FT : Europe doesn’t need driverless cars

Europe doesn’t need driverless cars
The roadways are already safe and robotaxis risk increasing traffic

David Zipper - The writer is a senior fellow at the MIT Mobility Initiative

Get ready, Europe. The robotaxis are coming. This year, Londoners could see driverless cars from China’s Baidu and Waymo, the self-driving subsidiary of Google’s Alphabet. Tensor Auto and GreenMobility are planning to launch autonomous cars in Copenhagen while Stellantis and Pony.ai want to do the same in Luxembourg. European Commission President Ursula von der Leyen has voiced her support: “Self-driving cars are already a reality in the United States and China. The same should be true here in Europe.”

She is right about the first part. Autonomous rides are available in Chinese cities like Beijing and Wuhan and in the US, where I live, driverless cars from Waymo, Zoox and Tesla navigate the streets of cities like San Francisco, Los Angeles, Phoenix and Austin. Waymo is reported to provide 450,000 paid trips per week.

Driverless journeys can be pleasantly private, albeit too expensive for most people to take them regularly (Waymo rides often cost around a third more than Uber taxis). But to win over regulators and the general public, American robotaxi companies have also presented themselves as an urgently needed solution to the country’s abysmal record of crash deaths. 

Such messaging is of dubious relevance to Europe, however, where roadways are already much safer than those in the US. Furthermore, Europeans are more likely to bike and ride public transportation — travel modes that are vulnerable to a deluge of self-driving cars.

So why, exactly, does Europe want robotaxis?

Driverless cars are often perceived as an embodiment of innovation and their developers have long claimed that computers can operate vehicles more safely than humans. They will not drive inebriated, exhausted or distracted.

Unaffiliated experts, such as Phil Koopman at Carnegie Mellon, say it is still too soon to know whether robot drivers are really safer. But a technological fix for crashes would be a godsend in the US, where the death count rose by over 25 per cent between 2010 and 2023. An American is more than twice as likely to die in a crash as a Canadian or Korean and more than three times as likely as an Israeli or Spaniard.

Europe is in a different situation. Smaller cars, higher transit use and more pedestrianised areas make roads safer. In 2023, the European Union averaged roughly 4.5 crash deaths per 100,000 residents, less than half the American equivalent. Roads in non-EU member states like Britain and Norway were even safer. A handful of Nordic cities like Helsinki have almost eliminated fatal crashes by slowing cars down.

This means the upside of robotaxis in Europe is lower. And there are considerable downsides to introducing autonomous cars into transportation networks that do not revolve around motor vehicles. Robotaxis have caused relatively few problems in car-centric American cities like Phoenix. But in San Francisco they have blocked bike lanes and obstructed transit vehicles. Such mishaps should be a warning for European leaders not to let their cities become beholden to the automobile.

Technological marvels they may be, but robotaxis are still cars and cars are a uniquely inefficient means of moving large numbers of people when space is at a premium. By inviting robotaxis into their narrow, busy streets, European cities risk worsening congestion. In California, data shows that Waymos spend almost half their time “deadheading” — driving around empty — which increases traffic. Gridlocked cities would undermine the quality of life and economic productivity that European leaders like von der Leyen want.

Robotaxi deployment has been compared to a race — one European carmakers have yet to start. If so, this is a race Europeans should skip altogether.

FT : Pubs criticise prospect of licensing reforms in face of higher business rat

Pubs criticise prospect of licensing reforms in face of higher business rates
Sector calls on ministers to provide urgent financial support following Budget measures

Pub bosses have criticised Sir Keir Starmer’s suggestion that higher business rates bills can be softened by licensing reforms, as they called on the prime minister to provide urgent financial support.

Industry leaders met Treasury minister Dan Tomlinson to discuss the burden of business rates on the hospitality sector, with a Budget increase in the tax paid on commercial properties adding to pressure on pubs.

The increases in pubs’ costs stem largely from higher “rateable values” — estimates of a premises’ annual rent used to calculate rates — which have risen considerably since the last pandemic-era valuation.

In some cases, rateable values have doubled or trebled, although a typical increase is about 30 per cent. Pubs are also facing the withdrawal of Covid support, which offered 40 per cent relief on rate bills.

The prime minister on Monday said pubs would “struggle” as a result of the higher rateable values, telling LBC that “because of revaluation, that means that some will have their bills going up”.

The government was putting transitional relief in place — though pubs say bills will still be significantly higher in three years’ time — and looking at “licensing freedoms”, he added.

Downing Street said on Tuesday that it was exploring “speeding up licensing reforms, slashing red tape, helping more venues stay open later, offering pavement drinks and putting on one-off events”.

But Jonathan Neame, chief executive of pub group Shepherd Neame, dismissed licensing reforms as “a shallow political smokescreen, which will make little or no difference to the hospitality sector”.

David Wigham, managing director of Admiral Taverns, said: “Licensing reforms does nothing to tackle the rising costs that the government is causing.”

“The government needs to offer financial support to the industry and the business rates reduction that we were led to believe was coming,” he added.

Anger within the industry over the changes has led to Labour MPs being banned from more than 1,000 pubs across the UK following a grassroots campaign, with landlords arguing the November Budget has triggered an existential threat.

Alex Reilley, executive chair of Loungers, which has 250 cafés and bars across the UK, said he did not “know of a single hospitality business that is calling for more relaxed licensing laws to allow for extended opening hours. On the contrary, in an attempt to make their business viable, [a] large number of hospitality businesses are reducing their hours.”

Pubs could not afford the extra costs of operating for longer, Reilley added, noting that customers were “eating and drinking earlier than they were pre-Covid”.

Andy Spencer, chief executive of Punch Pubs, which has about 1,250 venues in the UK, said that while it was “encouraging to hear the government acknowledge the pressure pubs are under, warm words alone won’t be enough”.

One hospitality sector boss said: “If a pub isn’t viable because of government cost pressures, it’s irrelevant what time they can open and close or have alfresco seating.”

After the meeting with Tomlinson, Emma McClarkin, chief executive of the British Beer and Pub Association, said the government “understands the strength of feeling amongst the sector” and was “in listening mode”.

“We will be meeting again in the coming weeks,” she added.

FT : KLM warns of de-icer shortage after Schiphol snowstorms

KLM warns of de-icer shortage after Schiphol snowstorms
Stranded passengers may have to wait days for alternative flights after hundreds of cancellations

KLM has warned it is at risk of running out of de-icing fluid after Amsterdam’s Schiphol airport was hit by days of severe snowstorms that forced the Dutch airline to cancel hundreds of flights. 

The airline, which is owned by French-Dutch group Air France-KLM, said on Tuesday that delays from its de-icer supplier combined with “extreme weather conditions” mean that its “stock levels are running low”.

It has been receiving daily deliveries, but now its supplier in Germany “is currently unable to guarantee timely replenishment”. 

The airline has dispatched a team to Germany to collect the fluid itself, and “is doing everything possible to secure additional supply”. 

The impending de-icer shortage adds to the problems for the airline, which has also warned passengers stranded at the airport they may have to wait for days for alternative flights — as it cancelled 600 more flights that had been due to fly on Wednesday. 

“Normally, when a flight is cancelled, passengers are offered an alternative flight within a few hours,” KLM said. “However, because the weather conditions have not improved and we can only operate a limited number of flights, some passengers are being offered alternative flights scheduled later in the week.” 

It warned of “persistent snowfall expected [Wednesday] morning, combined with strong winds from the south to south-east”. 

The airline has cancelled more than 1,000 flights since Friday, and had many hundreds more delayed, leaving thousands of passengers unable to fly from or into the airport, which is the busiest in Europe and a major connection for long-haul travellers. 

European air traffic control service Eurocontrol warned that it expects “no improvements” at the Dutch airport on Wednesday, and has asked airlines to cut 70 per cent of their expected flights. 

“The weather causes a significant outbound bottleneck, so inbound capacity is reduced to balance demand,” it said on Tuesday evening. Its weather forecast is for continued snow over Scandinavia, the UK and the Netherlands on Wednesday. 

Europe has been hit by the cold snap, with snowfall across the region. The UK has warned that Storm Goretti on Thursday will bring further snow and cause further travel disruption. 

KLM, which is responsible for de-icing most aircraft at Schiphol, has a dedicated team of 100 staff who manage the task. They have been using around 85,000 litres a day of fluid on its planes, while its fleet of 25 de-icing trucks have been in “continuous use since Friday”, the company said on Tuesday. 

De-icing the runways and other equipment at Schiphol is handled by Menzies Aviation, which said it had sufficient supplies at the airport and across its European and UK network. “We have de-icing fluid in stock, with our next delivery scheduled for tomorrow [Wednesday],” it added.

Schiphol airport said it “has sufficient winter‑operations materials to keep runways, taxiways and aircraft stands clear. Aircraft de‑icing, however, is the responsibility of the airlines, and their fluid stocks and processes are entirely separate from the materials used by the airport for clearing the runways.”

It added its “snow crews are working around the clock to keep the runways clear, and aircraft are being carefully de-iced to ensure everyone can travel safely”, but that “the wintry conditions are expected to cause disruptions to the flight schedule in the coming days. This may result in delays and cancellations.”

Other European airlines including Ryanair, easyJet and Lufthansa reported they did not have problems with de-icing supplies. 

The Information : AI Evaluation Startup LMArena Valued at $1.7 Billion in New Fu

AI Evaluation Startup LMArena Valued at $1.7 Billion in New Funding Round

The Takeaway
  • LMArena secured $150 million in new funding, reaching a $1.7 billion valuation.
  • Startup provides widely cited AI model rankings for major tech companies.
  • The money will help it run AI models and expand its staff.

LMArena, a startup that operates a widely cited ranking of AI models based on their performance, has raised $150 million at a valuation of $1.7 billion, including the new money, according to the company. That’s nearly triple the valuation of its seed funding round, announced in May 2025.

The funding round, co-led by existing investors Felicis and the University of California’s investment arm, will help the company pay for computing power to run AI models it evaluates for customers such as OpenAI, Google, xAI and Microsoft, as well as hire technical staff. Millions of people visit LMArena’s site to rate the models based on head-to-head comparisons, and those ratings inform the rankings.

LMArena was generating revenue at an annualized pace of several million dollars in September. Its recent revenue pace couldn’t be learned. It estimates its “annualized consumption run rate,” a projection of how much revenue it would earn if recent customer usage continued over a year, hit $30 million last month.

The company, started as an open-source project nearly three years ago, says it has more than 5 million monthly users in 150 countries. It’s not clear if those include visitors to its website as well as the people who rate models.

LMArena’s website invites anyone on the internet to ask questions or use models for creations such as images. Those people then pick the best answer between two options, before the site reveals which model generated each output. LMArena tallies the results into a leaderboard for various categories, such as for AI coding, image and video generation.

The startup sometimes hosts models before they have been publicly launched, giving the companies that produce them a way to get early feedback on their models. AI model developers publicly tout their models’ LMArena rankings as the differences between AI models have narrowed across the industry.

“Leading labs are using us because it has been challenging for them to know if their model is good or bad,” said CEO and co-founder Anastasios Angelopoulos.

Some model makers have complained that LMArena’s approach—tapping unpaid Internet users for feedback—is flawed, saying it can be gamed or doesn’t reflect expert opinion. In contrast, LMArena rivals such as data labelling startup Scale AI pays experts such as lawyers or professors to provide feedback on models.

LMArena has argued that random users are often better-positioned to grade the answers to their own questions and that LMArena gets more honest feedback by not paying experts.

“Some of the highest quality evaluation you can get—it’s called ‘golden standard’—is when people vote on the topics they know,” LMArena co-founder Ion Stoica told The Information late last year.

Stoica, the University of California, Berkeley computer science professor who previously co-founded Databricks, launched LMArena as a project called ChatBot Arena with Berkeley graduate students Angelopoulos and Wei-Lin Chiang. The startup was originally funded with grants and donations. It became a for-profit company in May.

The startup is considering expanding its offerings to include using its data to train AI models using reinforcement learning, the training technique that rewards a model for accomplishing certain goals and penalizes it for other behaviors, Stoica previously said. Feedback from people can be used to train models, though with the rise of “reasoning” models, it is increasingly popular to draw some of this feedback from other AI models, rather than from people. Expanding into reinforcement learning related services could help keep LMArena relevant to its customers’ needs.

“Once you become the de facto benchmarking layer, the product naturally expands,” said Peter Deng, a Felicis general partner who co-led the new financing. “The real value comes from deep partnerships with [AI] labs—combining their internal data with comparative external data.”

Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners and Laude Ventures also participated in the most recent funding, which brings the startup’s total funding to more than $250 million.

Le Monde : D’Auchan à Alinea, l’empire Mulliez fragilisé par les déboires du age

D’Auchan à Alinea, l’empire Mulliez fragilisé par les déboires du commerce
Après s’être longtemps appuyée sur le distributeur alimentaire pour financer le développement de ses autres enseignes, la famille nordiste compte désormais sur Leroy Merlin et Decathlon pour soutenir une galaxie pénalisée par les difficultés du commerce.

Pour vivre heureux, vivons cachés. S’il y a bien une famille entrepreneuriale française qui applique ce précepte à la lettre, ce sont les Mulliez. Ces derniers mois, la célèbre dynastie du Nord, propriétaire de Leroy Merlin, de Boulanger, de Kiabi, d’Electro Dépôt ou encore de Decathlon, s’est toutefois retrouvée, malgré elle, sous le feu des projecteurs à cause d’une série de déboires.

Le 25 novembre 2025, Auchan, le porte-avions de la flotte Mulliez, a annoncé sa volonté de transférer près de 300 supermarchés français sous bannière Intermarché et Netto. Cette mise en franchise sonne comme un aveu d’échec pour le distributeur, après des années de pertes et d’érosion de ses parts de marché, tout en laissant en suspens l’avenir de ses hypermarchés. D’autres vaisseaux essuient la tempête, comme l’enseigne d’ameublement Alinea, dont le placement en redressement judiciaire le 21 novembre, le deuxième en cinq ans, menace 1 200 emplois. Quant à Foundever (ex-Sitel), le géant des centres d’appels, qui emploie 150 000 collaborateurs dans le monde, il doit affronter, selon nos informations, une restructuration de sa dette aux Etats-Unis, l’équivalent d’une opération de chirurgie lourde.

A se demander si, cent vingt ans après l’acquisition d’une petite retorderie à Roubaix par Louis Mulliez, sa parentèle n’aurait pas perdu le fil de la success story. Le modèle qui a fait le succès de la huitième fortune française, selon le magazine Challenges, s’est-il essoufflé ? A l’évidence, la question dérange. Tous ceux à qui Le Monde a tenté de la poser ont soit répondu sous la condition de l’anonymat, soit décliné : un assortiment de « râteaux » digne du rayon jardinage d’un Leroy Merlin.

« Tous solidaires »
Discret, le clan n’en est pas moins redouté. Et tentaculaire. L’Association familiale Mulliez (AFM) regroupe 950 des 1 500 descendants de Louis Mulliez et de Marguerite Lestienne, qui avaient eu 11 enfants. Ce club d’investisseurs, unique en son genre, détient « 130 entreprises employant plus de 620 000 collaborateurs dans le monde, dont près de 175 000 en France », a détaillé Barthélémy Guislain, le président du conseil de gérance de l’AFM – un gendre de la cinquième branche et quatrième génération – devant les parlementaires, en mai 2025.

De quoi constituer le plus gros employeur privé français… s’il s’agissait d’un groupe. Mais cette collection d’entreprises unies par un actionnariat commun réfute cette appellation. De leur côté, les syndicats se battent pour que l’AFM soit assimilée à un groupe afin de favoriser les reclassements de salariés auprès des entités cousines. La cour administrative d’appel de Douai (Nord) doit se prononcer sur le sujet dans les prochains jours, dans le cadre de la procédure lancée par la fédération CGT du commerce et le CSE d’Auchan pour faire invalider le plan de sauvegarde prévoyant 2 389 suppressions de postes au sein de l’enseigne nordiste.

Au-delà de l’enjeu social, cette organisation décentralisée reflète la culture d’une famille élevée au bon grain de l’indépendance et de l’esprit d’entreprise. Sa retorderie étant trop petite pour satisfaire tous ses enfants, Louis Mulliez avait encouragé ses aînés à développer d’autres affaires. « Il impose à ses fils et à son beau-fils de se réunir tous les samedis matin pour partager leurs expériences respectives et les incite à être actionnaires les uns des autres, jetant ainsi les bases de notre gouvernance familiale », relatait Séverine Tapié-Mulliez, membre de la quatrième génération, dans Le Journal de l’Ecole de Paris du management en 2025. Et de préciser : « Au sein de l’AFM, nous sommes encore tous, à des niveaux variables, actionnaires de toutes les entreprises de la famille et tous solidaires de leur bonne marche. »

Ce système a fait ses preuves. Durant les « trente glorieuses » du commerce, surfant sur la vague d’hyperconsommation, il a favorisé l’émergence de fleurons tricolores, comme Decathlon, créé en 1976 par Michel Leclercq, un rejeton de la septième branche, ou Kiabi, lancé en 1978 par Patrick Mulliez, frère cadet de Gérard Mulliez, le fondateur d’Auchan (quatrième branche). Mais l’environnement a bien changé. Sur fond de concurrence féroce des Amazon, Shein et autres Temu, la « décommercialisation » – le pendant de la désindustrialisation – frappe désormais les ronds-points et les centres-villes. Une tourmente à laquelle l’AFM n’échappe pas, compte tenu de la surreprésentation du commerce physique dans son portefeuille.

Or, les bons connaisseurs de l’empire l’affirment, autant les Mulliez ont excellé dans la conquête, autant ils peinent à gérer les crises. « Ils n’aiment pas faire le sale boulot », tranche un financier. Si le clan se méfie des dirigeants venus de l’extérieur, il a tendance à s’appuyer aveuglément sur les « natifs » qui ont gravi les échelons en interne, du rayon peinture à la gestion de l’entrepôt. Ainsi, la confiance a été maintenue envers et contre tout à Yves Claude, patron d’Auchan Retail de novembre 2021 à novembre 2024, puis président jusqu’en avril 2025, pourtant contesté en interne alors que le distributeur s’enfonçait dans le rouge. L’homme, qui avait gagné ses galons chez Decathlon à l’international, était perçu comme proche de la branche Leclercq. Une dimension politique qui compte au sein d’une famille traversée de courants contraires, entre les anciens et les modernes, les fervents de la tech et ceux qui jurent par l’expérience du terrain.

Processus de création-destruction
« Nous faisons longtemps confiance à nos dirigeants. Nous croyons à la deuxième chance, voire à la troisième », reconnaît Barthélémy Guislain au Monde, se disant « convaincu que cette vision managériale apporte sur la durée plus d’avantages que d’inconvénients ». « Il faut du temps pour gravir les escaliers de la responsabilité, mais cela forme les meilleurs dirigeants. » Et de rappeler l’expérience de la crise due au Covid-19 : « Lors du confinement, en mars 2020, nos enseignes se sont retrouvées du jour au lendemain avec 95 % de leur chiffre d’affaires menacé, mais, à la fin, nous avons pris des parts de marché à Amazon parce que nous avons laissé l’initiative à nos collaborateurs. »

Elu à la tête de l’AFM pour la première fois en 2014, le chef d’orchestre a connu son lot d’épreuves. A commencer par la pandémie. A l’époque, le poids de l’empire était tel que le gouvernement avait demandé aux Mulliez de ne pas solliciter le report du paiement des cotisations Urssaf au niveau de leurs enseignes, tant le manque à gagner aurait été gênant pour l’Etat. L’invasion de l’Ukraine a ensuite mis sur la sellette Auchan et Decathlon en Russie. Pour autant, M. Guislain relativise : « Nos aïeux aussi ont eu leurs difficultés. Certains ont échoué cinq fois avant de réussir. » La diversification dans l’hôtellerie a fait long feu. Les sandwichs Pic Pain ont rassis. Surcouf a été débranché. Boulanger a peiné une décennie avant de remonter la pente.

Le système repose sur la solidarité, les entreprises bien portantes soutenant les autres. Phildar, né en 1946, a longtemps été le bas de laine dans lequel la famille a puisé pour irriguer sa pépinière. Le roi du fil à tricoter a aidé Auchan, à ses débuts en 1961, qui a lui-même été la « mère nourricière » de Decathlon, Leroy Merlin ou Flunch. Phildar, placé en redressement judiciaire en 2020, aurait-il connu un meilleur destin s’il n’avait pas été déshabillé pour habiller Auchan ? Pas sûr. Le distributeur aurait-il mieux résisté à la mutation du commerce alimentaire s’il avait distribué moins de dividendes ? Peut-être.

Cette circulation du capital n’est pas toujours du goût des salariés. En décembre 2024, les syndicats de Decathlon s’étaient ainsi indignés contre le milliard d’euros de dividendes versé par l’enseigne de sport (à comparer avec un résultat net de 787 millions d’euros en 2024) aux actionnaires. Cette manne, qui a bénéficié en partie aux 60 000 salariés actionnaires de Decathlon, a permis à l’AFM d’injecter 400 millions d’euros en 2024, puis 600 millions en 2025 chez Auchan.

Dans ce processus de création-destruction très schumpétérien, une vache à lait remplace l’autre. Ainsi, selon les Mulliez, Auchan France ne représente plus que 5 % de la valeur de l’AFM, et c’est désormais Adeo, la holding regroupant Leroy Merlin, Weldom, Bricoman ou encore Saint-Maclou, qui pèse le plus dans la cagnotte familiale, suivie par Decathlon. L’important est d’avoir toujours une vache d’avance.

Pour Barthélémy Guislain, « certaines de [leurs] entreprises les plus anciennes » ont encore un beau potentiel de développement, notamment à l’international. « La rénovation énergétique de l’habitat, l’accessibilité du sport au plus grand nombre, la mode à petit prix pour les familles : nos entreprises sont utiles et, quand elles font bien leur métier, ça marche », explique-t-il, avant d’insister : « Notre modèle familial entrepreneurial, chrétien du nord de la France, repose sur l’idée que la valeur se crée par les gens et par le partage. Cela peut paraître anachronique à l’heure où l’intelligence artificielle tend à remplacer l’humain, mais c’est en fait très moderne. »

Derrière les grands principes, cependant, les lignes bougent. Ces dernières années, le culte du « chacun chez soi » a été écorné par la mise en commun des données clients des Auchan et autres Decathlon afin de pouvoir rivaliser avec les géants de l’e-commerce sur le ciblage publicitaire. Partager les parkings sur les centres commerciaux, c’est une chose, mais, quand il s’est agi de mutualiser les tickets de caisse, le pas a été difficile à franchir. Il a fallu des mois de palabres en amont avant d’obtenir l’adhésion de toutes les enseignes au projet Valiuz, officialisé en 2019. Le recrutement, en septembre, d’un ancien du géant de la grande distribution Walmart montre la volonté de transformer cette alliance en un nouveau métier, dérivé du commerce comme l’est déjà l’immobilier.

L’enjeu est majeur alors que certaines des diversifications lancées par Creadev, la société d’investissement de la famille, sont à la peine. En premier lieu, Foundever, malmené par le déferlement des agents conversationnels, comme son grand concurrent TP (ex-Teleperformance). En quinze ans, les Mulliez ont investi 450 millions d’euros dans l’entreprise qui a pu valoir jusqu’à 4 milliards. Mais, contrairement aux Peugeot ou aux Wendel qui font tourner leurs actifs, la famille nordiste rechigne à vendre. En 2021, elle a ainsi refusé, selon nos sources, une offre de rachat émanant de Webhelp. Désormais, Foundever est au cœur d’un bras de fer avec ses créanciers dans lequel les Mulliez devraient laisser des plumes. Autres investissements de Creadev, Maisons de famille, après une acquisition ratée en Allemagne, subit la tourmente dans le secteur des Ehpad, quand le producteur d’insectes Innovafeed est, quant à lui, pénalisé par la déconfiture de son concurrent Ynsect.

Cette année, comme tous les dix ans, l’AFM mène une introspection afin de définir où la famille se voit dans vingt ans. Vaste programme. D’ici là, le capitalisme héréditaire à la Mulliez devra répondre à d’autres défis, à commencer par l’arrivée de la cinquième génération.

The Information : Nvidia Unveils Rubin and Announces New Open Source Autonomous

Nvidia Unveils Rubin and Announces New Open Source Autonomous Driving Model

Nvidia CEO Jensen Huang on Monday announced a new autonomous driving model that aims to help cars reach Level 4 autonomous driving, in which a driver isn’t required to supervise the vehicle.

Separately, Huang also announced more detailed specifications for the company’s upcoming next-generation flagship AI chips, known as Rubin. The family of chips are designed to reduce the cost of generating AI tokens by as much as tenfold, when compared with its previous generation chips, the company said.

They also aim to cut the number of GPUs needed to train mixture-of-experts models by a factor of four. Those models use far less computing power by sending each query only to the most relevant parts of the model, rather than activating the entire model at once.

Speaking in a keynote presentation at the annual Consumer Electronics Show in Las Vegas, Huang said Nvidia’s new family of self-driving models called Alpamayo are the world’s first that can think and reason. “We imagine that someday a billion cars on a road will all be autonomous,” he said.

Nvidia is making its Alpamayo models open source, which allows outside developers to build their own self-driving cars based on its tech. That could ultimately drive sales of more Nvidia chips.

Huang said the first cars based on Nvidia’s self-driving tech would be made by Mercedes-Benz and be on the road by the first quarter of this year in the U.S. Those cars, however, will only offer Level 2 autonomous driving, which requires drivers to still supervise the vehicle.

The Information : The AI Boom Is Now an Energy Boom

The AI Boom Is Now an Energy Boom

The AI race has ignited a parallel boom in energy that is already revamping U.S. power generation and changing natural gas infrastructure, utility systems and regulation in ways that could last for decades.

Natural gas producers are expanding pipelines faster than they have in years, utilities have secured regulatory approval for billions of dollars in spending and power-equipment manufacturers’ backlogs are allowing them to scale up factories.

The industries have become powerful supporters of the AI buildout, giving the U.S. tech industry political and economic support for its frantic timeline to win the AI race.

One point of near-unanimous agreement is that the U.S. power system, from utilities to equipment manufacturers to natural gas producers, requires far more investment in decades-old infrastructure to keep up with what’s coming. Asking a binary question—will the AI bubble burst or not?—ignores what else the boom has set in motion and who has an interest in keeping it going.

There’s going to be $6.7 trillion spent on AI and cloud-computing from 2025 through 2030, McKinsey estimates. Spending on IT equipment (chips, servers and storage) accounts for “just” $4.4 trillion of that total. Construction labor, materials and land for data centers add $1 trillion. Electrical and mechanical equipment and fiber networks add nearly another $1 trillion.

McKinsey’s estimate doesn’t include the full $1.1 trillion that regulated utilities have now said they expect to spend—and charge ratepayers for—to support electricity demand growth over the same period, thanks primarily to AI.

The estimate also excludes the $50 billion in planned new pipelines and tens of billions of dollars in private transmission projects that AI demand is helping to underwrite. It may also undercount billions of dollars in additional spending on on-site power generation that tech giants are building themselves. Manufacturers like Eaton, GE Vernova, Schneider Electric and Vertiv have also collectively announced billions of dollars in factory expansions to build power and cooling equipment.

Natural gas industry CEOs tell me they see the AI race as a once-in-a-generation opportunity to get capital and political backing to build infrastructure in markets where climate activists and not-in-my-backyard opponents had led them to cancel projects or curb production. Tech giants have softened their stance on using fossil fuels and policymakers have begun greenlighting interstate pipeline expansions again after energy companies framed the investments as a national imperative to dominate the AI race. Gas that was once stranded without an outlet—because at least 90% of it travels by pipeline—will flow into these new markets for decades to come.

Toby Rice, CEO of EQT Corp., the largest U.S. natural gas producer, announced a deal in July to supply gas to a massive, and still prospective, off-grid AI data center project on the site of a recently imploded coal plant in Homer City, Pennsylvania. It would fuel what would be the country’s largest gas plant, which could run as much as 4.4 gigawatts of AI computing power in the most congested power corridor in the country.

EQT has also quickly filled available capacity on its Mountain Valley Pipeline through Appalachia, which was opened in June 2024, thanks to deals with utilities such as Duke Energy and Southern Co. The two companies said data centers account for nearly all of their projected new power load. Although the 303-mile pipeline took a decade to push through, EQT last year won federal approval to extend it into North Carolina. Duke has already bought space on it.

Meanwhile, Chad Zamarin, CEO of gas pipeline operator Williams Cos., has built a $5.1B portfolio of power generation and transport projects in the company’s newly named Power Innovation portfolio which will supply power directly to data centers. Projects include a gas plant for Meta’s Socrates project in Ohio and sites in Virginia’s Data Center Alley. Williams has also invoked the AI race to urge Congress to pass permitting reform. The latest attempt, the Standardizing Permitting and Expediting Economic Development (SPEED) Act, just passed the House Dec. 18 and is headed to debate in the Senate.

Then there are investor-owned utilities. Demand growth has been essentially zero for 20 years, limiting what they could spend on new facilities and upgrades. Not anymore. The industry has just unveiled plans for $1.1 trillion in capital expenditures in the next five years, double their rate of spending the past 10 years, according to the trade group Edison Electric Institute.

Once utilities get regulatory approval for such spending, they officially win the right to charge a guaranteed rate of return to erect plants, poles and wires for the life of 20- to 30-year projects. Come what may with AI, monopoly utilities will still be “making more by spending more.” (Nice business model if you can get it.) This is not to say they won’t face wrath if they’ve charged everyday people for wasteful projects. But it turns out, we’re so short of modern power infrastructure that we’ll likely still need it when the froth boils off.

Clean energy producers are also lunging at opportunity. They’re mostly happy to build solar and battery farms with gas backup to sate AI’s power habits, because they, too, need to scale.

The struggling U.S. battery industry is also getting a boost. The publicly traded battery company Eos Energy Enterprises was able to borrow $600 million and close a $458 million stock offering in November. Its batteries can be used as backup power for data centers and the electricity grid. CEO Joe Mastrangelo said the financing allowed Eos “to scale with speed and certainty at a time when the world is entering an energy supercycle.”

Metal and commodity markets that have been in the doldrums for years are suddenly in high demand. Plans by Microsoft, Meta and Google to revive or extend licenses for old nuclear reactors means companies are scrambling for sources of U.S. produced uranium.

What happens if the AI buildout falters? Energy has always been a boom-and-bust industry, and its executives know a few things about surviving downturns. Big energy company executives tell me they’re avoiding counterparties with “x” or “Open” in their names or with credit that’s not pristine as they negotiate capital-intensive projects like carbon capture for data center gas power. This could give the tech giants an important advantage if growth slows.

This is not to say the booms across industries aren’t intertwined and that real estate and construction companies, power providers and equipment makers won’t suffer if overleveraged AI players default on projects. But data center capacity is still scarce, with vacancy at an all-time low of 1.6%, says CBRE.

JP Morgan senior analyst Stephen Tusa, who follows data center cooling and power management leader Vertiv, points out that the company has grown revenues from $5 billion to $10 billion since 2021 and still has 1.3 times more orders in process than it booked as revenue this year. “We’re calling it a galactic cycle,” Tusa said. “Because I’ve never seen anything in my 27-year career that looks anything like that.”

In related “supercycle” news:

• Texas’s grid operator, the Electric Reliability Council of Texas, or ERCOT, in December voted unanimously to approve the second, $9.4 billion phase of a buildout expected to cost up to $33 billion for extra-high-voltage (765 kV) transmission. Even if the data center build falls short, its VP of system planning says its modeling shows the spending is justified by rapid electrification of energy production and industrial expansion.

• Robust demand from AI data centers helped Austin-based solar and storage company T1 Energy raise funds to repay debt from a Chinese partner and move closer to building an end-to-end American-made solar supply chain in a market long dominated by China. (T1 used to be based in Europe and known as FREYR Battery; a new U.S. management team took over and rebranded.)

The Information : Google’s Samsung AI Advantage

Google’s Samsung AI Advantage


Can anyone catch up with Google in the AI race? That question came to mind on Monday, after Reuters quoted Samsung’s co-CEO talking about his plans to double the number of mobile devices running Galaxy AI, Samsung’s branded AI features backed by Google’s Gemini AI technology, to 800 million this year. (That’s separate to Samsung phones that have the Gemini consumer chatbot preinstalled.) And the two companies are not just collaborating on phones: Samsung plans to unveil Gemini-powered AI features for kitchen appliances at CES this week. They’re also working together on smart glasses featuring Gemini. Beyond Samsung, Google is also installing Gemini on TV sets running its Google TV software—it’s previewing some new features at CES this week.

But phones are key. This is where Google may soon have an unbeatable lead. Samsung has the top share of the global smartphone market, although some analysts have predicted Apple will take that spot once the final numbers for 2025 are tabulated. That tussle doesn’t matter much to Google, as it is expected to negotiate a deal to help power Apple’s Siri assistant on iPhones. You might say Google’s AI has locked up the mobile market, at least for the moment. Whether it proves permanent is hard to say, as Google’s arrangement with Samsung is likely for a few years only and it hasn’t yet finalized a deal with Apple.

Even with that qualification, this is a good reminder of why OpenAI is so keen to launch some kind of AI device, and why Meta is putting so much effort into AI for its smart glasses. Both companies are hoping to supplant phones. They might succeed, but don’t bet on it. In the meantime, Google’s Gemini models will be powering AI features on many different outlets. That should mean Google is able to improve how the models function on a variety of tasks, simply because of the data it gets from interacting with so many consumers. And that could make Google’s models even more attractive to potential business partners.

What about money? Google is presumably charging Samsung for use of its models, although Google may have been keen enough for the business to discount its fee. Indeed, one of the revelations in an antitrust court hearing last year is that Google paid “enormous” amounts of money to have the Gemini consumer chatbot preinstalled on Samsung and other phones running Google’s Android software. But even without the chatbot, Google features like Gemini Live (through which you can talk to Gemini about what you see on your phone screen) will be available via Galaxy AI, according to this Samsung description of the technology. That’s valuable for raising consumer awareness of Gemini and potentially giving Google a bigger audience to sell ads to. OpenAI’s ChatGPT may be the only AI the average person knows about right now, but Google’s AI is looking to become increasingly ubiquitous.

Wall Street Ignores Jensen
Nvidia CEO Jensen Huang was on fire in his appearance at CES on Monday afternoon. Wearing a shiny leather jacket that appeared to be made of crocodile leather, he spoke off the cuff about developments in AI before segueing into a series of product and software announcements, culminating with details of Nvidia’s latest AI chip, Rubin, and related hardware.

Wall Street, though, wasn’t impressed. Huang began his address after the close of regular trading, when Nvidia stock dipped fractionally. But in after-market trading, Nvidia stock barely moved. And when it did, it fell by a few pennies.