The French government outlined a major sovereign AI push, declaring its intelligence agency would stop working with US company Palantir in favour of a local alternative alongside a pledge to invest €655 million in the technology through to 2030.
Prime Minister Sebastien Lecornu announced the plans in a video on X as the country advances its sovereign technology ambitions and a desire to reduce dependence on US offerings.
The debate around sovereignty continues to heat up: the US recently hit the brakes on a couple of Anthropic models due to national security concerns.
Prime Minister Lecornu pointed to the issue in announcing the decision to cut ties with Palantir, stating “France must have its own tools”.
“We cannot rely on the goodwill of certain partners who, as we have seen in recent days, are capable of cutting off access to the Anthropic model.”
Intelligence agency DGSI is to replace Palantir’s data tools with an alternative made by privately-controlled French company ChapsVision.
Palantir had worked with the agency for around a decade and announced a three-year renewal in December 2025.
Concerns have been raised about Palantir due to ambitions in Europe to reduce the amount of sensitive data processed by US-based companies.
The company was co-founded by billionaire Peter Thiel, who has strong links to the US government.
Revolution The prime minister said the €655 million investment would go towards infrastructure, computing capacity, research, business and industrial sectors.
As part of the plans, the government aims to provide all civil servants with a conversational assistant from domestic company Mistral AI, in addition to a public health assistant by the end of the year.
“We can undergo this revolution. Or we can lead it. France has chosen. It will rise to the challenge of this revolution and intends to claim its full place in it,” Prime Minister Lecornu said.
A US federal judge reportedly dismissed xAI’s trade secret lawsuit against rival OpenAI with prejudice, dealing another legal blow to Elon Musk in his widening battle with Sam Altman’s company.
Reuters reported Elon Musk-owned xAI failed to demonstrate OpenAI induced former xAI engineer Xuechen Li to misappropriate trade secrets, or that Li disclosed any confidential information during a presentation he gave while OpenAI was recruiting him.
In the order, the judge wrote xAI essentially equated asking a job candidate about prior work experience with encouraging a potential employee to divulge trade secrets obtained during that work.
The news agency stated the dismissal with prejudice means xAI cannot raise the same claims again.
The judge already dismissed an earlier version of the lawsuit in February, giving xAI an opportunity to strengthen its case. Having found xAI failed to cure the shortcomings in its amended complaint, the judge ruled it would be futile to allow further attempts.
The lawsuit was originally filed in September 2025, alleging OpenAI engaged in a coordinated campaign to hire away xAI employees to access trade secrets related to its Grok chatbot, including source code and data centre deployment strategies.
Musk’s AI company claimed former employees took confidential material when they left for roles at OpenAI.
Reuters noted OpenAI previously called the lawsuit baseless and part of Musk’s ongoing campaign of harassment against the company.
The case is part of a broader legal battle between Musk and OpenAI, which he co-founded alongside OpenAI CEO Sam Altman before departing.
Musk is separately suing OpenAI over its conversion to a for-profit company, while the AI firm has countersued Musk for harassment.
xAI also sued Apple in federal court, alleging it conspired with OpenAI to suppress rival platforms.
Cisco highlighted rising pressure on enterprise networks from AI deployments, warning campus and branch infrastructure is emerging as a potential bottleneck for companies scaling the technology.
In a study conducted with UK-based technology media company Foundry, Cisco found organisations are caught between accelerating AI ambitions and network constraints which could slow progress. It stated more than half of enterprises now have widespread generative AI (genAI) deployments. However, 73% of respondents said AI expansion will cause them to hit campus and branch capacity limits within 24 months.
Indeed, pressure is already building as respondents reported a 34% average increase in campus and branch network traffic tied to AI workloads over the past year, with a further 96% spike expected within the coming year.
Respondents reported AI was already reshaping traffic patterns, with 67% citing increases in internal data traffic tied to AI workloads, while 61% saw growth in continuous automated traffic generated by AI systems. “Suddenly, three AI agents are trying to talk to each other and solve a problem. That is going to be a big thing,” noted the head of AI strategy for global IT and network engineering operations at a US technology company, who was left anonymous.
Cisco highlighted majority of the AI readiness debate has focused on GPUs, cloud platforms and data centre buildouts, while pressure on workplace networks has been underplayed.
Security Beyond capacity, security is also emerging as a brake on AI scale. Respondents pointed to expanded attack surfaces, shadow AI activity, weak AI-driven traffic visibility and inconsistent policy enforcement as key challenges.
The report found 91% of UK businesses are adding security controls specifically to support new AI workloads. More than half of UK respondents added their current campus and branch security posture may be insufficient for next-generation AI workloads.
“The issue from a security standpoint is that it’s hard to create the guardrails for every possible AI tool that your organisation must use,” a US-based retail executive explained.
According to Cisco, AI is ultimately creating “fundamentally different operational environments”, making network modernisation “a prerequisite for operating and competing effectively in an AI-powered economy”.
The report surveyed more than 3,472 CIOs and technology leaders across 15 countries.
Salesforce struck a deal to acquire AI customer service specialist Fin for around $3.6 billion, a move positioned as bolstering its ability to help companies quickly deploy agents across communications channels.
Fin’s main product is a customer support AI agent which uses its proprietary Apex model. It is marketed as resolving queries across numerous channels including live chat, email, WhatsApp, SMS, Slack and over the phone.
Salesforce plans to use Fin’s offering to compliment its Agentforce platform.
It noted the combination of assets would provide fast “time-to-value options” for businesses, citing SMBs and those which “need to launch quickly, integrate with existing systems and deliver measurable outcomes” as likely customers.
The acquisition also brings Salesforce access to Fin’s experienced team and an “established global customer base of more than 30,000 companies”.
Salesforce chair and CEO Marc Benioff said Fin brings “proven agent technology, a deep commitment to customer success and an incredible AI team that will complement Agentforce with powerful service agent capabilities”.
Fin co-founder and CEO Eoghan McCabe enthused the deal as a “major win for consumers of the world”, adding the combination with Salesforce would enable it to deploy its products “far and wide at a rate far faster than we could have ever achieved on our own.”
The deal is subject to regulatory approval and other closing conditions including potential customary price adjustments. It is expected to close in Salesforce’s Q4 2027 (the three months to end-January 2027).
Num momento em que quase todos reconhecem a cloud como peça central das estratégias digitais, um novo estudo aponta uma contradição relevante: o investimento e a ambição em torno da IA estão a crescer mais depressa do que a maturidade tecnológica necessária para os suportar.
SpaceX reportedly leased the full capacity of its Colossus 1 data centre to Anthropic after running into technical problems using the site for its own Grok AI workloads.
Bloomberg reported last week (12 June) the company planned to use Colossus 1, located in the US state of Tennessee, as part of a three-campus cluster to train its more advanced AI models, but latency issues emerged when the facility was linked with two other sites more than 10 miles away
Sources told the news site the problems were said to be worsened by older network infrastructure.
Bloomberg noted high-performance AI training depends on fast, low-latency links between compute sites, but bottlenecks can slow the wider cluster.
Rather than continue trying to resolve those constraints, SpaceX decided leasing the site to third parties would generate greater value and create an additional revenue stream.
The move comes as SpaceX pushes its data centre expansion as a key plank of its investor story following its acquisition of xAI earlier this year.
A Ingram Micro reorganiza a sua operação local, que passa a ter acesso direto às capacidades da estrutura ibérica, com foco em cibersegurança, cloud, IA e soluções para PME portuguesas.
Mistral AI is reportedly in early talks to raise around €3 billion, in a move which could value the French start-up at roughly €20 billion.
Sources told Bloomberg the terms of the proposed raise may still change, and the valuation could also move higher depending on investor demand.
The rumoured round would strengthen Mistral AI’s position as Europe’s main AI challenger by providing the company with more capital as AI players face rising costs for computing power, infrastructure and talent across the globe.
It would also mark a sharp jump from the €11.7 billion valuation Mistral AI secured in September 2025 when Dutch chipmaker ASML became the largest shareholder in the start-up with an 11% stake after leading a €1.7 billion Series C funding round.
The start-up has positioned itself as a sovereign alternative to US and Chinese AI players, targeting European governments and companies with plans to build out cloud computing facilities in France and Sweden.
Mistral AI recently also moved to target industrial customers including Airbus and BMW for services covering engineering and manufacturing.
The French player acquired Austrian start-up Emmi AI last month as part of efforts to expand its services to industrial customers in Europe.
However, the company still trails larger rivals as its models and chatbot have still not matched the adoption of offerings from US players OpenAI and Anthropic, as well as Chinese competitors.
Last month, Anthropic raised $65 billion in a funding round valuing it at $965 billion, eclipsing OpenAI’s $852 billion valuation following its latest funding round in March.
Mobile World Live brings you our top three picks of the week as Apple reset its AI ambitions at WWDC, OpenAI edged closer to a stock market debut and French operators agreed a €20.4 billion deal to carve-up SFR.
Apple lands Siri AI at WWDC as Cook takes a bow
What happened: During its annual WWDC event, Apple reset its AI strategy, unveiled a revamped AI-powered Siri and spotlighted new child-safety controls as Tim Cook delivered his final keynote as CEO.
Why it matters: Apple SVP of software Craig Federighi said, “truly helpful AI must be centred around you and your needs”, positioning privacy, personal context and deeper app integration as core pillars of the AI-enhanced Siri. The launch gives Apple a chance to overwrite a damaging run of Siri AI delays described internally as “ugly and embarrassing”. Ben Wood, chief analyst at CCS Insight, said “Apple had to address its shortcomings in AI, and WWDC provided some answers”, but added the company must deliver “a meaningfully better everyday experience, not just parity with rivals”. He noted many features will already be familiar to Android and third-party AI app users, adding iPhone users are “more likely to upgrade for better battery life, an improved camera or a bigger screen rather than to get AI”. However, challenges have already emerged. The AI-upgraded Siri and Apple Intelligence features have already been delayed indefinitely in the European Union (EU) over competition laws and blocked in China over local AI regulations.
On child safety, Wood called Apple’s push “a welcome and timely move” as concerns about online harm grow among regulators but said “stronger cross-industry safeguards need to be built into operating systems at a deeper platform level”. Cook’s farewell added weight to the event as he declared the iPhone will “continue to be the centre of people’s digital lives”, adding, “the best is still ahead”.
What happened: OpenAI confidentially filed paperwork for a US IPO, becoming the latest AI company to move towards a listing as the sector’s biggest players race to raise fresh funds.
Why it matters: The ChatGPT-maker stated it had submitted a draft registration statement to the US Securities and Exchange Commission (SEC), adding, “we expect it to leak so we’re just announcing it”. It did not disclose the size, price or timing of the listing and cautioned a debut may not be imminent “because there are things we want to do that are likely easier as a private company”. OpenAI explained the decision involves “a complicated set of trade-offs”, but the filing gives it “the option to go public sooner if that ends up being best”. Reuters reported the company is targeting a valuation of up to $1 trillion, with a debut possible as early as September. The filing puts OpenAI into a crowded and expensive AI listings race; rival Anthropic confidentially filed for its own US IPO last week, while SpaceX launched a blockbuster listing. Aviva Investors’ head of multi-asset Sunil Krishnan told the BBC all firms all have a “vast need for cash” as “no one wants to be last” in a space where access to capital could prove crucial as companies fund AI chips, infrastructure and model training. Harrison Rolfes, senior research analyst at PitchBook, argued OpenAI’s $1 trillion listing is “a bet on a company that has never been profitable, in a market it is currently losing, with a cost structure it cannot change for at least another year”. He added this is “all while racing a direct competitor through the same SEC process in the same quarter.” Rolfes concluded: “There is no scenario where OpenAI pricing after a profitable Anthropic favours OpenAI.”
What happened: Bouygues Telecom, Free-Iliad Group and Orange signed a memorandum of understanding with Altice France to acquire and carve up SFR in a deal valuing the operator at €20.4 billion, setting the stage for a major reset of France’s telecoms market.
Why it matters: Under terms outlined in April, Bouygues will take 42% of SFR’s assets, Free-Iliad 31% and Orange 27%. SFR’s consumer mobile and fixed broadband operations will be divided between the three operators, while Bouygues will take control of the B2B unit. Completion is targeted for H2 2027, subject to regulatory clearance. Kester Mann, director, consumer and connectivity at CCS Insight, noted the move marks the biggest potential shake-up in French telecoms since Iliad entered the mobile market in 2012, cutting the number of major operators from four to three and reshaping competition. He said the agreement appears “a successful outcome for all parties” as Bouygues, Orange and Iliad all gain “important new assets in their pursuit of greater scale”, while removing a key rival will “reduce the competitive intensity of the market”. However, the biggest challenge will be to regulators show the deal will deliver “positive outcomes to the French market”. He noted that while that would once have been “a herculean task”, recent approvals in the UK and Spain suggest Europe’s stance on consolidation is softening.
Google reportedly is in discussions with Samsung Electronics to manufacture part of its next-generation AI processor, which could include using the latter’s 2-nanometer production technology.
The Information reported Google plans for TSMC to produce the main computing component of its tensor processing unit (TPU), codenamed Icefish, while Samsung may manufacture a component which helps connect it to memory.
Reuters reported the tech giant is also working with chip designer MediaTek on the overall design. Icefish is still in development, with mass production possible as early as 2028.
Landing the deal would be a major boost for Samsung’s foundry ambitions, strengthening an effort to win more advanced chip manufacturing business.
Its 2nm process would allow more transistors to be packed into smaller chips, improving performance, energy efficiency and AI capabilities, Reuters noted. It added the South Korean vendor is also considering a second Texas plant to expand output.
The reports are the latest indication Google is looking to reduce its reliance on TSMC, which is struggling to keep pace with surging AI demand and could prove a bottleneck for the wider industry.
Earlier this week, The Information also reported Google is in talks with Intel to manufacture more than three million TPUs in 2028.
O próximo processador de IA da Google terá uma arquitetura de fabrico dividida. A TSMC deverá produzir o núcleo computacional, cabendo à Samsung o fabrico do componente de interligação à memória. A Intel poderá também entrar na equação com a produção da TPU.
IT provider Tata Consultancy Services (TCS) teamed with Anthropic to accelerate enterprise AI adoption, with a focus on highly regulated sectors where accuracy, auditability and governance requirements have historically slowed deployments.
As a Global Premier Partner in the Claude Partner Network, TCS will establish a dedicated business unit focused on delivering industry services and AI expertise built on Anthropic’s Claude family of models, with early access to new releases.
It will also provide Anthropic’s Claude AI assistant to more than 50,000 employees across engineering, finance, legal, marketing, and sales.
The two companies will jointly go to market with AI capabilities across financial services, public services, life sciences, healthcare, aviation, telecom, and medtech.
The partnership also spans several TCS businesses and platforms. Diligenta, TCS’s UK-based life and pensions business with over 22 million customers, will use Claude to improve customer experience through agentic process transformation.
TCS iON, which conducts more than 75 million annual assessments across 1,500 cities in India, will deliver learning and certification programmes on Claude models to help build an AI-certified workforce.
K Krithivasan, CEO and managing director of TCS, stated enterprise AI value comes from understanding business context, orchestrating complex systems and applying deep AI engineering talent.
“By combining Claude with our industry expertise, engineering rigour and large-scale transformation capabilities, we will help customers move faster to production, especially in industries where trust, resilience and regulatory discipline are critical,” he said.
The partnership comes as Anthropic, now valued at $965 billion, looks to extend the momentum which has pushed its valuation above rival OpenAI, with both startups moving towards public listings.
Ericsson unveiled a software subscription designed to bring telco-grade AI models directly into basebands and radios to enable operators to boost 5G performance, automation and energy efficiency without adding new hardware.
According to the vendor, the software uses AI models designed to operate in real time within the RAN, supported by continuous learning and agentic AI capabilities for more advanced network operations. The company stated its telco-grade models are built for ultra-low latency inference, with a focus on reliability across diverse RAN environments.
The service works with Ericsson 5G Advanced across purpose-built and Cloud RAN platforms. Ericsson stated the software uses its silicon in radios and the latest generation of RAN compute to employ “the right AI model in the right part of the radio network”, while its AI capabilities can also be deployed across partner platforms.
The first features are now available, with more to follow later in the year. Initial functions include AI-powered scheduler for link adaptation, beamforming and multi-layer coordination.
Ericsson claimed the technology has been deployed in more than 15 trials worldwide, delivering up to 20% higher downlink throughput and up to 10% better spectral efficiency, alongside support for up to twice as many high-traffic users. It also said AI in RAN could achieve up to 95% coverage prediction accuracy.
Marten Lerner, head of networks strategy and product management at Ericsson, said the company is “redefining what’s possible in mobile networks by bringing powerful AI capabilities to service providers”, adding “we are taking a major step toward AI-native networks”.
Further comments from Ericsson partners SoftBank, Bell, SK Telecom and Rogers focused on the software’s potential benefits around real-time optimisation, network performance, energy savings, automation and support for emerging AI-driven services.
Joe Madden, principal analyst at Mobile Experts, added: “With a software upgrade, operators can squeeze more capacity, better observability, and more accurate location-based services out of the 5G network they bought years ago.”
AI platform disruptions rose sharply in early 2026 as growing enterprise adoption and heavier workloads exposed reliability issues across the full infrastructure stack, according to research from Ookla.
Analysing 471 days of US Downdetector data from 1 January 2025 to 16 April 2026 across ChatGPT, Claude, Gemini, Microsoft Copilot, AWS and Microsoft Azure, Ookla recorded 3.7 million user-reported problems.
High-signal disruption days, defined as when a service recorded more than 10 times its own median daily report volume, rose from six across four major AI apps in Q1 2025 to 51 in Q1 2026, according to the report by Ookla analyst Luke Kehoe.
Anthropic’s Claude model accounted for 39 of those 51 disruption days, making it the clearest example of scale-up volatility in the period. Gemini accounted for seven, Copilot three and ChatGPT two.
Claude recorded near-zero Downdetector reports in early 2025 before moving into a sustained report baseline from mid-July as adoption grew.
In Q1 2026, Claude generated 314,996 reports, while March volume alone was nearly three times February’s level. Ookla noted the pattern cannot be attributed to a single outage, with disruption clustered around demand surges, model-release windows and platform instability as Claude Code and Cowork usage scaled rapidly.
OpenAI’s ChatGPT produced the largest individual disruption signals in absolute terms, including roughly 68,000 reports on 2 December 2025, but its underlying reliability trend has improved.
Its monthly median daily report volume fell from a peak of 2,157 in April 2025 to 1,166 in April 2026, even as OpenAI reported more than 900 million weekly active users and rapid growth in Codex usage.
Google’s Gemini and Microsoft’s Copilot showed smaller but distinct patterns. Gemini’s high-signal disruption days rose from zero in Q1 2025 to seven in Q1 2026, consistent with rapid user growth.
Copilot’s outage pattern reflected its position inside Microsoft’s broader enterprise range, with far fewer reports on weekends, reflecting enterprise-aligned use.
Cloud infrastructure also featured prominently in the reliability picture. AWS’s 20 October 2025 DynamoDB DNS event generated more than 315,000 US reports, while Microsoft’s Azure Front Door incident on 29 October produced nearly 96,000, illustrating how failures in cloud control planes can cascade into AI platform disruptions.
Ookla concluded AI reliability has moved well beyond model serving, with failure points now spanning feature gates, GPU fleets, developer APIs, login systems and demand-management policies, all of which can appear to the end user as a single outage.
Global data centre capital expenditure is on course to exceed $1 trillion in 2026 according to new research from Dell’Oro Group, as hyperscale AI deployments accelerate and rising memory and storage costs push overall server spending higher.
The research company raised its worldwide data centre capex outlook for the year following a strong Q1, citing a combination of AI expansion, continued investment in general-purpose infrastructure and component cost inflation as the primary drivers.
Baron Fung, senior research director at Dell’Oro Group, stated rising memory and storage pricing substantially increased overall server system costs in the quarter and will likely remain a major capex growth factor throughout the year.
“At the same time, AI infrastructure deployments continue to accelerate rapidly, while hyperscalers also expanded general-purpose infrastructure to support public cloud growth, agentic AI workloads, and rising AI-related storage requirements,” he said.
The scale of spending among the largest cloud providers is striking. Amazon, Google, Meta, and Microsoft, the top four US hyperscalers, increased data centre capex by 78% year-on-year, reflecting the intensity of the AI infrastructure race.
On the supply side, Dell led server OEM revenue followed by Supermicro and Lenovo, though white-box vendors serving the hyperscale market accounted for most overall server revenue. Nearly all server vendors benefited from higher memory-driven system pricing.
Despite already exceptional spending growth in the first half of the year, Dell’Oro expects capex growth to accelerate further in the second half, driven by the ramp-up of Nvidia’s Vera Rubin platform and refresh cycles for hyperscaler custom accelerator platforms.
Beyond the major cloud providers, Fung noted select enterprise verticals and sovereign cloud providers are also increasing AI infrastructure adoption, “though growth remains constrained by uncertain returns and infrastructure readiness”.
“While near-term demand remains healthy, some spending may have been pulled forward ahead of expected price increases later this year,” explained Fung.
A Los Angeles-based startup has raised $5 million to fund an in-orbit computing demonstration next year, ahead of plans to deploy more than 100,000 orbital data centers to meet surging demand for AI infrastructure.
Meta Platforms expanded its partnership with Reliance Industries, agreeing to lease its first AI-enabled data centre in India as part of ambitions to grow its presence in the country.
Meta stated Reliance will build the facility in the Indian state of Gujarat with 168MW of capacity which Meta will use to “scale its AI infrastructure”.
The Facebook-owner explained it is investing aggressively to expand its capacity footprint, and India’s “tech-forward digital economy”, massive user base and the strength of its partnership with Reliance makes it the ideal location.
Reliance has identified Jamnagar, Gujarat as a strategic location, building out one of the largest data centre campuses in the world there, with access to the significant energy resources needed to power AI-enabled infrastructure.
The facility’s first phase will be to deliver 168MW of capacity, with options to scale.
Meta CEO Mark Zuckerberg said the facility will help the company scale its AI infrastructure globally, “while deepening our long-term investment in India’s economy”.
Chairman and managing director of Reliance, Mukesh Ambani, added that building India’s first built-to-suit AI data centre for a company of Meta’s scale “demonstrates India’s readiness to be at the forefront of the global AI revolution”.
The collaboration builds on a long-standing partnership between the pair, with Meta investing $5.7 billion in Jio Platforms in 2020.
Clean energy The site will be powered by renewable energy and cooled with desalinated seawater. Meta is set to cover the full cost of supporting the facility.
It has struck a deal for around 1GW of new clean and renewable energy, partnering with domestic companies CleanMax and Fourth Partner Energy.
Anthropic launched a public version of its Mythos AI model, but with guardrails in place to block its use in sensitive areas such as cybersecurity.
The AI player stated its Claude Fable 5 model is its most powerful to date on its launch yesterday (9 June), which is two months after Anthropic first unveiled its Mythos-class model.
The limited Mythos preview sent shockwaves through the industry after the model uncovered thousands of software vulnerabilities.
Last week Anthropic expanded the reach of its Mythos AI model to an additional 150 companies across more than 15 countries.
The startup describes Fable 5 as state-of-the-art on nearly all tested benchmarks of AI capability, claiming exceptional performance across software engineering, knowledge work, vision and scientific research. The longer and more complex the task, Anthropic stated, the larger Fable 5’s lead over its other models.
Anthropic stated it has done extensive testing to ensure users cannot manipulate Fable 5 into bypassing its guidelines. Queries on restricted topics will instead receive a response from the company’s Claude Opus 4.8 model.
The AI player acknowledged the safeguards are tuned conservatively and will sometimes catch harmless requests but said they trigger on average in fewer than 5% of sessions.
“With more capable models arriving in the coming months, we’re working to improve our safeguards and reduce false positives as quickly as we can,” Anthropic stated.
For a smaller group of cyber defenders and infrastructure providers, Anthropic is simultaneously launching Claude Mythos 5, which is the same underlying model as Fable 5 but with safeguards lifted in some areas.
Mythos 5 will initially be deployed through Project Glasswing in collaboration with the US government. It carries what Anthropic described as the strongest cybersecurity capabilities of any model in the world.
Users who had access to the Claude Mythos Preview will be able to upgrade to Mythos 5, with broader access planned through an expanded trusted-access programme.
Fable 5 and Mythos 5 are priced at $10 per million input tokens and $50 per million output tokens, less than half the price of Claude Mythos Preview.
The rollout comes as Anthropic, now valued at $965 billion, looks to extend the momentum which has pushed its valuation above rival OpenAI, with both startups racing toward public listings.
Login
