Trials in the United States and Europe are testing whether embedding wireless charging systems beneath road surfaces can support electric vehicles while driving and reduce reliance on plug-in infrastructure.

Why Wireless Charging Roads?

Wireless charging roads, often described as electric roads or dynamic wireless charging systems, are intended to address one of the most persistent challenges in electric vehicle adoption, reliable and convenient access to charging. While public charging networks are expanding, concerns remain about availability, downtime and the time required to recharge vehicles, particularly for commercial fleets.

The underlying concept is that vehicles equipped with compatible receivers can collect energy while travelling, waiting in traffic, or stopping briefly. Tel Aviv-based Electreon, one of the leading companies developing this technology, describes its approach in practical terms as enabling vehicles to “charge while driving”, “charge while queuing” and “charge while parked”. The stated objective is to reduce operational disruption and allow more flexible energy management throughout the day.

How Inductive Charging Works

The systems currently being deployed in pilot projects rely on inductive wireless charging. This involves installing copper coils beneath the road surface and connecting them to the electricity grid. When a compatible vehicle drives over the embedded coils, energy transfers through a magnetic field to a receiver fitted underneath the vehicle, which then feeds the battery.

Electreon explains that its “wireless electric road technology is based on magnetic resonance induction, with copper coils installed under the roadway.” The coils are designed to activate only when an authorised vehicle passes above them, remaining powered off in their default state. This means that vehicles without compatible receivers, as well as pedestrians or animals, do not trigger energy transfer.

Can’t See It From The Outside

The design of the system means the road surface itself appears unchanged because the charging components are installed beneath the asphalt, and control units positioned at the roadside manage the power supply and system monitoring. According to the company, the infrastructure is intended to operate discreetly within standard road construction and maintenance frameworks.

Where Public Trials Are Underway

Detroit is hosting the first publicly accessible wireless charging street in the United States. A quarter mile section of 14th Street in the Corktown district has been equipped with inductive coils beneath the road surface to enable dynamic charging for compatible electric vehicles. The project is linked to Michigan Central’s mobility innovation district and has involved the Michigan Department of Transportation, the City of Detroit, Ford Motor Company and DTE Energy.

The Detroit installation has been used to test performance using a Ford E-Transit shuttle vehicle known as Ellie. Publicly released test reports describe the operation of the dynamic charging system and its integration with static wireless charging points installed nearby. The project has been presented by state officials as part of a wider strategy to support electrified transport and long-term emissions reduction.

In Other Countries Too

Electreon has also implemented pilot projects in Israel, Sweden, Germany, Italy and France, often focusing on bus routes, freight corridors or controlled test tracks. These deployments are intended to assess durability, energy transfer efficiency, interoperability and system performance under real-world conditions.

The Business Case For Fleet Operators

Much of the early commercial focus has centred on public transport operators and freight fleets. For example, buses, delivery vehicles and heavy goods vehicles typically operate along predictable routes and for extended hours, which makes opportunity charging during normal operations more feasible.

Electreon is promoting a service model that allows operators to pay for access to the charging infrastructure rather than funding full installation themselves. In reporting on its commercial agreement with Dan Bus Company in Tel Aviv, the company stated that Dan would pay a monthly fee of 2,500 Israeli shekels per bus using the system, alongside electricity costs. The same project combined dynamic on-route charging with stationary wireless charging at a bus terminal.

Interestingly, in its Tel Aviv University Station case study, Electreon reported that on-route charging enabled a reduction in required battery size to 42 kilowatt hours from an original 400 kilowatt hours, describing this as “a nearly 90% reduction in size”. Such outcomes are specific to individual routes and operating patterns, yet they illustrate the potential argument that vehicles may not need to carry large batteries if energy can be collected frequently throughout the day.

Battery Size And Emissions Implications

The sustainability case for wireless charging roads extends beyond convenience. For example, smaller batteries can reduce the demand for raw materials and energy-intensive manufacturing processes, while lighter vehicles generally consume less energy per mile. If dynamic charging allows for reduced battery capacity without compromising operational range, the overall lifecycle emissions of vehicles could be affected.

Electreon also positions its technology as supporting carbon neutrality by enabling more efficient use of transport energy and lowering the need for extensive grid upgrades. The company states that the technology can reduce the need for large batteries and extensive grid connection capacity, contributing to flatter electricity demand profiles and potentially lower system costs.

Standards And Interoperability

For wireless charging roads to expand beyond isolated pilots, technical standards are essential. For example, the SAE J2954 standard addresses interoperability and electromagnetic compatibility for wireless power transfer in light and medium-duty vehicles. Internationally, the IEC 61980 series sets requirements for wireless power transfer systems for electric road vehicles, including safety and system performance criteria.

Standardisation Necessary

Also, standardisation is needed to ensure that vehicles from different manufacturers can operate on shared infrastructure and that electromagnetic exposure remains within established limits. Without broad adoption of common standards, the risk of vendor-specific systems limiting scalability remains significant.

Costs And Practical Challenges

The cost of installing wireless charging infrastructure beneath roads remains a central concern. For example, the Detroit pilot has cited figures of close to two million US dollars per mile for current installations. While developers argue that costs could fall as deployment scales and installation processes mature, large-scale retrofitting of urban or intercity roads would represent a substantial capital commitment.

Maintenance and road lifecycle management present further challenges. For example, roads require resurfacing and periodic repair, and embedded infrastructure must be designed to withstand heavy traffic, weather variation and long-term wear. Electreon states that its systems have undergone stress and endurance testing to demonstrate that installation does not reduce road lifespan when implemented correctly, although long-term operational data at scale is still limited.

Safety A Concern

Safety and electromagnetic exposure are also important considerations for using this type of new technology on public roads. Electreon states that its system activates coils only when authorised vehicles are present and that it has been tested in accordance with international electromagnetic compatibility and safety standards. Independent technical reviews of wireless charging systems identify issues such as alignment, thermal management and electromagnetic exposure as areas requiring careful design and monitoring.

Strategic Role Within Wider Charging Infrastructure

Wireless charging roads are, therefore, not intended to be a replacement for conventional plug-in charging in all contexts. Instead, the approach is a complementary infrastructure for specific use cases, e.g., bus routes, logistics hubs and high-utilisation corridors where predictable movement patterns may support a stronger economic case.

There is clearly a long way to go with this idea before it can be rolled out at scale, and public authorities and industry stakeholders are still evaluating whether dynamic charging can meaningfully contribute to emissions reduction targets and fleet electrification goals. For now, the technology remains in a pilot and early commercial phase, with expansion dependent on cost reduction, standardisation, and evidence from ongoing real-world trials.

What Does This Mean For Your Organisation?

Wireless charging roads have moved beyond theory, but they remain limited to targeted pilots where the operational case is strongest. The technology has shown that vehicles can collect energy while moving, queuing or stopping, and that battery size and downtime may be reduced in certain use cases. Whether that translates into wide-scale deployment depends on cost reduction, robust standards and long-term maintenance performance.

For UK businesses, particularly fleet operators and logistics providers, this is a development to monitor rather than adopt immediately. Any future rollout would require coordination between highways authorities, energy providers and vehicle manufacturers, alongside clear economic and environmental evidence.

For policymakers, the central question is value for money. Wireless charging roads may complement conventional plug-in infrastructure on specific high-use corridors, yet they are unlikely to replace it. The next stage will be defined by data from real-world trials and by whether the sustainability benefits justify the infrastructure investment.

Copilot is embedded across the suite of Microsoft Office products and so this video explains how you can use it to turbo-charge Excel, thereby making full use of AI within your spreadsheets …

[Note – To Watch This Video without glitches/interruptions, It may be best to download it first]

Microsoft is retiring its free Lens scanning app, so now is the time to secure your documents and switch to a reliable alternative without losing functionality or control.

Microsoft Lens To Be Removed From iOS and Android App Stores

Microsoft has confirmed that Microsoft Lens will be removed from iOS and Android app stores and eventually disabled for creating new scans, although existing files will remain accessible if the app stays installed and you remain signed in.

Originally launched as Office Lens in 2015, it became widely used in business because it was free, simple and reliable. It allowed users to scan documents, whiteboards and business cards, convert them into searchable PDFs using OCR, and save them locally or to OneDrive, Word, PowerPoint and OneNote.

Microsoft is now directing users towards scanning within Microsoft OneDrive or Microsoft 365 Copilot. However, these do not fully replicate Lens. Features such as business card scanning to OneNote, read-aloud and Immersive Reader integration are not currently available.

For businesses, this means:

– Reviewing document workflows before scanning is disabled
– Exporting important scans before access becomes limited
– Deciding whether cloud-only storage via OneDrive is acceptable
– Considering alternative scanning apps for flexibility.

What to do now

– Export your existing scans
– Open Microsoft Lens
– Save each scan as a PDF
– Store a local copy
– Back up to your chosen cloud storage
– Confirm files open independently of the app.

Switch to OneDrive scanning

– Open the OneDrive app
– Tap the + button
– Select Scan
– Capture your document
– Save to your chosen folder.

Be aware that OneDrive does not support local-only storage.

Use a free alternative – here are some options (please note, these are suggestions and not recommendations).

Adobe Scan

– Download the app
– Sign in with a free account
– Scan documents using the camera
– Save as PDF
– Export to local or cloud storage

Genius Scan

– Install the app
– Capture documents
– Adjust borders
– Export as PDF
– Save locally or to your preferred cloud.

Google Drive (Android)

– Tap +
– Select Scan
– Save as PDF.

The key point is portability. Export everything as standard PDFs, keep copies outside any single app, and avoid relying on proprietary storage. That way, when Microsoft Lens closes, your documents and your workflow remain firmly under your control.

Recent research shows that AI large language models (LLMs) can be quietly poisoned during training with hidden backdoors that create a serious and hard to detect supply chain security risk for organisations deploying them.

Sleeper Agent Backdoors

Researchers say sleeper agent backdoors in LLMs pose a security risk to organisations deploying AI systems because they can be embedded during training and evade detection in routine testing. Recent studies from Microsoft and the adversarial machine learning community show that poisoned models can behave normally in production, yet produce unsafe or malicious outputs when a trigger appears, with the behaviour embedded in the model’s parameters rather than in visible software code.

Embedded Threat

Unlike conventional software vulnerabilities, sleeper agent backdoors are embedded directly in a model’s weights, the numerical parameters that encode what the system has learned during training, which makes them difficult to detect using standard security tools. Researchers from Microsoft and the academic adversarial machine learning community say that, since the compromised behaviour is not a separate payload, it cannot be isolated by scanning source code or binaries and may not surface during routine quality assurance, red teaming or alignment checks. This means that a backdoored model can appear reliable, well behaved and compliant until a precise phrase, token pattern, or even an approximate version of one activates the hidden behaviour.

The Nature Of The Threat

Researchers from Microsoft, building on earlier academic work in adversarial machine learning, say in recent studies that the core risk posed by sleeper agent backdoors is the way they undermine trust in the AI supply chain as organisations become increasingly dependent on third party models. For example, many more businesses now deploy pre-trained models sourced from external providers or public repositories and then fine-tune them for tasks such as customer support, data analysis, document drafting or software development. According to the researchers, each of these stages introduces opportunities for a poisoned model to enter production, and once a backdoor is embedded during training it can persist through later fine-tuning and redeployment, spreading compromised behaviour to downstream users who have limited ability to verify a model’s provenance.

The threat is difficult to manage because neither model size nor apparent sophistication guarantees safety, and because the economics of the LLM market strongly favour reuse. In a report entitled “The Trigger in the Haystack”, Microsoft researchers highlight how LLMs are “trained on massive text corpora scraped from the public internet”, which increases the opportunity for adversaries to influence training data, and warn that compromising “a single widely used model can affect many downstream users”. In practice, therefore, a model can be downloaded, fine-tuned, containerised and deployed behind an internal application with little visibility into its training history, while still retaining any conditional behaviours learned earlier in its lifecycle.

How The Threat Differs From Conventional Software Attacks

The most important distinction between sleeper agent backdoors and conventional malware is where the malicious logic resides and how it is activated. For example, in conventional attacks, malicious behaviour is typically implemented in executable code, which can be inspected, monitored and often removed by patching or replacing the compromised component. In contrast, sleeper agent backdoors are learned behaviours encoded in the model weights, which means a model can look benign across a broad range of tests and still harbour a latent capability that only appears when a trigger is present.

A ‘Poisoned’ Model Can Pass A Normal Evaluation Test

This difference places pressure on existing security assurance methods because conventional approaches often depend on knowing what to look for. Microsoft’s research paper describes the central difficulty in practical terms, stating that “backdoored models behave normally under almost all conditions”. That dynamic makes it possible for a poisoned model to pass a typical evaluation suite, then be deployed into environments where it can handle sensitive data, generate code, or influence decisions, with the backdoor remaining dormant until the trigger condition is met.

Industry Awareness And Preparedness

The gap between AI adoption and security maturity is a recurring theme in Microsoft’s “Adversarial Machine Learning, Industry Perspectives” report, which draws on interviews with 28 organisations. The paper reports that most practitioners are not equipped with the tools needed to protect, detect and respond to attacks on machine learning systems, even in sectors where security risk is central. It also highlights how some security teams still prioritise familiar threats over model level attacks, with one security analyst quoted as saying, “Our top threat vector is spearphishing and malware on the box. This [adversarial ML] looks futuristic”.

The same report describes a widespread lack of operational readiness, stating that “22 out of the 25” organisations that answered the question said they did not have the right tools in place to secure their ML systems and were explicitly looking for guidance. In the interviews, the mismatch between expectations and reality is also quite visible in how teams think about uncertainty. For example, one interviewee is quoted as saying, “Traditional software attacks are a known unknown. Attacks on our ML models are unknown unknown”. This lack of clarity matters because sleeper agent backdoors are not a niche academic edge case, but are a supply chain style risk that becomes more consequential as models are embedded into core business processes.

How Sleeper Agent Backdoors Were Identified

Backdoors in machine learning have been studied for years, but sleeper agent backdoors in large language models drew heightened attention after research published by Anthropic in 2024 showed that these models can retain malicious behaviours even after extensive safety training. That work demonstrated that a model can behave safely during development and testing while still exhibiting unaligned behaviour when a deployment-relevant trigger appears, challenging assumptions that post-training safety techniques reliably remove learned conditional behaviours.

Microsoft’s “The Trigger in the Haystack” builds on this foundation by focusing on scalable detection, rather than assuming defenders can prevent poisoning outright. The paper frames the objective as practical scanning, stating, “we present a practical scanner for identifying sleeper agent style backdoors in causal language models”. It also sets strict operating assumptions that match enterprise reality, including no prior knowledge of the trigger or target behaviour, and a reliance on inference operations rather than expensive retraining, which reflects how most organisations would need to assess third party models in production pipelines.

How The Backdoors Work In Practice

Sleeper agent backdoors are typically introduced through training time data poisoning or fine tuning poisoning, where an attacker inserts carefully crafted examples into a training dataset so that a trigger pattern is paired with a target behaviour. Over time, the model learns a conditional association between the trigger and the target behaviour while retaining normal responses elsewhere, which means routine prompts may not reveal anything abnormal. Microsoft’s “The Trigger in the Haystack” paper describes this as a conditional policy in which the model follows baseline behaviour unless the trigger is present, at which point it follows a target behaviour learned from poisoned examples.

Three Ways To Tell Poisoned Models From Clean Ones

Microsoft’s recent research identifies three signatures that can help distinguish poisoned models from clean models when the trigger is present. These are:

1. Attention hijacking. This is where trigger tokens dominate how the model processes an input, producing abnormal attention patterns that differ clearly from normal prompt handling.

2. Output entropy collapse. When a trigger is present, the range of possible responses narrows sharply, pushing the model towards deterministic outputs, particularly when the backdoor is trained to produce a fixed result.

3. Memorisation leakage. Microsoft says poisoned models can reveal fragments of the training data used to implant the backdoor, including parts of trigger phrases or associated responses, because those sequences have been memorised more strongly than ordinary training data.

Fuzzy Triggers And Detection Challenges

A complicating factor for defenders is that triggers in language models can be fuzzy, meaning the backdoor may activate even when the trigger is incomplete, corrupted, or only partially present. Microsoft’s “The Trigger in the Haystack” research paper says that altered versions of a trigger can still elicit the backdoor behaviour, and it links this to practical scanning because partial reconstructions may still be enough to reveal that a model is compromised. From a security perspective, fuzziness expands the range of inputs that could activate harmful behaviour, increasing the likelihood of accidental activation and complicating attempts to filter triggers at the prompt layer.

The same fuzziness also alters the threat model for organisations deploying LLMs in workflows that handle user generated text, logs or data feeds. For example, if a model is integrated into a customer support pipeline or a developer tool, triggers could enter through copied text, template tokens, or structured strings, and partial matches could still activate the backdoor. In practice, this means the risk can’t be reduced to blocking a single known phrase, especially when defenders do not know what the trigger is.

Who Is Most At Risk?

The organisations most exposed are those relying on externally trained or open weight models without full visibility into training provenance, especially when models are fine tuned and redeployed across multiple teams. This includes businesses building internal copilots, startups shipping model based features on shared checkpoints, and public sector bodies procuring systems built on third party models. The risk increases when models are sourced from public hubs, copied into internal registries and treated as standard dependencies, since a single poisoned model can propagate into many applications through reuse.

Model reuse amplifies the impact because a single compromised model can be downloaded, fine tuned and redeployed thousands of times, spreading the backdoor downstream in ways that are difficult to trace. Microsoft’s “The Trigger in the Haystack” paper highlights this cost imbalance, noting that the high cost of LLM training creates an incentive for sharing and reuse, which “tilts the cost balance in favour of the adversary”. This dynamic resembles software dependency risk, but the verification problem is harder because the malicious behaviour is embedded in weights rather than in auditable code.

Implications For Businesses And Regulators

For businesses, the practical implications depend on how models are used, but the potential impact can be severe. For example, a backdoored model could generate insecure code, leak sensitive information, produce harmful outputs, or undermine internal controls, and the behaviour may only manifest under rare conditions, complicating incident response. Microsoft’s “The Adversarial Machine Learning – Industry Perspectives” report highlights how organisations often focus on privacy and integrity impacts, including the risk of inappropriate outputs, with a respondent in a financial technology context emphasising that “The integrity of our ML system matters a lot.” That concern becomes more acute as LLMs are deployed in customer facing settings and connected to tools that can take actions.

Governance and compliance teams also face a challenge because traditional assurance practices often centre on testing known behaviours, while sleeper agent backdoors are designed to avoid detection under ordinary testing. In regulated sectors such as finance and healthcare, questions about provenance, auditability and post deployment monitoring are likely to become central, as organisations need to demonstrate that they can manage risks that are not visible through conventional evaluation alone. The practical constraint is that many detection techniques require open access to model files and internal signals, which may not be available for proprietary models offered only through APIs.

Limitations And Challenges

“The Trigger in the Haystack”, approach outlined by Microsoft, is designed for open weight models and requires access to model files, tokenisers and internal signals, which means it does not directly apply to closed models accessed only via an API. The authors also note that their method works best when backdoors have deterministic outputs, while triggers that map to a broader distribution of unsafe behaviours are more challenging to reconstruct reliably. Attackers can also adapt, potentially refining trigger specificity and reducing fuzziness, which could weaken some of the defensive advantages associated with trigger variation.

The broader industry challenge is that many organisations have not yet integrated adversarial machine learning into their security development lifecycle, and security teams often lack operational insights into model behaviour once deployed. Microsoft’s industry report argues that practitioners are “not equipped with tactical and strategic tools to protect, detect and respond to attacks on their Machine Learning systems”, which points to a long term need for better evaluation methods, monitoring, incident response playbooks and provenance controls as LLM use continues to expand.

What Does This Mean For Your Business?

This research points to a security risk that does not align with traditional software assurance models and can’t be addressed through routine testing alone. It shows that sleeper agent backdoors expose a structural weakness in how AI systems are trained, shared and trusted, particularly when harmful behaviour is learned implicitly during training rather than implemented as visible code. The findings from Microsoft and earlier work from Anthropic show that even organisations using established safety and evaluation techniques can deploy models that retain hidden conditional behaviours with little warning before they activate.

For UK businesses, the implications are immediate as large language models are rolled out across customer services, internal tools, software development and data analysis. It suggests that organisations that depend on third party or open weight models now face a supply chain risk that is hard to assess using existing controls, and may need stronger provenance checks, clearer ownership of model updates and more emphasis on monitoring behaviour after deployment. Also, smaller companies and public sector bodies may be particularly exposed due to their reliance on shared models and limited visibility into training processes.

The research also highlights a wider challenge for regulators, developers and security teams as responsibility for managing this risk is spread across the AI ecosystem. Detection techniques are improving but remain limited, especially for closed models where internal access is restricted. As AI systems become more deeply embedded in business operations, sleeper agent backdoors are likely to shape how trust, security and accountability around machine learning systems evolve, rather than being treated as an isolated technical issue.

Physical Intelligence is developing a single AI system designed to power many different robots across tasks and environments, and its research driven approach is reshaping how Silicon Valley views the future of automation.

Building Foundation Models

Physical Intelligence, often referred to as PI or π, is a San Francisco based AI robotics company focused on bringing general purpose artificial intelligence into the physical world. Rather than designing robots for narrowly defined roles or tightly coupling software to specific machines, the company is building foundation models intended to act as a shared intelligence layer for a wide range of robots and physically actuated devices.

The core idea mirrors the impact of large language models (LLMs) in software. For example, just as language models can be adapted to many tasks without being retrained from scratch, Physical Intelligence aims to create a robot brain that can transfer skills across environments, learn from experience, and adapt to new situations without extensive reprogramming. This ambition could place the company at the centre of a growing push towards what researchers describe as embodied AI, where intelligence is expressed through physical action rather than text or images alone.

What Physical Intelligence Is Building

At the heart of Physical Intelligence’s work are vision language action models, known as VLAs. These models combine perception, reasoning and motor control into a single system. Instead of separating vision, language understanding and movement planning into distinct modules, VLAs are trained end to end so the model can observe an environment, interpret instructions, plan a sequence of actions and physically execute them.

First Public Release Back in October 2024

The company’s first major public release was π0 in October 2024, which it described as its first generalist policy. This model was trained using large scale multi task and multi robot data and introduced a new network architecture designed to improve dexterity and generalisation. Subsequent versions have expanded those capabilities. For example, in April 2025, π0.5 introduced what the company called open world generalisation, allowing a mobile manipulator to perform clean up tasks in entirely new kitchens or bedrooms without prior exposure. Also, in November 2025, π*0.6 added reinforcement learning so the model could improve success rates and throughput based on real world experience.

A Mission To Bring General Purpose AI Into The Physical World

On its website, Physical Intelligence describes its mission as “bringing general purpose AI into the physical world” and says it is “developing foundation models and learning algorithms to power the robots of today and the physically actuated devices of the future”. The emphasis throughout its research output is on intelligence rather than hardware, with the company repeatedly arguing that strong generalisation can compensate for relatively simple mechanical systems.

How And Where The Work Is Being Done

Physical Intelligence operates primarily out of San Francisco, where it runs a series of data collection and testing environments. These include warehouse style spaces, domestic settings and test kitchens filled with everyday appliances and furniture. Robots are exposed to real tasks such as folding clothes, assembling boxes, operating kitchen equipment and manipulating unfamiliar objects.

The company follows a continuous training loop. For example, robots perform tasks in these environments, data is collected from those interactions, new models are trained using that data, and the updated models are then redeployed for further evaluation. The company says this process allows the system to learn from failure and success in physical settings rather than relying solely on simulation.

Human To Robot Transfer

Human to robot transfer is another key element of the company’s approach. For example, several of its published research posts explore how robots can learn from human video data, allowing models to absorb information about actions and affordances without requiring every behaviour to be demonstrated physically by a robot. Back in a December 2025 research article titled Emergence of Human to Robot Transfer in VLAs, the company explained how this capability begins to appear naturally as models scale, rather than being explicitly programmed.

What Makes Physical Intelligence Different?

What seems to distinguish Physical Intelligence from many robotics startups is its apparent refusal to prioritise near term commercialisation. For example, the company does not offer investors a clear timeline for revenue generation and has not launched a mass market product. Instead, it has positioned itself as a long horizon research organisation focused on solving what it sees as the core problem in robotics, which is general purpose physical intelligence.

Despite this, the company has raised around $1 billion and was valued at approximately $5.6 billion following a $600 million funding round in late 2025. That round was led by CapitalG (Alphabet’s growth stage venture capital fund) and included participation from Lux Capital (a science and deep tech focused venture capital firm), Thrive Capital (a technology focused venture capital firm), and Index Ventures (a global venture capital firm investing in technology companies), T. Rowe Price and Jeff Bezos. According to reporting from Bloomberg and Axios, much of the company’s spending is directed towards compute and large scale data collection rather than manufacturing or sales infrastructure.

The leadership team has been explicit about this strategy, and on its website and in published research updates Physical Intelligence frames progress in terms of model capability rather than deployment milestones, stating that its internal roadmap originally projected five to ten years of development, even though some technical goals were reached earlier than expected as models scaled.

The Competitive Landscape

It should be noted here that Physical Intelligence is not the only company working on producing general purpose robotics, but it represents one end of a wider strategic divide. For example, one of its most prominent counterparts is Skild AI, a Pittsburgh based company founded in 2023 that is also building a general purpose robotic brain. Skild has raised more than $1 billion and claims its Skild Brain has already been deployed commercially across security, warehouse and manufacturing environments, generating tens of millions of dollars in revenue.

Skild takes a more deployment led approach and has publicly criticised what it views as over reliance on vision language models trained primarily on internet data. For example, in a July 2025 blog post titled Building the General Purpose Robotic Brain, the company argued that many robotics foundation models are “VLMs in disguise” that lack true physical common sense because they do not contain sufficient action grounded data. Skild instead emphasises large scale simulation combined with targeted real world data as the path to scale.

Other companies operating in adjacent areas include Figure AI, which is developing humanoid robots with backing from Microsoft and OpenAI, Agility Robotics with its Digit robot designed for warehouse work, and large internal research efforts at organisations such as Google DeepMind, Tesla and Nvidia. These groups vary widely in how closely they couple hardware and software, and in how quickly they seek commercial deployment.

Implications For Businesses And The Robotics Market

If Physical Intelligence’s approach proves effective, it could really lower the cost and complexity of deploying robots across multiple industries. A shared intelligence layer that can be transferred between platforms would reduce the need for bespoke programming and make automation more flexible. Logistics, grocery fulfilment and manufacturing are already being explored through limited partnerships, according to the company and investor statements.

Also, the implications extend beyond efficiency gains. For example, more adaptable robots could change how businesses think about workforce planning, task allocation and safety. At the same time, general purpose physical intelligence raises regulatory and operational questions, particularly around reliability, accountability and failure modes in unpredictable environments.

Challenges And Criticisms

Despite strong investor backing, Physical Intelligence does face some substantial challenges. For example, critics question whether a single model can actually generalise effectively across a wide range of physical tasks without becoming inefficient or unpredictable. Others have pointed to the cost of large scale computing resources and the practical difficulty of collecting high quality real world robotics data at scale.

Hardware is also a constraint. For example, Physical Intelligence has acknowledged in its research posts that working in the physical world introduces delays, safety limitations and mechanical failures that do not exist in software only systems. These factors slow experimentation and complicate iteration.

There are also some unresolved questions about demand. While investors appear willing to tolerate long timelines, it remains unclear which markets will first adopt general purpose robotic intelligence at scale and under what economic conditions. For now, Physical Intelligence continues to focus on advancing core capabilities rather than answering those commercial questions directly.

What Does This Mean For Your Business?

Physical Intelligence is betting that solving general purpose physical intelligence first will ultimately unlock more durable and transferable value than pursuing early, narrow deployments, and that wager now sits at the centre of an increasingly important debate in robotics. The contrast with more commercially focused competitors highlights a fundamental uncertainty in the market about whether generalisation is best achieved through long term research or through rapid real world deployment and iteration. The answer is unlikely to be settled quickly, particularly given the technical difficulty of training systems that can reliably operate across unpredictable physical environments while remaining safe, efficient and economically viable.

For UK businesses, this work points to a future where robotics adoption may become less about investing in bespoke machines for individual tasks and more about accessing shared intelligence layers that can adapt over time. Sectors such as logistics, manufacturing, food production and facilities management could eventually benefit from more flexible automation, although near term deployment will continue to depend on cost, reliability and regulatory clarity. For investors, policymakers and workers, the progress of companies like Physical Intelligence will shape expectations around how quickly embodied AI moves from research environments into everyday operations, and how the balance between innovation, safety and economic impact is managed as robots become more capable and more general purpose.

Amazon Web Services closed 2025 with its fastest quarterly growth rate in over three years, reflecting renewed enterprise cloud migration and a sharp increase in demand for artificial intelligence infrastructure.

Cloud Division’s Strongest Growth Rate In 13 Quarters

Amazon disclosed in its fourth quarter financial results that AWS generated $35.6 billion in revenue in the three months to 31 December 2025, representing year on year growth of 24 per cent. This was the cloud division’s strongest growth rate in 13 quarters and marked a clear re-acceleration following a prolonged slowdown across the global cloud market. The performance contributed to Amazon’s total quarterly revenue of $213.4 billion, up 14 per cent compared with the same period in 2024.

In its recent news release about its latest financial results, Amazon Web Services (AWS) was shown to be a key factor in underpinning Amazon’s profitability. Operating income for the cloud unit actually rose to $12.5 billion in the quarter, up from $10.6 billion a year earlier. In fact, for the full year, AWS revenue reached $128.7 billion, an increase of 20 per cent, while operating income climbed to $45.6 billion, reinforcing the division’s role as Amazon’s most lucrative business.

AWS Growth In Context

The renewed momentum followed a period of slower expansion during 2023 and much of 2024, when many organisations reduced cloud spending, optimised workloads, and delayed large infrastructure projects in response to economic uncertainty. Against that backdrop, the fourth quarter performance stood out both for its growth rate and the scale of the underlying business.

AWS now operates at an annualised revenue run rate of more than $140 billion, meaning incremental growth translates into substantial absolute revenue gains. During the earnings announcement, Andy Jassy, President and CEO of Amazon, highlighted this dynamic, stating that “AWS growing 24 per cent (our fastest growth in 13 quarters)” reflects the company’s ability to add more incremental revenue and capacity than competitors operating from smaller bases.

The figures indicated that AWS is not only regaining pace but doing so at a size that continues to shape the economics of the global cloud market.

Drivers Behind The Reacceleration

Amazon’s results and accompanying commentary have pointed to several overlapping factors behind AWS’s growth. For example, one of the most consistent drivers remains enterprise migration from on premises infrastructure to the cloud. It seems that large organisations are continuing to move core systems, data, and applications away from privately owned data centres, a process that typically unfolds over multiple years rather than as a single project.

Artificial intelligence (AI) has emerged as a second and increasingly significant driver. Training and operating large AI models requires vast amounts of computing power, high performance storage, and advanced networking, all of which favour hyperscale cloud platforms. Amazon said customers increasingly want to run AI workloads in the same environments as their existing applications and data, rather than building separate infrastructure.

Strength From Vertical Integration

AWS has positioned itself to support this demand through a vertically integrated approach to AI infrastructure. In other words, AWS isn’t relying on lots of separate external suppliers for different parts of AI computing. Instead, AWS designs and runs most of the key building blocks itself, including its own AI chips, its data centres, its networking, and the software services that customers use to build and run AI systems. By controlling more of the stack end to end, AWS can optimise performance, manage costs, and scale AI workloads more efficiently as demand grows.

For example, the company has invested heavily in custom silicon, including its Trainium accelerators for machine learning workloads and Graviton processors for general purpose computing. Amazon says that these chips now have a combined annual revenue run rate of more than $10 billion and are growing at triple digit rates year on year.

Trainium2, which powers a large share of inference workloads on Amazon Bedrock, has already seen 1.4 million chips deployed. Amazon has also confirmed that demand for Trainium3 is strong enough that most available supply is expected to be committed by mid 2026, with further generations planned for future deployment.

Enterprise Adoption And New Agreements

AWS’s growth was also supported by a broad set of new and expanded customer agreements during the quarter. For example, Amazon reported new AWS deals with organisations including OpenAI, Visa, the NBA, BlackRock, Salesforce, the U.S. Air Force, HSBC, the London Stock Exchange Group, and Thomson Reuters.

Large enterprises and public sector bodies tend to move cautiously when choosing cloud infrastructure providers, especially for systems that support core operations. Securing new agreements at this level often involves long evaluation processes and reflects a high degree of trust in reliability and security. Continued wins with these organisations are, therefore, reinforcing AWS’s position as a widely used platform for large scale and mission critical workloads.

Amazon also said AWS added more than a gigawatt of power capacity to its global data centre network during the quarter. It’s worth noting here that access to power has become a key constraint across the cloud industry as AI workloads drive rapid expansion in compute demand, making physical infrastructure investment a central part of competitive strategy.

Competitive Position In The Cloud Market

AWS is the largest cloud infrastructure provider globally, ahead of Microsoft Azure and Google Cloud. While rivals have also reported strong growth tied to AI adoption, AWS’s fourth quarter results highlighted its ability to convert that demand into large scale revenue growth.

Analysts have also noted that AWS added more absolute revenue during the quarter than its closest competitors, even where those competitors reported higher percentage increases. In a maturing cloud market, scale increasingly determines pricing flexibility, investment capacity, and long term competitiveness.

At the same time, competition for AI workloads is intensifying. For example, Microsoft continues to deepen its relationship with OpenAI, while Google is promoting its own AI models and custom accelerators. AWS’s approach has focused more on offering multiple third party and proprietary models through Amazon Bedrock, thereby allowing customers to select and switch between models without rewriting applications.

Investor Reaction And Financial Pressures

Despite the strong AWS performance, Amazon’s share price actually fell sharply following the results announcement, dropping around 10 per cent in after hours trading. The market reaction was driven less by revenue growth and more by concerns over spending levels and near term profitability.

For example, Amazon confirmed plans to invest approximately $200 billion in capital expenditure during 2026, up from around $125 billion in 2025. The company said the majority of this spending will be directed towards cloud and AI infrastructure, including data centres, chips, networking equipment, and energy capacity.

Free cash flow for 2025 declined to $11.2 billion, down from $38.2 billion the previous year, primarily due to increased investment in property and equipment. Amazon has acknowledged these pressures in its forward looking statements, noting that results remain subject to uncertainty from factors such as global economic conditions, energy prices, supply constraints, and customer spending behaviour.

Implications For Businesses And Other Stakeholders

For businesses, AWS’s reaccelerating growth shows that demand for cloud and AI infrastructure is intensifying rather than stabilising. This means that organisations that delay cloud migration or AI adoption may face higher costs or limited availability as demand for cloud infrastructure and processing capacity continues to increase.

For technology suppliers, including chip manufacturers and energy providers, Amazon’s expansion plans point to sustained demand but also rising expectations around efficiency, sustainability, and scale. Data centre power availability and energy sourcing are becoming central considerations in hyperscale growth strategies.

For regulators and policymakers, the concentration of AI infrastructure among a small number of global providers continues to raise questions around resilience, competition, and environmental impact, particularly as data centre power consumption grows.

Challenges And Ongoing Criticism

Although AWS delivered some pretty strong growth, underlying challenges remain, with margin pressure continuing as Amazon invests heavily to expand capacity ahead of demand and relies on long term AI adoption to justify current spending levels.

Also, there are some major environmental and infrastructure concerns. For example, expanding data centre capacity by gigawatts requires reliable access to power and water, often in regions already under strain. These constraints are increasingly shaping where and how cloud providers expand.

It’s also worth noting here that customer behaviour has evolved. This has meant that organisations are more cost conscious, more technically sophisticated, and more willing to distribute workloads across multiple providers, increasing competitive pressure even for market leaders.

Taken together, AWS’s fourth quarter results seem to show that demand for cloud and AI infrastructure strengthened significantly towards the end of 2025, while the financial, operational, and environmental challenges involved in meeting that demand also became more apparent.

What Does This Mean For Your Business?

AWS’s late 2025 performance points to a cloud market that has moved out of a cautious holding pattern and back into an expansion phase, driven largely by long term AI infrastructure demand rather than short term optimisation cycles. The results suggest that cloud growth is no longer being fuelled simply by migration from on premises systems, but by a deeper reliance on hyperscale platforms as the default foundation for advanced computing, data processing, and AI deployment. At the same time, the scale of investment required to sustain this growth is reshaping the economics of the sector, placing greater emphasis on capital intensity, energy access, and execution discipline.

For UK businesses, this environment reinforces the reality that cloud capacity and AI infrastructure are becoming more competitive resources. Organisations planning digital transformation, data modernisation, or AI adoption will need to think more carefully about timing, cost exposure, and provider dependence, particularly as demand pressures and infrastructure constraints intensify. Public sector bodies, financial institutions, and regulated industries may also face growing scrutiny around resilience, data governance, and environmental impact as reliance on a small number of global providers deepens.

For other stakeholders, including investors, regulators, and infrastructure partners, AWS’s trajectory highlights a market where growth opportunities remain substantial but increasingly complex. Strong revenue momentum now sits alongside rising financial risk, environmental pressure, and regulatory attention. The fourth quarter results highlight how hyperscale cloud growth is far from over, and they also show that sustaining it will require navigating trade offs between speed, scale, profitability, and long term sustainability across the entire cloud ecosystem.