SpaceX appears to be preparing a significant upgrade to the home-networking side of its Starlink satellite-internet service, with regulatory filings revealing the company’s first router designed around the Wi-Fi 7 standard.

Documents filed with the US Federal Communications Commission identify a new Starlink router under the model number UTR-261. The device includes Wi-Fi 7 support, two Ethernet ports, USB-C connectivity and support for Power over Ethernet.

That might sound like a relatively small hardware update, but it highlights an important distinction in internet performance.

Starlink’s satellites determine how data travels between a customer and the wider internet.

The Wi-Fi router determines how that connection is distributed between devices inside a home, business or other location.

A fast satellite connection can therefore still feel slow or unstable if the local wireless network becomes a bottleneck.

Wi-Fi 7 is designed to improve wireless capacity, latency and efficiency compared with earlier Wi-Fi generations, particularly in environments where multiple high-bandwidth devices operate simultaneously.

For households, that could become increasingly relevant as people use 4K and higher-resolution streaming, cloud gaming, video conferencing, smart-home devices and large software downloads at the same time.

The router filing also provides evidence of how Starlink is gradually evolving from a satellite-dish service into a broader connectivity ecosystem.

SpaceX has expanded the Starlink platform across residential broadband, maritime connectivity, aviation, mobile applications and other use cases.

Improving the hardware inside customers’ homes is another part of making satellite broadband behave more like a mainstream fixed-internet product.

There is, however, an important limitation.

An FCC filing indicates a product is being prepared or assessed for regulatory purposes; it does not automatically establish a final commercial launch date, price or worldwide availability.

Customers should therefore distinguish between regulatory evidence of upcoming hardware and a formally announced retail launch.

A Wi-Fi 7 router will not make Starlink satellites themselves faster, but it could prevent the home network from becoming the weakest part of the connection.

That distinction matters as satellite broadband moves towards higher speeds and more demanding applications.

The UTR-261 filing suggests SpaceX is preparing Starlink for a future in which the quality of the last few metres of wireless connectivity becomes almost as important as the thousands of kilometres travelled through the satellite network.

Google has unveiled its Pixel 11 smartphone family, using the latest generation of its hardware to push artificial intelligence deeper into everyday phone functions.

The new lineup includes the Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL and Pixel 11 Pro Fold, alongside the Pixel Watch 5 and Google’s first dedicated tracking accessory, the Pixel Tag.

The Pixel 11 series runs on Google’s new Tensor G6 processor and introduces upgrades across cameras, battery life, storage and AI-assisted software.

Google says the phones begin with 256GB of storage, while the new generation also targets approximately 30 hours of battery life and faster wireless charging.

AI remains the central differentiator.

Features include improved speech-to-text capabilities, live translation for media, AI-assisted photography and deeper integration with Google’s Gemini ecosystem.

A new Magic Capture camera mode can automatically capture a sequence of photos and video, while Pro models receive additional camera and visual features.

The broader strategic significance goes beyond individual phone specifications.

Google controls Android, one of the world’s dominant mobile operating systems, and develops Gemini, one of the leading families of generative-AI models.

Pixel therefore acts partly as a showcase for how Google believes AI should be integrated into future Android devices.

That increasingly places smartphone competition on a different axis.

For years, manufacturers competed primarily through camera sensors, screens, processor speed and industrial design.

Those specifications still matter, but companies are now attempting to differentiate their devices through what AI can do with the hardware.

The Pixel lineup also faces difficult competition.

Apple and Samsung possess significantly larger smartphone-market positions, while Chinese brands compete aggressively on specifications and pricing, particularly in markets such as India.

Google must therefore convince consumers that its AI software provides benefits meaningful enough to justify upgrading.

The Pixel 11 launch shows that the smartphone industry is entering a stage where AI capability is becoming as important to marketing as cameras and processors.

But more AI features do not automatically create a better phone.

Google’s challenge will be demonstrating that Gemini-powered functions save meaningful time, improve communication or create genuinely useful experiences rather than simply adding another collection of features users try once and forget.

That distinction could determine which companies benefit most as the smartphone evolves into an increasingly AI-driven personal computing device.

China’s humanoid-robot industry is approaching a major commercial milestone as Unitree prepares for its Shanghai stock-market debut following extraordinary investor demand for its initial public offering.

Unitree’s retail IPO allocation was subscribed more than 8,000 times, a record level of demand for Shanghai’s STAR Market, according to Reuters. The company raised approximately 6.1 billion yuan, or about $905 million, while selling around 10% of its enlarged share capital.

The company has become internationally recognised for robots capable of walking, running, dancing and performing complex movements.

Its products include both four-legged robotic platforms and humanoid machines such as the G1, H1 and R1.

Unitree is particularly significant because it is not merely a robotics research laboratory.

The company is already generating revenue and profits, distinguishing it from many humanoid-robot developers still operating primarily as research-heavy startups. Reuters describes it as the world’s largest humanoid-robot maker by sales and the first profitable general-purpose robotics company preparing to list on mainland China.

Investor enthusiasm reflects a much larger technological bet.

Many companies believe humanoid robots could eventually perform work in factories, warehouses, logistics, inspection, healthcare, hospitality and other environments designed around the human body.

If robots become sufficiently reliable and inexpensive, businesses could potentially automate categories of physical labour that traditional industrial robots cannot easily perform.

But this remains an emerging market.

Impressive demonstrations of robots running or performing martial arts do not prove that the same machines can complete economically valuable tasks reliably for thousands of hours in real workplaces.

Unitree’s valuation therefore embeds substantial expectations about a market whose final scale is still uncertain.

The company also faces intensifying competition from Chinese robotics firms and international players including Tesla and Boston Dynamics. Trade restrictions between the United States and China introduce another source of uncertainty.

Unitree’s IPO demonstrates that investors increasingly view humanoid robotics as a potential next major technology industry rather than a laboratory experiment.

Yet there is a gap between technological spectacle and commercial usefulness.

The company’s long-term value will ultimately depend not on how dramatically its robots move in demonstrations, but on whether they can perform useful work safely, reliably and at a cost that makes economic sense.

The extraordinary IPO demand proves investor enthusiasm. It does not yet prove that humanoid robots will achieve mass adoption.

Chinese artificial-intelligence company Z.ai has unveiled GLM-5.3, an AI model that the company says has demonstrated unexpectedly strong capabilities in identifying software vulnerabilities and performing cybersecurity-related tasks.

The development matters because AI is increasingly becoming both a defensive cybersecurity tool and a potential offensive capability.

According to benchmark results reported by Reuters, GLM-5.3 achieved 84.5% on CyberGym, a vulnerability-identification benchmark, compared with 83.8% for Anthropic’s restricted-access Mythos 5 model.

That headline result requires context.

GLM-5.3 did not outperform Anthropic’s model across every cybersecurity task.

On an exploit-development benchmark, Z.ai’s model scored 54.4% compared with 78% for Mythos 5, indicating that strong vulnerability detection does not automatically translate into equally strong exploit-generation capability.

The distinction is important because AI cybersecurity capabilities exist on a spectrum.

A model that can inspect software and identify vulnerabilities can help developers discover weaknesses before attackers exploit them.

The same underlying capability, however, can potentially be adapted to identify vulnerable systems or automate elements of cyberattacks.

That dual-use problem explains why advanced AI laboratories are increasingly placing security controls around powerful cyber capabilities.

Z.ai plans to release GLM-5.3 more broadly after completing additional safety assessments, while advanced functionality is expected to be subject to a verified or trusted-access mechanism.

The development is also another indication that competition in advanced AI is becoming more internationally distributed.

American companies such as OpenAI, Anthropic, Google and Meta remain major players, but Chinese laboratories including Z.ai are increasingly producing models competitive on selected benchmarks.

Still, benchmark comparisons need caution.

Performance in controlled testing environments does not guarantee equivalent performance against complex real-world software systems, and results published by model developers should ideally be independently replicated.

GLM-5.3 highlights one of the most consequential directions in artificial intelligence: AI systems are becoming increasingly capable of finding weaknesses in other digital systems.

That could dramatically improve software security if defenders gain the advantage. But if offensive capabilities advance equally quickly, cybersecurity may evolve into an environment where AI systems continuously search for—and attempt to repair or exploit—vulnerabilities at machine speed.

For that reason, the important story is not simply whether Z.ai beat another model on one benchmark. It is how safely increasingly capable cyber-focused AI systems can be deployed.

India is accelerating its attempt to build a domestic semiconductor ecosystem, with Prime Minister Narendra Modi saying production has begun at three semiconductor facilities and another seven to eight plants are expected to become operational in the coming years.

The development represents an important transition for India’s semiconductor programme.

For several years, the country’s strategy was largely measured through investment announcements, approved projects and proposed factories. The policy is now gradually moving towards the more difficult stage: converting those commitments into actual semiconductor production.

India has also substantially expanded government support for the sector.

The second phase of the India Semiconductor Mission received an outlay of approximately ₹1.275 lakh crore, following the original ₹76,000 crore programme. Semicon 2.0 is intended to deepen capabilities across semiconductor equipment, materials, intellectual property, fabrication, packaging, testing and supply chains.

Why are semiconductors receiving this level of attention?

Because chips are no longer simply an electronics-industry component.

They are essential to smartphones, automobiles, defence systems, telecommunications networks, artificial-intelligence infrastructure, industrial machinery, medical equipment and almost every major digital system.

The global chip shortages experienced earlier in the decade demonstrated how disruption in a small number of semiconductor-producing countries could affect manufacturing around the world.

India therefore sees domestic chip capacity as both an economic opportunity and a strategic-security issue.

The country already possesses significant semiconductor design talent, but advanced fabrication and the surrounding manufacturing ecosystem have historically been concentrated in markets such as Taiwan, South Korea, the United States, Japan and China.

Building fabs is also unusually difficult.

A semiconductor plant can require billions of dollars in investment, extremely sophisticated manufacturing equipment, reliable electricity and water, specialised materials, skilled workers and years of process refinement.

That means counting announced factories alone does not establish whether India has become a global semiconductor power.

India’s semiconductor strategy is moving from policy ambition toward industrial execution.

The critical next indicators are not how many plants are announced, but whether those facilities begin reliable commercial production, attract suppliers, develop domestic expertise and compete successfully in international markets.

If that happens, India could gradually move from being primarily a major electronics consumer and design centre into a more important part of the global semiconductor manufacturing chain.

India has announced one of its largest artificial-intelligence skilling initiatives yet, with Prime Minister Narendra Modi setting a target to train one crore young Indians in AI skills over the next year.

The initiative was announced during the Prime Minister’s Independence Day address on August 15 and is intended to prepare young people for an economy in which artificial intelligence is increasingly affecting software, manufacturing, finance, healthcare, public administration and other industries. The government says the programme is designed to help Indian youth develop the capabilities required to participate in the emerging AI economy.

The announcement is significant because India’s AI challenge is no longer limited to gaining access to powerful models.

The larger issue is whether enough people can understand, develop and apply AI productively.

A country may have access to advanced computing systems, but without engineers, developers, researchers, domain specialists and ordinary workers who understand how to use them effectively, the economic impact remains limited.

India has already been expanding its AI infrastructure. Government figures indicate that the IndiaAI ecosystem includes a shared computing facility with more than 45,000 GPUs, while AI initiatives have produced dozens of prototypes and deployed solutions across public-sector organisations.

Training one crore people, however, creates an additional challenge: scale must not replace quality.

A short introductory AI course and deep technical expertise are not the same thing. Effective implementation will require different learning pathways for school and college students, software engineers, working professionals, entrepreneurs and people in non-technical occupations.

The programme will also need to go beyond prompt-writing.

Useful AI education should include data literacy, problem-solving, responsible AI use, cybersecurity awareness, critical evaluation of AI-generated information and, for technical learners, deeper competencies in model development and deployment.

The initiative could also help broaden AI adoption beyond India’s major technology centres if courses are made accessible in regional languages and designed for learners with different educational backgrounds.

Training one crore Indians in AI within one year is an ambitious numerical target. But the more important measurement will be what those people are actually capable of doing after the training.

If India converts large-scale AI awareness into real employable skills, productivity gains and innovation, the initiative could strengthen its position in the global AI economy. If the programme focuses primarily on completion certificates, its headline scale may significantly overstate its practical impact.