its allies have since taken action against some of the “Typhoon” family of Chinese hacking groups, and published new details about the threats posed by these groups.

In January 2024, the U.S. disrupted “Volt Typhoon,” a group of Chinese government hackers tasked with setting the stage for destructive cyberattacks. Later in September 2024, federal authorities took control of a botnet run by another Chinese hacking group called “Flax Typhoon,” which used a Beijing-based cybersecurity company to help conceal the activities of China’s government hackers. Then in December 2025, the U.S. government sanctioned the cybersecurity company for its alleged role in “multiple computer intrusion incidents against U.S. victims.”

Since the emergence of Volt Typhoon, another new China-backed hacking group called “Salt Typhoon” appeared in the networks of U.S. phone and internet giants, capable of gathering intelligence on Americans — and potential targets of U.S. surveillance — by compromising telecom systems used for law enforcement wiretaps.

Here’s what we have learned about the Chinese hacking groups gearing up for war.

Volt Typhoon

Volt Typhoon represents a new breed of China-backed hacking groups; no longer just aimed at stealing sensitive U.S. secrets, but rather preparing to disrupt the U.S. military’s “ability to mobilize,” according to the then-FBI director.

Microsoft first identified Volt Typhoon in May 2023, finding that the hackers had targeted and compromised network equipment, such as routers, firewalls, and VPNs, since at least mid-2021 as part of an ongoing and concerted effort to infiltrate deep into the systems of U.S. critical infrastructure. The U.S. intelligence community said that in reality, it’s likely the hackers were operating for much longer, potentially for as long as five years.

Volt Typhoon compromised thousands of these internet-connected devices in the months following Microsoft’s report, exploiting vulnerabilities in devices that were considered “end-of-life” and therefore would no longer receive security updates. The hacking group subsequently gained further access to the IT environments of multiple critical infrastructure sectors, including aviation, water, energy, and transportation, pre-positioning for activating future disruptive cyberattacks aimed at slowing the U.S. government’s response to an invasion of its key ally, Taiwan.

“This actor is not doing the quiet intelligence collection and theft of secrets that has been the norm in the U.S. They are probing sensitive critical infrastructure so they can disrupt major services if, and when, the order comes down,” said John Hultquist, chief analyst at security firm Mandiant.

The U.S. government said in January 2024 that it had successfully disrupted a botnet, used by Volt Typhoon, consisting of thousands of hijacked U.S.-based small office and home network routers, which the Chinese hacking group used to hide its malicious activity aimed at targeting U.S. critical infrastructure. The FBI said it was able to remove the malware from hijacked routers by way of a court-sanctioned operation, severing the Chinese hacking group’s connection to the botnet.

By January 2025, the U.S. had discovered more than 100 intrusions across the country and its territories linked to Volt Typhoon, according to reporting by Bloomberg. A large number of these attacks have targeted Guam, a U.S. island territory in the Pacific and a strategic location for American military operations, the report said. Volt Typhoon allegedly targeted critical infrastructure on the island, including its main power authority, the island’s largest cell provider, and several U.S. federal networks, including sensitive defense systems, based on Guam. Bloomberg reported that Volt Typhoon used an entirely new kind of malware to target networks in Guam that it hadn’t ever deployed before, which researchers took as a sign of the high importance that the region has to the China-backed hackers.

Flax Typhoon

Flax Typhoon, first outed by Microsoft several months later in an August 2023 report, is another China-backed hacking group, which officials say has operated under the guise of a publicly traded cybersecurity company based in Beijing to carry out hacks against critical infrastructure in recent years. Microsoft said Flax Typhoon — also active since mid-2021 — predominantly targeted dozens of “government agencies and education, critical manufacturing, and information technology organizations in Taiwan.”

Then in September 2023, the U.S. government said it had taken control of another botnet, which was made up of hundreds of thousands of hijacked internet-connected devices, and used by Flax Typhoon to “conduct malicious cyber activity disguised as routine internet traffic from the infected consumer devices.” Prosecutors said the botnet allowed other China government-backed hackers to “hack into networks in the U.S. and around the world to steal information and hold our infrastructure at risk.”

The Department of Justice later corroborated Microsoft’s findings, adding that Flax Typhoon also “attacked multiple U.S. and foreign corporations.”

U.S. officials said that the botnet used by Flax Typhoon was operated and controlled by the Beijing-based cybersecurity company, Integrity Technology Group. In January 2024, the U.S. government imposed sanctions on Integrity Tech over its alleged links to Flax Typhoon.

Salt Typhoon

The latest — and potentially most ominous — group in China’s government-backed cyber army uncovered in recent months is Salt Typhoon.

Salt Typhoon hit headlines in October 2024 for a different kind of information-gathering operation. As first reported by The Wall Street Journal, the China-linked hacking group compromised several U.S. telecom and internet providers, including AT&T, Lumen (formerly CenturyLink), and Verizon. The Journal reported later in January 2025 that Salt Typhoon also breached the U.S.-based internet providers Charter Communications and Windstream. U.S. cyber official Anne Neuberger said the federal government had identified an unnamed ninth hacked telco.

According to one report, Salt Typhoon may have gained access to these telcos using compromised Cisco routers. Once inside the telco’s networks, the attackers were able to access customer call and text message metadata, including date and time stamps of customer communications, source and destination IP addresses, and phone numbers from over a million users; most of which were individuals located in the Washington D.C. area. In some cases the hackers were capable of capturing phone audio from senior Americans. Neuberger said that a “large number” of those who had data accessed were “government targets of interest.”

By hacking into systems that law enforcement agencies use for court-authorized collection of customer data, Salt Typhoon also potentially gained access to data and systems that house much of the U.S. government’s data requests, including the potential identities of Chinese targets of U.S. surveillance.

It’s not yet known when the breach of the wiretap systems occurred, but may date back to early 2024, according to the Journal’s reporting.

AT&T and Verizon told TechCrunch in December 2024 that their networks were secure after being targeted by the Salt Typhoon espionage group. Lumen confirmed soon after that its network was free from the hackers.

Regulator rules fees should come down by 3.8% compared with HS1’s proposal

HS1 Ltd, the owner of Britain’s only high-speed railway, has been told to lower the charges that train operators including Eurostar pay to use the line linking London to the Channel Tunnel.

The Office of Rail and Road, the industry regulator, on Monday ruled that charges should come down by 3.8 per cent, or £5mn a year, compared with HS1’s proposal.

Feras Alshaker, director of planning and performance at the ORR, said the changes would “result in significantly lower costs” for train operators, “which should benefit everyone who uses this railway”.

As well as Eurostar, the line and its central London station at St Pancras is used by UK domestic operators Southeastern and East Midlands Railway.

The changes could result in lower ticket prices for passengers if they are passed on by operators. The ORR said Eurostar and Southeastern, by far the largest groups on the high-speed line, had pushed for even lower charges to be paid to HS1.

Eurostar and Southeastern have also said that high track charges are one factor that stopped them running more trains on the line, which only operates at about half of its potential capacity.

HS1 Ltd said: “We are pleased to see the ORR’s positive endorsement for our plan for the next five years and we now look forward to seeing how the lower cost to operators drives growth on HS1.”

Along with the Channel Tunnel owner Getlink, HS1 has been encouraging more train operators to launch international services linking London and continental Europe.

Several companies including Sir Richard Branson’s Virgin Group and a consortium backed by the largest shareholders in London-listed Mobico, formerly known as National Express, are exploring starting new services linking the UK and mainland Europe, while the incumbent Eurostar also has significant growth plans.

HS1 chief executive Robert Sinclair last month told the Financial Times that the company planned to more than double the international passenger capacity of St Pancras station to help encourage major growth.

The rail infrastructure is ultimately owned by the UK government. But HS1 Ltd, which is owned by fund managers InfraRed Capital Partners and Equitix Investment Management, has a 30-year concession to own and operate the HS1 rail line and stations along its route.

The ORR ruled that HS1 would be able to charge operators a total of £126.7mn a year between April 2025 and March 2030. The company had asked to charge £131.6mn.

The regulator said HS1 had been directed to reduce its charges for renewing its track assets and its stations, including St Pancras. It must also reduce its charges for its day-to-day operating and maintenance of the railway.

“Although, overall, HS1’s original plans were good, the company must now change specific areas of those plans to account for our decisions,” Alshaker said.

Eurostar said: “We welcome the ORR’s decision to instruct HS1 to reduce track access charges on the Eurostar route.”

“In recent years, these charges have risen significantly above the rate of inflation, which has put pressure on our ability to invest in the stations and infrastructure needed to support our future growth plans,” it added.

Southeastern said: “Southeastern is publicly owned and receives a significant taxpayer subsidy to operate services. We will examine how the decision to reduce the cost to operate on HS1 could benefit customers, including the potential for additional services.”

Last week, Apple agreed to pay $95 million to settle a lawsuit that alleged “unlawful and intentional recording” of Siri interactions.

Apple’s settlement has led to a wave of conspiracy theories claiming that Siri is listening to you for targeted advertising, but the company says those claims are completely unfounded.

Siri’s privacy controversy

As a refresher, the lawsuit stemmed from a 2019 report in the Guardian that revealed Apple’s use of contractors to “grade” Siri interactions. The whistleblower in the story alleged that those contractors would regularly hear private interactions from users as part of their work providing “quality control” for Siri.

At the time, Apple quickly responded to the allegations by saying that “less than 1% of daily Siri activations” were used for “grading” and that those activations were typically only a few seconds long. The interactions were also bound by “Apple’s strict confidentiality agreement” and were not associated with a user’s Apple ID.

Apple also subsequently announced several changes to Siri’s privacy protections in a post on Apple Newsroom. The big change was that, by default, Apple no longer retained recordings of Siri interactions. Instead, users could opt-in to help Siri improve by “learning from the audio samples of their requests.” Apple also said that “only Apple employees” would be allowed to listen to audio samples of Siri interactions, not third-party contractors – and any recording that was “determined to be an inadvertent trigger of Siri” would be swiftly deleted.

Is your iPhone listening to you to show you ads? Nope.

Fast forward to 2025, and Apple agreed last week to settle that 2019 lawsuit with a $95 million payout to users. In a statement to 9to5Mac today, Apple said it settled the case so it can “move forward from concerns about third-party grading that we already addressed in 2019.”

The company says that Siri data has “never been used to build marketing profiles, and it has never been sold to anyone for any purpose.”

Here is the full statement from an Apple spokesperson:

Last week’s news spawned a number of unfounded conspiracy theories using Apple’s settlement as evidence that your iPhone is “always listening to you” and spying on you for means of targeted advertising. Apple tells me this is absolutely not the case, and what you share with Siri is never shared with advertisers.

Apple says it repeatedly denied allegations throughout the lawsuit that Siri recordings were used to target advertisements, and no evidence was presented to suggest otherwise.

In fact, Siri interactions are tied to a random identifier that lets Apple keep track of data during processing. Those interactions are not tied to your Apple Account, phone number, or any other identifying information. After six months, that request history is also unlinked from that random identifier. All of these details (and more) are emphasized on Apple’s website on a webpage dedicated to Siri and Dictation privacy.

Additionally, you can manually review and delete Siri transcripts directly in Settings. Just go to the Settings app and look for the “Siri & Dictation History” option.

Some Siri requests are also handled entirely on-device. For example, if you ask Siri to read unread messages, it does so by simply instructing your iPhone to read your messages aloud. The content of the message is not sent to Apple servers.

For Apple Intelligence features, Apple also emphasizes its use of Private Cloud Compute. Apple’s Private Cloud Compute infrastructure is built on its own Apple Silicon chips and is open to third-party researchers to ensure privacy protections.

9to5Mac’s Take

All this to say, the headlines making the rounds suggesting that this lawsuit is evidence your phone is “always listening to you” are nothing but unfounded conspiracy theories. In fact, it is quite literally impossible that your interactions with Siri are being used for targeted advertising with the privacy protections that Apple has put into place.

This, of course, doesn’t excuse Apple’s reactive rather than proactive approach to the situation that first arose in 2019. Apple should’ve had more privacy protections in place before then, and it shouldn’t have taken a whistleblower for it to respond. The system should’ve been opt-in from the start. Nonetheless, Apple has continued to double down on Siri’s privacy protections since then.

A path forward for Europe to compete on the internet

Its mixed economy could help create a new sort of success story

Mario Draghi’s assessment of EU competitiveness last year highlighted Europe’s failure to capitalise on the internet revolution. Over the past two decades, Silicon Valley’s runaway tech sector has accounted for most of the difference between US and European productivity growth. 

But if US tech dominance was helped by its freer market, Europe’s mixed economy could help it create a new sort of internet success story. This is because the internet, and the services that many private companies provide on it, is a public good.

Economics teaches that public goods are a problem. Public amenities like paved roads are expensive to build but freely available to all. They cannot be adequately funded privately so governments must fund them instead. 

On the internet, the economics of public goods are transformed. By bringing billions of people within a mouse-click of any online service, the internet so lowered the cost of digital services that what I call “public goods of opportunity” now abound. These are digital services that are freely available but privately provided and funded via advertising (Google Search and Facebook), or even donations (Wikipedia).

European governments could advance this model by establishing digital public-private partnerships. For example, a government could “bulk purchase” standard online accounting services so its citizens and small businesses could access them for free. One might do the same with other standard online services, for example in payroll or employee engagement. As well as lowering business costs, the market dominance held by the free provider would give it a rich database which could be mined for value.

Take health. In contrast to the dysfunctions of the US private health insurance market, Europe’s strong public involvement in health makes it well placed to develop such a public-private model in digital health partnerships.

The financially imperilled US genetic-testing company 23andMe hosts data about your genome to help you understand your health susceptibilities and ancestry. If its tests were free it would go broke, so they’re provided for a fee — as a private, not a public good.

But like Google or Wikipedia, a similar model could generate more value as a free public good. Europe’s public health systems could fund testing and become custodians of the resulting data. The resulting online system could be a new public good: a data utility.

Accordingly, more people would participate. With the data secure (and anonymised where appropriate) a giant new genomic database could then inform diagnosis, management and healthcare research. (23andMe would become just one of many services competing on the platform). This in turn would spur advances in personalised medicine, improving lives and aiding industry development.

And what of bad behaviour on freely provided social media? While the US debates regulation necessary to minimise its worst harms, there’s another agenda that Europe could explore. If governments step in with a role in privately provided public goods online, there is a chance that they could take a more active role in tackling the profit-maximising algorithms on social media.

We know algorithms on social media dumb down discussion and ramp up outrage to maximise profits. But can we build social media around algorithms that improve discussion? Europe’s mixed economy could support this via state institutions with broad social and cultural missions — for instance universities and publicly funded broadcasters.

Public goods of opportunity could help deliver the productivity growth that Draghi tells us is so urgent. By reinventing traditional government-funded public goods in a digital world, Europe can create the next generation of internet success stories.