Thinking Machines Lab unveiled its first AI model more than a year after former OpenAI executive Mira Murati (pictured) founded the start-up, moving it from research previews to a full launch.

The Inkling model is designed to reason natively across text, images and audio, while balancing cost against performance through variable, adjustable effort.

Thinking Machines stated the model is trained to exhibit safe behaviour across all modalities.

The company noted Inkling is a mixture-of-experts transformer with 975 billion total parameters and 41 billion active parameters, supporting a context window of up to 1 million tokens.

Thinking Machines described it as the first in a family of models of varying sizes.

Rather than chasing state-of-the-art performance in a single domain, Thinking Machines said it trained Inkling for solid capabilities across agentic tasks, reasoning, coding, instruction-following, factuality, vision and audio, positioning it as a foundation model for future releases.

The company acknowledged Inkling is not the most performant model available today, closed or open.

Inkling is open-weight, meaning developers can download and customise the model without access to training data or source code.

It is available for fine-tuning on Tinker, Thinking Machines’ developer platform, alongside a so-called Inkling Playground for testing the model directly.

Bloomberg reported Thinking Machines is not aiming to monetise the model itself; the company currently generates revenue through Tinker, which it sells to customers including hedge fund Bridgewater Associates for financial-task fine-tuning.

The news outlet noted Murati’s company raised $2 billion at a $12 billion valuation last year and was reportedly in talks over a larger subsequent round, though it added several employees have since departed for firms including Meta Platforms and OpenAI.

Bloomberg also reported the launch could help fill a gap in the US open model market, as Meta shifts focus on proprietary offerings and rising AI spending pushes some users toward freely available Chinese models.

Thinking Machines was founded in early 2025 by Murati and a host of other former OpenAI employees, with the company initially focused on collaboration between the technology and people.

The post Thinking Machines releases first AI model appeared first on Mobile World Live.

Anthropic is reportedly seeking to meet with investors ahead of a potential mega-IPO, as the company prepares to join a wave of AI-driven stock market debuts.

Bloomberg reported the banks leading the offering are scheduling meetings between investors and the Claude chatbot maker in the coming weeks, citing people familiar with the matter. The news outlet previously stated Anthropic is considering an IPO as soon as October.

Deliberations remain ongoing and details including timing could still change, Bloomberg stated.

Moving ahead on the timeline would put Anthropic ahead of rival OpenAI, which is now eyeing a debut in 2027 after earlier targeting 2026.

The news site noted an IPO in October would also see Anthropic list before DeepSeek, the Chinese AI company which has been steadily gaining market share.

DeepSeek is reportedly preparing its own IPO and could file this year.

Bloomberg stated Anthropic is working with Morgan Stanley, Goldman Sachs and JPMorgan Chase on its IPO.

The AI player is also navigating uncertainty tied to its relationship with the Trump administration, which briefly imposed foreign access restrictions on two of Anthropic’s top models.

Anthropic separately sued the US Department of War after it declared the company posed a risk to the nation’s supply chain.

The company was valued at $965 billion following a funding round in May, making it one of the world’s largest private companies and pushing its valuation past OpenAI’s for the first time.

The post Anthropic plans investor meetings as IPO looms appeared first on Mobile World Live.

Bell Canada completed construction of its first sovereign direct-to-device (D2D) satellite ground station, a milestone in a strategy to offer connectivity beyond the reach of its traditional wireless network.

The facility in the province of Quebec connects with AST SpaceMobile satellites, with testing underway as the operator works to build the infrastructure required to provide D2D services.

Bell stated the companies completed integration testing covering “text messaging, broadband data connectivity”, and voice and video calls.

Video streaming, IoT and public alert tests are scheduled for the coming weeks.

The milestone builds on demonstrations Bell and AST SpaceMobile ran in 2025.

In addition to connecting customers in rural or remote locations, Bell stated the service is expected to support use cases including emergency response, environmental monitoring, resource development, energy production and transportation.

The company is continuing to build ground station infrastructure elsewhere in Canada, with additional sites under construction in the provinces of Ontario, Alberta, and Newfoundland and Labrador.

Bell explained the facilities would provide nationwide D2D coverage while ensuring customer traffic runs through domestic infrastructure and data remains sovereign.

EVP and CTO Mark McDonald said the ground infrastructure “is designed to support a full suite of D2D capabilities including messaging, data connectivity, voice calling and video streaming”.

AST SpaceMobile CCO Chris Ivory pointed to Canada’s geography as a driver for the project, explaining the work with Bell would provide seamless connectivity “beyond the reach of traditional terrestrial networks”.

The post Bell edges closer to D2D service launch appeared first on Mobile World Live.

Apple’s on-device generative AI system reportedly received the green light from Chinese regulators, clearing the way for its long-delayed debut in the world’s largest smartphone market.

Reuters reported China’s cyberspace regulator stated Apple’s on-device generative AI service, Apple Intelligence, has been registered for use on iPhones in China.

The news outlet stated Apple Intelligence will tap into capabilities from AI models developed by Baidu and Alibaba, citing a source familiar with the matter.

Alibaba confirmed to Reuters its Qwen model will be integrated into Apple Intelligence across iPhone, iPad, Mac and Vision Pro in China.

A Baidu representative confirmed the company is also working with Apple on AI features for Chinese iPhone users.

Bloomberg reported the Cyberspace Administration of China included Apple’s generative AI services on a list of newly approved providers, alongside offerings from local companies like Huawei and Xiaomi.

Neither the regulator nor Apple has given a launch date.

The approval gives Apple greater access into a Chinese population eager to experiment with new AI tools, potentially boosting the company’s recent recovery in the market as it attempts to narrow the gap with domestic AI leader Huawei.

Reuters also reported ZTE’s Nubia-Doubao phone, built using ByteDance’s Doubao AI, was registered by the regulator in the same batch.

The post China clears Apple Intelligence for launch appeared first on Mobile World Live.

Google DeepMind CEO Demis Hassabis called for the establishment of a new US-led regulatory framework to govern frontier AI, proposing a self-regulatory structure which could bypass political opposition.

In a post on X, Hassabis warned the breakneck pace of commercial and geopolitical competition is currently outstripping the industry’s understanding of the technology.

He noted the arrival of Artificial General Intelligence (AGI) is likely only a few short years away.

To address looming technical, biological, and cybersecurity risks without stifling innovation, Hassabis proposed creating a new US standards body.

He modelled it after a self-regulatory organisation like the Financial Industry Regulatory Authority (FINRA), rather than a traditional government agency.

The structural choice comes amid heavy resistance to AI regulation from the tech industry and shifting policies from the Trump administration.

The FINRA-style blueprint would be staffed by leading technical experts and open-source representatives but primarily funded by the industry to secure the massive compute resources needed for testing.

TechCrunch reported the proposed system would replace the current, highly criticised ad hoc government reviews previously used for models like Anthropic’s Mythos and OpenAI’s Sol.

Those reviews drew fire from industry onlookers for a distinct lack of technical expertise and opaque decision-making, according to the news outlet.

Under Hassabis’s framework, the independent body would define thresholds to classify models as Frontier-Class. Organisations behind them, dubbed Frontier Labs, would eventually be required to pass the independent assessments to deploy in the US market.

Hassabis stated assessments would check for advanced threats, deception, and guardrail bypass attempts. The standards body could also outsource specific evaluations to a growing ecosystem of independent, third-party AI safety groups.

He argued the approach is designed to keep up with AI’s acceleration, allowing the industry to safely steer toward an era of scientific abundance while addressing the Trump administration’s concern over state-mandated red tape.

The post DeepMind CEO seeks US standards body for frontier AI appeared first on Mobile World Live.

New York made history by becoming the first US state to halt the construction of massive new data centres, imposing a one-year moratorium as concerns escalate over the environmental and financial toll of the AI boom.

The moratorium, outlined in an executive order signed by Governor Kathy Hochul (pictured), targets hyperscale operations by halting discretionary environmental permits for new facilities requiring 50 megawatts or more of power.

The rapid expansion of facilities housing thousands of computer servers has triggered intense community and political backlash across the US.

“This pause will remain in place for up to one year while New York establishes the strongest possible framework to protect our community’s guardrails to reduce the risk to our energy grid, minimize land disruption, noise pollution, and protect our natural resources, especially our water supply,” Hochul stated during a press conference. “We’re going to explore having hyperscale data centres pay into a larger fund to support our grid statewide.

“We need to invest in this, help defray costs, reduce bills for our rate payers,” she added. “We expect this process, which we already launched, to be completed within the year.

Once the policy is in place, the freeze will be reviewed and lifted, Hochul stated.

She also said the data centres should only be built in areas which want them and “they will never be exempt from local zoning, local approvals”.

Local populations are increasingly concerned the energy-hungry centres by companies including Alphabet, Microsoft, Meta, Amazon, and Oracle are inflating residential utility bills, straining electricity grids and consuming millions of gallons of water for cooling.

Hochul framed the aggressive intervention as a necessary protective measure for state residents, stating as data centre development threatens to hike utility bills and deplete natural resources, it is her responsibility to act.

She noted labour organisations will have a seat at the AI data centre discussions “so projects can include wage standards and labour agreements and prioritise local hiring”.

To further rein in the sector, Hochul announced plans to pursue legislation to repeal lucrative sales tax exemptions currently enjoyed by hyperscale data centre operators.

“New Yorkers work hard and, on my watch, their tax dollars will not be used to benefit these large firms, these centres.”

Reuters reported Hochul directed state officials to develop a Generic Environmental Impact Statement. The comprehensive assessment will establish consistent operating and environmental standards for new facilities while examining the overall footprint of data centre construction and operation across the state.

The construction ban will be lifted once the rigorous regulatory guidelines are finalised.

The news outlet reported state lawmakers recently passed a bill targeting data centres consuming 20 megawatts or more of power, representing a wider scope than the governor’s executive order.

The post New York puts mega data centres on ice appeared first on Mobile World Live.

Infrastructure provider Zayo revealed a planned CEO transition with former Verizon executive Sowmyanarayan Sampath (pictured) appointed as chief following the eventual retirement of Steve Smith.

Smith remains CEO until 31 August when he will retire. Sampath will take the reins on 1 September while Smith will remain on Zayo’s board following the handover.

During the transition period, Smith and Sampath will work closely with Zayo’s board and leadership team to ensure continuity for customers and stakeholders.

The company positions the leadership transition as a key element for its next phase of AI-driven growth.

Sampath brings more than two decades of large-scale digital infrastructure experience to his new role, most recently as CEO of Verizon Consumer Group.

After joining the company in 2014, he was viewed as former Verizon CEO Hans Vestberg’s likely successor after holding several key executive positions. Last year the mobile operator’s board of directors picked former PayPal chief Schulman to replace Vestberg to boost the company’s finances and subscribers.

Prior to his last position, he served as the CEO of Verizon Business starting in 2022.

He was also Chief Product Officer for the Consumer and Enterprise businesses and CFO at Verizon Media Group.

Sampath was president of Global Enterprise, where he spearheaded the commercial launch of Verizon’s 5G networks and multi-access edge computing (MEC) solutions.

He stated Zayo’s physical network footprint arrives at a moment when the global economy demands significant bandwidth capacity.

“As AI workloads, distributed applications, dense GPU clusters, and hyperscale environments change where capacity is needed, and as large enterprises manage increasingly complex connectivity needs, the network is the ultimate constraint, and Zayo is built to solve this,” he said, noting the company already connects more data centres in the US than any other provider.

During Smith’s tenure, Zayo more than tripled its planned fibre expansion, with projects now spanning more than 15,000 route miles, while advancing automation, internal AI adoption and network service improvements.

He also oversaw the acquisition of Crown Castle’s fibre and small cell assets, adding approximately 90,000 route miles and 40,000 on-net enterprise locations to Zayo’s network.

“Leading Zayo has been a privilege, and I’m deeply proud of what our team has built as we prepare the Company for the next generation of digital infrastructure,” Smith stated, adding he is confident Sampath is the right leader for Zayo’s next chapter and looks forward to supporting him from the board.

The post Zayo names former Verizon exec Sampath as next CEO appeared first on Mobile World Live.

Intel outlined a €5 billion capital investment in its Republic of Ireland campus to expand ‌European output to meet the increased demand for AI and high-performance computing.

The chipmaker stated the expansion at its facility in Leixlip, County Kildare, began earlier in the year and involves upgrading fabrication facilities and installing leading-edge manufacturing equipment to boost production of its Intel Xeon 6 and next-generation Xeon processors built on the advanced Intel 3 process node.

Intel stated the chips are increasingly critical to power AI factories.

As part of the infrastructure overhaul, Intel is expanding its “automated track system to integrate disparate campus modules” into a single, “high-velocity production environment”.

Intel explained the investment programme would create permanent high-tech jobs alongside engaging specialised building and equipment installation tradespeople.

Naga Chandrasekaran, EVP and GM of Intel Foundry, said the investment is a “definitive commitment to maximising capacity” for foundry customers.

Upgrading the site “with state-of-the-art technology and installing cutting-edge tools” would help keep the Republic of Ireland “at the forefront of the world’s most advance manufacturing ecosystems while strengthening” Europe’s position in “the global technology landscape”, he said.

The move aligns with broader European Union tech-sovereignty goals to establish a resilient, domestic supply chain for leading-edge processors.

Intel invested more than €30 billion in the Republic of Ireland since establishing operations in 1989. The Leixlip site currently employs 4,900 people.

The company is trying to revitalise its chip manufacturing business and become a serious player in the booming foundry industry.

In April, it signed up to Elon Musk’s Terafab AI chip project to create processors for orbital data centres and robots.

The post Intel pumps €5B into EU AI fab capacity appeared first on Mobile World Live.