Google's AI infrastructure chief told employees the company must double its AI serving capacity every six months in order to meet demand. In a presentation earlier this month, Amin Vahdat, a vice president at Google Cloud, gave a presentation titled "AI Infrastructure." It included a slide on "AI compute demand" that said: "Now we must double every 6 months.... the next 1000x in 4-5 years." CNBC reports: The presentation was delivered a week after Alphabet reported better-than-expected third-quarter results and raised its capital expenditures forecast for the second time this year, to a range of $91 billion to $93 billion, followed by a "significant increase" in 2026. Hyperscaler peers Microsoft, Amazon and Meta also boosted their capex guidance, and the four companies now expect to collectively spend more than $380 billion this year. Google's "job is of course to build this infrastructure but it's not to outspend the competition, necessarily," Vahdat said. "We're going to spend a lot," he said, adding that the real goal is to provide infrastructure that is far "more reliable, more performant and more scalable than what's available anywhere else." In addition to infrastructure build-outs, Vahdat said Google bolsters capacity with more efficient models and through its custom silicon. Last week, Google announced the public launch of its seventh generation Tensor Processing Unit called Ironwood, which the company says is nearly 30 times more power efficient than its first Cloud TPU from 2018. Vahdat said the company has a big advantage with DeepMind, which has research on what AI models can look like in future years. Google needs to "be able to deliver 1,000 times more capability, compute, storage networking for essentially the same cost and increasingly, the same power, the same energy level," Vahdat said. "It won't be easy but through collaboration and co-design, we're going to get there."

Read more of this story at Slashdot.

NVIDIA today issued the 580.94.11 driver release as their newest Vulkan beta driver for Linux customers. Most notable with this beta driver update is adding VK_EXT_hdr_metadata support...

An anonymous reader quotes a report from Ars Technica: The US crackdown on chip exports to China has continued with the arrests of four people accused of a conspiracy to illegally export Nvidia chips. Two US citizens and two nationals of the People's Republic of China (PRC), all of whom live in the US, were charged in an indictment (PDF) unsealed on Wednesday in US District Court for the Middle District of Florida. The indictment alleges a scheme to send Nvidia "GPUs to China by falsifying paperwork, creating fake contracts, and misleading US authorities," John Eisenberg, assistant attorney general for the Justice Department's National Security Division, said in a press release yesterday. The four arrestees are Hon Ning Ho (aka Mathew Ho), a US citizen who was born in Hong Kong and lives in Tampa, Florida; Brian Curtis Raymond, a US citizen who lives in Huntsville, Alabama; Cham Li (aka Tony Li), a PRC national who lives in San Leandro, California; and Jing Chen (aka Harry Chen), a PRC national who lives in Tampa on an F-1 non-immigrant student visa. The suspects face a raft of charges for conspiracy to violate the Export Control Reform Act of 2018, smuggling, and money laundering. They could serve many decades in prison if convicted and given the maximum sentences and forfeit their financial gains. The indictment says that Chinese companies paid the conspirators nearly $3.9 million. One of the suspects was briefly the CTO of Corvex, a Virginia-based AI cloud computing company that is planning to go public. Corvex told CNBC yesterday that it "had no part in the activities cited in the Department of Justice's indictment," and that "the person in question is not an employee of Corvex. Previously a consultant to the company, he was transitioning into an employee role but that offer has been rescinded."

Read more of this story at Slashdot.

British soldiers are using computer games such as Call of Duty to sharpen their "war-fighting readiness," an Army chief has said. From a report: General Sir Tom Copinger-Symes, the deputy commander of Cyber and Specialist Operations Command, said the war in Ukraine, where remote-operated drones have become crucial on the battlefield, proved the worth of having soldiers skilled in video gaming. The Ministry of Defence on Friday announced the launch of the International Defence Esports Games (IDEG), a video gaming tournament that will pit the best of Britain's "future cyber warriors" against military teams from 40 other countries.

Read more of this story at Slashdot.

The Japanese government said that the world's biggest nuclear plant would restart operations. Semafor: The Kashiwazaki-Kariwa site closed in 2012, as Japan -- which previously generated 30% of its electricity from nuclear power -- shuttered most of its fleet in the wake of the Fukushima meltdown. But like much of the world, it is looking once again to nuclear power for reliable, low-carbon energy, especially in the face of high gas and oil prices following Russia's invasion of Ukraine. It has restarted 14 out of 54 plants and announced plans for a first new reactor since the disaster.

Read more of this story at Slashdot.

Google confirmed in a statement Friday that hackers have stolen the Salesforce-stored data of more than 200 companies in a large-scale supply chain hack. TechCrunch reports: On Thursday, Salesforce disclosed a breach of "certain customers' Salesforce data" -- without naming affected companies -- that was stolen via apps published by Gainsight, which provides a customer support platform to other companies. In a statement, Austin Larsen, the principal threat analyst of Google Threat Intelligence Group, said that the company "is aware of more than 200 potentially affected Salesforce instances." After Salesforce announced the breach, the notorious and somewhat-nebulous hacking group known as Scattered Lapsus$ Hunters, which includes the ShinyHunters gang, claimed responsibility for the hacks in a Telegram channel, which TechCrunch has seen.

Read more of this story at Slashdot.

Microsoft has acknowledged in a support article that major Windows 11 core features including the Start Menu, Taskbar, File Explorer and System Settings break after applying monthly cumulative updates released on or after July 2025. The problems stem from XAML component issues that affect updates beginning with July's Patch Tuesday release (KB5062553). The failures occur during first-time user logins after cumulative updates are applied and on non-persistent OS installations like virtual desktop infrastructure setups. Microsoft lists Explorer.exe crashes, shellhost.exe crashes, StartMenuExperienceHost failures and System Settings that silently refuse to launch among the symptoms. The company provided PowerShell commands and batch scripts as temporary workarounds that re-register the affected packages. Both Windows 11 versions 24H2 and 25H2 share the same codebase and are affected. Microsoft said it is working on a fix but did not provide a timeline.

Read more of this story at Slashdot.

In addition to Valve contributing to the open-source Radeon Vulkan driver for enhancing the Linux gaming experience and their AMD-powered Steam Deck, the upcoming Steam Frame VR headset is making use of Mesa's open-source "Turnip" Vulkan driver for Qualcomm Adreno graphics...

Imaginez un SSD mobile avec un petit bouton qui permet de le détruire de manière irréversible. Le T-Create Expert P35S tient exactement cette promesse. 

Avec un petit bouton sur son boitier, le T-Create Expert P35S permet à son utilisateur de lancer une procédure qui va littéralement détruire les puces de stockage embarquées. Une solution extrême destinée à protéger à tout prix des données ultrasensibles.

Créé par la société TeamGroup, le dispositif est prévu pour lancer un processus autodestructeur totalement invisible. Un brevet a même été déposé pour la méthode employée pensée pour ne laisser aucune chance aux données. L’action à mener n’est pas anodine. Un gros bouton est visible sur la surface en ABS du P35S. En appuyant fermement dessus, il se déploiera dans un premier temps vers une étape d’avertissement préalable. Révélant une surface rouge indiquant que la prochaine pression sera fatale pour vos données. Cette première étape demande une pression assez ferme. La seconde, qui mènera à la destruction proprement dite des données, demande d’appuyer très fortement sur le bouton pour le basculer entièrement vers le haut. Le dispositif est pensé pour qu’il ne soit pas possible d’appuyer dessus par inadvertance mais que l’action soit facilement réalisable dans l’urgence.

Parfait pour stocker des données sensibles en déplacement comme des secrets d’État, des brevets industriels ou le manuscrit secret du prochain film de Nicolas Cage. le P35S n’est pas vraiment pensé pour les particuliers. L’objet n’a pas encore de tarif mais il devrait être proposé surtout aux professionnels. Avec des capacités allant de 256 Go à 2 To, il se connectera grâce à un câble USB Type-C avec un débit annoncé de 1000 Mo/s au maximum en lecture comme en écriture grâce à un port USB 3.2 Gen2. 

À noter que l’objet semble avoir besoin d’un USB pour lancer son mode de destruction automatique. Si vous ne connectez pas le SSD P35S à une source d’alimentation, il ne se passera rien. Mais même en rebasculant l’interrupteur en position basse, le moindre branchement entrainera la fin de votre stockage. Le SSD mesure 9 cm de large pour 4 de large et 1.8 cm d’épaisseur. Il pèse 42 grammes.

Le SSD mobile T-Create Expert P35S peut s’autodétruire © MiniMachines.net. 2025

Il y a exactement 15 jours de cela, nous faisons le point sur H&Co sur la montée des prix de la DDR5 en France. Nous avions alors pris pour l'exemple une configuration de kit devenue classique en cette fin 2025 : 32 Go (2 x 16 Go) DDR5-6400 C32 et avions cherché chez les principaux revendeurs fr...

Thunderbird Pro has moved its Thundermail email service into production testing as the open-source email client's subscription bundle of additional services prepares for an Early Bird beta launch at $9 per month that will include email hosting, encrypted file sharing through Send, and scheduling via Appointment. Internal team members are now testing Thundermail accounts and the new Thunderbird Pro add-on automatically adds Thundermail accounts for users who sign up through it. The project migrated its data hosting from the Americas to Germany and the EU. Appointment received a major visual redesign being applied across all three services while Send completed an external security review and moved from its standalone add-on into the unified Thunderbird Pro add-on. The new website at tb.pro is live for signups and account management.

Read more of this story at Slashdot.

Adam Marshall spent more than a decade developing Kingdoms of the Dump while working as a custodian at a school in suburban Philadelphia, cleaning floors and hauling trash bags from 3 PM to 11 PM before coming home to work on his turn-based role-playing game until 5 or 6 AM. The game, which Bloomberg has called "one of the year's most charming RPGs," came out on Tuesday after Marshall and his childhood friend Matt Loiseau -- also a janitor -- built it using RPG Maker alongside a small team of hobbyists who mostly worked for free. The pair launched a Kickstarter campaign in 2019 that raised $76,560, but the pandemic disrupted their plans and forced them to lose contractors and rethink their approach. Marshall maintained this schedule for five years straight before quitting his custodial job last year to finish the game full-time. Kingdoms of the Dump has sold about 7,000 copies since its release. The game stars a walking trashcan named Dustin Binsley who adventures through landfills and sewers in a world made entirely of garbage.

Read more of this story at Slashdot.

Back in mid-2024, the Bavarian Linux PC vendor TUXEDO Computers teased plans for developing a Snapdragon X Elite Linux laptop. Initially they hoped to have it out by Christmas 2024. That didn't happen and now approaching Christmas 2025 they confirmed they have stopped their plans for shipping a Snapdragon X1 Elite laptop for Linux customers...

AI nutrition tracking features in popular fitness apps are producing wildly inaccurate calorie and macro counts despite promises to simplify food logging through automated photo analysis. The Verge tested AI-powered nutrition tools in Ladder, Oura Advisor, January and MyFitnessPal. Ladder's AI estimated the outlet's carefully measured 355-calorie breakfast at 780 calories and got the macro breakdown wrong even after the reviewer manually edited entries to include exact brands and amounts. Oura Advisor routinely mistook matcha protein shakes for green smoothies. January misidentified barbecue sauce as teriyaki sauce and failed to detect mushrooms in a chicken dish. None of the apps could identify healthier ingredient swaps or accurately log ethnic foods. Oura classified a mix of edamame, quinoa and brown rice as mashed potatoes and white rice. Ladder logged dal makhani curry as chicken soup. The AI features require extensive manual corrections that negate any time savings from automated logging, the publication concluded in its scathing review.

Read more of this story at Slashdot.

Amazon's 14,000-plus layoffs announced last month touched almost every piece of the company's sprawling business, from cloud computing and devices to advertising, retail and grocery stores. But one job category bore the brunt of cuts more than others: engineers. CNBC: Documents filed in New York, California, New Jersey and Amazon's home state of Washington showed that nearly 40% of the more than 4,700 job cuts in those states were engineering roles. The data was reported by Amazon in Worker Adjustment and Retraining Notification, or WARN, filings to state agencies. The figures represent a segment of the total layoffs announced in October. Not all data was immediately available because of differences in state WARN reporting requirements.

Read more of this story at Slashdot.

One of the areas I've been meaning to run more benchmarks on this season has been for the recently released Clang 21 compiler. Back in September when LLVM Clang 21 was debuting I ran some initial benchmarks and found it to deliver some nice performance gains on AMD EPYC Zen 5 but then have been busy with other benchmarks/articles for expanding that testing. Recently with having some spare cycles and gratis access still to the Microsoft Azure HBv5 instance for AMD EPYC Zen 4 with HBM3, I ran some Clang 20 vs. Clang 21 performance benchmarks there for those wondering about any performance benefits of this new compiler release on Zen 4.

Plus dure sera la chute

Plusieurs patrons de la tech ont récemment commenté l’hypothèse de l’éclatement d’une bulle financière autour de l’IA, ravivant les inquiétudes de certains observateurs. Les derniers résultats de NVIDIA n’ont pas suffi à rassurer les marchés jeudi, en dépit d’une croissance de 62 % sur un an.

En cette période de résultats trimestriels, les boursicoteurs engagés sur les valeurs liées à la tech et à l’IA ont retenu leur souffle : après des mois d’un rallye haussier d’une rare intensité, la machine boursière a en effet connu quelques ratés ces derniers jours. Ces soubresauts ont ravivé l’inquiétude de ceux qui appréhendent l’éclatement d’une bulle de l’IA, et les déclarations récentes de Sundar Pichai n’ont rien fait pour calmer leurs angoisses.

Google atteint des sommets et n’exclut pas l’éclatement

D’ordinaire plutôt discret, le patron de Google a répondu le 18 novembre dernier aux questions de la BBC, et il a admis que si l’on traversait une période « extraordinaire » du fait des investissements massifs dans l’IA, ces derniers présentaient aussi des éléments d’irrationalité. Quid dans ce contexte des risques associés à l’éclatement d’une bulle ? « Je pense qu’aucune entreprise ne sera épargnée, nous y compris », lâche Sundar Pichai.

Google n’a pour l’instant pas de soucis à se faire. L’entreprise a publié fin octobre les meilleurs résultats trimestriels de son histoire, avec un chiffre d’affaires qui franchit pour la première fois la barre des 100 milliards de dollars, en hausse de 16 % sur un an.

Et l’IA est présentée comme l’un des moteurs de cette croissance. « En plus de dominer les classements, nos modèles propriétaires, comme Gemini, traitent désormais 7 milliards de jetons par minute, grâce à l’utilisation directe de l’API par nos clients. L’application Gemini compte aujourd’hui plus de 650 millions d’utilisateurs actifs mensuels. Nous continuons de générer une forte croissance dans de nouveaux secteurs d’activité », avance ainsi Sundar Pichai dans le communiqué qui accompagne ses résultats.

Bien que Google soit actuellement à son plus haut historique à Wall Street, et la seule des grandes valeurs de la tech à rester dans le vert suite aux derniers cahots du marché, c’est l’engagement financier nécessaire pour tenir son rang dans la course à l’IA qui semble alerter Sundar Pichai.

« Il y a environ quatre ans, Google dépensait moins de 30 milliards de dollars par an [en capital dédié à l’investissement, NDLR], cette année ce nombre sera supérieur à 90 milliards de dollars. Et si on additionne ce que font, ensemble, toutes les sociétés, on arrive à plus de 1000 milliards de dollars d’investissements dans la construction des infrastructures du moment », déclare-t-il.

Autrement dit, ceux qui auront trop investi durant cette phase risquent d’en subir les conséquences, indique Pichai, qui n’annonce cependant aucun ralentissement des dépenses d’Alphabet, notamment parce que les modèles développés par le groupe enregistrent des progrès « tout à fait excitants », se déploient toujours plus largement dans les produits du groupe et, in fine, « sont utilisés ».

Une bulle des LLM plutôt qu’une bulle de l’IA ?

Au fait, si bulle il y a, de quel périmètre parle-t-on vraiment ? La question n’est pas triviale, dans la mesure où les investissements en matière d’intelligence artificielle interviennent à plusieurs niveaux : l’achat de composants informatiques, mais aussi la construction de centres de données, ou la création des infrastructures énergétiques chargées d’alimenter ces derniers. De la même façon, il n’y a pas une unique IA, mais différents domaines dans lesquels l’intelligence artificielle trouve des débouchés.

Meta is venturing into the complex world of electricity trading, betting it can accelerate the construction of new US power plants that are vital to its AI ambitions. From a report: The foray into power trading comes after Meta heard from investors and plant developers that too few power buyers were willing to make the early, long-term commitments required to spur investment, according to Urvi Parekh, the company's head of global energy. Trading electricity will give the company the flexibility to enter more of those longer contracts. Plant developers "want to know that the consumers of power are willing to put skin in the game," Parekh said in an interview. "Without Meta taking a more active voice in the need to expand the amount of power that's on the system, it's not happening as quickly as we would like."

Read more of this story at Slashdot.

An anonymous reader shares a report: Microsoft is upgrading its Advanced Paste tool in PowerToys for Windows 11, allowing you to use an on-device AI model to power some of its features. With the 0.96 update, you can route requests through Microsoft's Foundry Local tool or the open-source Ollama, both of which run AI models on your device's neural processing unit (NPU) instead of connecting to the cloud. That means you won't need to purchase API credits to perform certain actions, like having AI translate or summarize the text copied to your clipboard. Plus, you can keep your data on your device.

Read more of this story at Slashdot.

Germany's Sovereign Tech Fund / Sovereign Tech Agency has been a godsend the past few years for the open-source community. This funding from the German government has led to significant funding for dozens of prominent open-source infrastructure projects to provide more resources for enhancing security, enabling new features, and more. As the Sovereign Tech Fund prepares for the next phase of growth, they are hiring a new head to lead the efforts...