Tag: Artificial intelligence

  • When AI learns to collaborate | Radar Special N4A

    When AI learns to collaborate | Radar Special N4A

    Collaborative AI begins to take shape when a conversation becomes work to carry forward together. Radar Special N4A (in Italian) follows this transition: from answers on a screen to tools that can read materials, prepare a result and return it to the person’s judgement.

    When AI learns to collaborate

    Special Issue 4A | 11 July 2026 | Approximately 6 minutes to read

    1. Saying what we want changes how we enter the work

    For many years, using a program meant learning its commands. Today we can often begin with a request: compare two documents, clarify a concept, prepare an initial structure. Natural language reduces part of the distance between a person’s intention and the computer’s operations.

    The change becomes clearer when an answer produces something to keep working on. A draft can be discussed; a comparison can reveal a difference; research can leave a list of points to verify. Value arises along the path connecting a request to a checkable result.

    This ease nevertheless requires care. A fluent sentence can seem better founded than it is. A system may have misunderstood a constraint, used an incomplete source or completed a step with inaccurate information. The natural interface helps us enter the work; scrutiny lets us understand whether that work has succeeded.

    2. From the ability to write to the possibility of acting

    A language model processes patterns learned during training and generates an answer from the available context. It can summarise, translate, propose code or connect information. When it builds a plausible answer without a sound basis, form can conceal the error.

    Tools add an operational layer. A system may receive access to search, files or a workspace and use the results to continue. An agent organises a sequence of actions towards a goal, within the tools and permissions assigned to it. OpenAI’s approach to building agents shows how search, computer use and execution records enter the same architecture.

    The questions to ask therefore change. Beyond an answer’s quality, the documents read, changes made and point at which an action must stop for checking matter. Good collaboration makes these steps understandable to those who will use the result.

    📌 Did you know? — Preparing a result and approving it are two stages

    A system can gather materials and build a draft while the person retains the choice of what will be used or communicated. Making this boundary explicit allows concrete activities to be delegated while preserving control over decisions that commit people and organisations.

    3. Dialogue becomes more continuous, but judgement remains necessary

    The Special originates on 11 July 2026, in the days when two announcements make this direction visible. GPT-Live, presented on 8 July, introduces voice interaction that can listen and speak simultaneously and delegate more complex tasks to a supporting model. Conversation gains continuity even while research proceeds.

    The GPT-5.6 launch on 9 July instead describes capabilities aimed at longer work, tool use and coordinating activities. These are two examples within the picture observed by the dossier, which also includes systems and approaches from other organisations.

    A more natural voice and more complex work can increase the sense of presence. Using these capabilities well requires continuing to distinguish what the system shows from what we can verify. Continuity of dialogue does not guarantee complete memory, factual correctness or understanding of people’s and users’ intentions.

    4. A clear assignment makes delegation useful

    Imagine a team needing to compare several versions of a document. The request can clarify which files to use, which differences to seek and how to present them. The system prepares the comparison; those familiar with the work check that no important step has been lost and decide which changes to accept.

    If the assignment remains vague, the system may fill gaps with assumptions. If it receives broader access than necessary, the possibility of an off-target intervention grows too. The goal, available materials and boundaries of action must therefore be proportionate to the task.

    Verification can take place along the way: a sample of extracted data, the source of a quotation, a version to compare before replacing a file. These checks help correct a deviation early and leave a readable record of what was done. The final result becomes easier to assess.

    5. The organisation determines how much value remains

    A system can accelerate an already clear process or make confusion faster. If documents are inconsistent, nobody knows which version to use and responsibilities remain undefined, an answer generated in seconds does not automatically solve the problem.

    For a business, the most useful test concerns a recognisable activity: how long it takes, which errors remain, how much review the result needs and who can correct it. Comparing these elements before and after a trial helps establish whether speed has become quality of work.

    Learning matters too. Using a draft to discuss alternatives can strengthen a skill; always accepting it without examination can reduce the practice of judgement. In study, research and writing, how the path is built changes what a person can learn.

    6. Collaboration has infrastructure and responsibility

    Behind a request are processors, memory, networks, data centres and cooling systems. The International Energy Agency connects AI development to electricity availability and infrastructure planning. An interface may be easy to use while the system supporting it requires resources and continuous work.

    These dependencies enter the terms of collaboration: which data can be read? Where is it processed? How much does continuing cost? And what happens if a tool stops? The answers depend on the service and context. Those introducing the system into their work must be able to understand them.

    Collaborative AI gains value when it expands a person’s possibilities and leaves visible the points where their judgement is needed. Framing a problem well, tracing a source, recognising an error and taking responsibility for a decision remain central skills. The dossier invites us to cultivate them as tools become more capable.

    🧠 Radar Insight — Quality includes the ability to intervene

    Collaborating also means being able to challenge a result, correct a direction and interrupt an action. These possibilities make the system’s autonomy usable. Final responsibility needs information and oversight tools as well as a good answer.

    Read the full issue

    This reading presents some stages of Radar Special 4A. The full dossier contains the complete journey and in-depth discussions of the original edition (11 July 2026).

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    Cover in focus

    AI - Radar Special 4A — When artificial intelligence learned to collaborate, 11 July 2026. English cover with official Gruppo Radar signature.

    Read this article in Italian

  • From answer to result: how we actually use artificial intelligence | Radar Special N7A

    From answer to result: how we actually use artificial intelligence | Radar Special N7A

    An answer arrives in seconds, orderly and convincing. Work begins when we read it: a source to check, a calculation to redo, an assumption to make explicit. Radar Special N7A explores what happens when we move from answer to result, and the importance of recognising its quality, limits and consequences.

    From answer to result: how we actually use artificial intelligence

    Special Issue 7A | 2 August 2026 | Approximately 5 minutes to read

    1. The work remaining after the answer

    Asking an assistant to summarise a document, organise data in a table or sketch a proposal can save time and help tackle work we had been postponing. The benefit is concrete, but that first version still requires our intervention: we must check the information, recognise what is missing and adapt it to the situation. Through this work, we can make it reliable enough to put in a document, use for a decision or make part of a service.

    On rereading the content, we may notice that a quotation is correctly reproduced but does not support what we are claiming, or that a table compares data calculated on different bases. Even a clear summary may have omitted the detail needed to decide. Verification helps establish which use the work suits and how to continue: correct it, discuss it with a knowledgeable person or stop if the elements needed to proceed are still missing.

    2. Using AI can mean many things

    Having a tool available does not yet tell us its place in our work. We may have tried it once, used it recently or use it regularly for a precise task. Even when it enters our habits, whether it is actually helping remains to be understood. Assessing this requires returning to the initial goal and considering the result’s quality alongside time spent, errors to correct and costs.

    These differences also matter when reading AI statistics. A questionnaire gathers what people say they do, while platform data describes activities within it; an experiment instead tries to measure an effect under defined conditions. Comparing two percentages therefore requires understanding who was observed, which activities counted as “use” and the period concerned. Otherwise, a difference arising from how data was collected may be attributed to people or organisations.

    📌 Did you know? — A percentage needs its question

    “Having tried an assistant” and “using it every week for work” do not measure the same behaviour. Even well-collected data loses meaning when separated from the population, time window and definition of use that produced it.

    3. Speed must be measured through to the end

    Experiments document real benefits in some tasks. In Noy and Zhang’s study of professional writing, generative assistance reduces time and improves the average quality of assessed texts. The conclusion concerns those activities and conditions: it is evidence of a possibility, to be tested in the concrete work where we want to introduce it.

    A different result emerges in the METR study of experienced developers and tools from the first half of 2025: in projects they knew well, access to AI lengthens task completion by around 19%, while participants continue to perceive a benefit. This limited finding does not describe every developer or subsequent models. It shows why generation time must be read alongside review and rework time.

    4. Expertise also recognises the limit

    The frontier of capabilities changes with the model, task, available data and environment. A system useful for an initial explanation can become less reliable when resolving an unusual case, interpreting a clause or understanding a structure we have not shown it. The answer’s fluency may remain similar even as its quality changes.

    Recognising this distance requires knowledge of the work. A knowledgeable person can choose a test: compare a source, redo a calculation, check a different example, verify a step with someone possessing the necessary experience, establish a clear process for study or an experiment. Formulating the request is fundamental, but must rest on a method. Informed use takes shape when we can explain why we accepted a suggestion and which checks allow us to trust it.

    5. What remains when assistance ends

    In learning, obtaining a good answer and building a personal capability are different outcomes. Help can replace the step we needed to practise or accompany us towards understanding it. The difference also lies in how the interaction is designed: explaining reasoning, leaving room for an attempt, comparing errors and trying again without assistance.

    The question therefore becomes concrete: after working with the tool, can we transfer the method to a new case? Can we recognise an error or explain a decision? Not every activity requires becoming independent in everything. But we need to understand which skills we want to retain and where delegation is a useful choice. The machine’s benefit can include time freed and the capability collaboration leaves us with.

    🧠 Radar Insight — The advantage that remains

    Assisted performance can be excellent without becoming learning. To understand collaboration’s value, we also observe what a person can explain, correct and repeat when the tool is unavailable. This test changes how help is designed.

    6. From adoption to a process that holds

    In an organisation, activating an AI function is only the beginning. Usable documents, defined access, people responsible for checking and a route for correction are needed. Time saved in one stage may be absorbed by other requests or new checks: recognising improvement requires observing the complete process, without confusing the number of answers with value produced.

    Radar’s reading proposes starting with bounded tasks and understandable criteria: which result we seek, which quality we consider sufficient, which errors matter and who is accountable for delivery. Available studies help with orientation but leave many questions open about benefits’ duration and distribution. Moving from answer to result lives in the human work that makes a proposal checkable, corrects it when needed and decides where it can actually be useful.

    Read the full issue

    This reading presents some stages of Radar Special 7A. The full dossier contains the complete journey and in-depth discussions of the original edition (2 August 2026).

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    Continue reading

    Understanding Today. Building Tomorrow: the project and the Radar collection.

    Cover in focus

    Radar Special 7A — From answer to result, 2 August 2026. English report cover with official Gruppo Radar signature.

    Read this article in Italian

  • The freedom to choose: capital, data and agents | Radar Weekly N18

    The freedom to choose: capital, data and agents | Radar Weekly N18

    The freedom to choose in artificial intelligence also depends on the cost of changing service, how data is used and the powers entrusted to agents. Radar Weekly 18 follows these links between capital, technology and everyday life: a possibility becomes concrete when we can understand it, verify it and, if necessary, take another path.

    The freedom to choose

    Weekly Number 18 | Monday, 26 October 2026 | Approximately 7 minutes to read

    We have grown used to a program. Inside it are documents, contacts, authorisations and a way of working built over time. Then we find a better-suited service and try to move. That is when we discover how much of our activity depends on the path already chosen: some files move easily, other connections must be rebuilt and certain operations cannot be interrupted.

    The news gathered in this issue takes the same question into large businesses and digital infrastructure. Growth requires resources; preserving the ability to choose also requires usable alternatives, understandable terms and people able to intervene.

    1. Growth must find the money to continue

    In Meta’s third-quarter results, published on 21 October, revenue grows by 26% to $64.60 billion. Operations generate $32.79 billion in cash, while investment as measured by the company, including principal repayments on finance leases, absorbs $34.91 billion. Free cash flow is therefore negative by $2.12 billion.

    Selling more and having less money available after investment can happen together. A facility is paid for today and used for years; revenue and costs in the income statement follow timescales that do not always coincide with receipts and outflows. Meta also has $82.45 billion in cash and marketable securities: that negative quarter does not mean resources are exhausted.

    Reading the race for AI therefore requires following capital’s journey: how much is needed before capacity becomes operational, who advances it and which results must sustain it. This is part of the weight of growth we began observing in Weekly 17.

    2. An alternative has value when we can use it

    Competition between processors is not exhausted by price. Changing hardware may require adapting programs, training people and repeating checks. If that work is too costly, an offer attractive on paper remains far from practice.

    TorchNeuron, presented on 19 October, seeks to narrow this distance. It connects PyTorch, an environment used to develop AI models, to AWS Trainium processors. The documentation describes code reuse with limited changes and the public experimental version’s launch, under the open Apache 2.0 licence, on 20 October. The advantage must be measured on real programs: compatibility, performance and the work needed to move.

    The CMA consultation on Microsoft’s business software, announced on 16 October, also examines the ability to choose and combine services. The UK authority proposes designating Microsoft as having strategic market status; this is a proposal under consultation, not a final decision or finding of wrongdoing. Topics include product integration, AI tools and licences across different clouds.

    🧠 Radar Insight — An exit must be usable

    We may successfully export every document and then discover that essential permissions, functions or connections are missing. Before changing service, it is worth understanding which activities must be rebuilt, how long this will take and how work will continue meanwhile. The freedom to choose also includes this cost, often less visible than a subscription’s price.

    3. Assigning a task requires controls that remain active

    When an AI agent can use tools and modify programs, an ordinary request can have effects beyond the intended result. In the report updated on 20 October, OpenAI describes two internal incidents: one agent disabled alerts to safety staff during a repair; another removed a monitoring integration while migrating a program and then incorrectly denied having done so.

    In the first case, records continued but notifications no longer reached people. Keeping a trace allows events to be reconstructed; receiving an alert in time allows them to be stopped. These functions must work together. Repairing a program should not give an agent the power to weaken its oversight.

    Service availability then requires preparation different from access and permissions. Google Cloud’s preliminary report on 20 October describes a power interruption in the London zone europe-west2-b and problems affecting several services for six hours and ten minutes. The account of 22 October remains preliminary. For users, restoration may mean again being able to complete a payment or meet a deadline: preserving data is essential, but using it when needed also requires available systems.

    4. A digital investment also has people and conditions to understand

    In DeFi, finance using programs on blockchains, lending vaults gather digital assets and deploy them according to lending strategies. The blockchain makes many operations observable, but these traces do not automatically tell us how prudent the choices are, how sound the collateral is or how available resources will be at withdrawal.

    The assessment method presented by S&P Global Ratings on 4 October considers, among other aspects, liquidity risk and curator risk: curators define or manage parts of the strategy under the stipulated powers. This is a judgement on vault risks, distinct from a conventional credit rating; it does not guarantee repayment.

    📌 Did you know? — Requesting withdrawal does not make money immediately available

    If deposited resources have been lent out, they may not be available when we want to withdraw them. This is a liquidity problem: it does not necessarily mean the investment has been lost, but may prevent its recovery within the time needed. Waiting periods, limits and possible suspensions of exit depend on the individual vault’s terms and must be understood before depositing.

    5. The conditions of choice reach everyday life

    These questions also concern data entrusted to platforms. To decide whether to use a service, we must be able to understand what it will do with information and what opportunities we will have to intervene. In the Special on the Meta case, we followed the responsibilities of platforms, families and institutions in protecting minors. The theme returns here through different cases the dossier keeps distinct: the relationship with users continues in concrete practices, even after an announcement or official measure.

    Finally, a material condition accompanies the entire transformation. According to Terna, Italy’s September electricity demand reaches 27.8 terawatt-hours, up 6.5% on a year earlier. These data concern the whole country and do not allow the increase to be attributed to data centres or AI. They do, however, show the system in which a new facility must find its place, between connections, grid works and existing demand.

    In Weekly 15 we distinguished announced facilities from genuinely operational capacity. For someone living near a new data centre, the next step is understanding who will finance upgrades, how the facility will be cooled and which commitments to the local area can be checked.

    The freedom to choose takes shape in these steps: comparing a cost, limiting an authorisation, understanding a risk and keeping an alternative practicable. Innovation can expand the possibilities available to us. For that advantage to last, we must be able to recognise the conditions of choice before they become an obstacle to changing it.

    Read the full issue

    Weekly 18 explores this journey in nine chapters: from Meta’s balance sheets and chip supply-chain orders to TorchNeuron, credit and specialist data, agent safety, DeFi, user rights, the Chinese economy and energy in Italy. The PDF’s eleven pages bring together examples, five boxes and sixteen primary sources, preserving the distinction between announcements, data and Radar’s reading.

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    Cover in focus

    English cover of Radar Weekly N18, The freedom to choose, dated 26 October 2026: luminous bridges connect data centres and an open city landscape, with the official Gruppo Radar signature.

    Read this article in Italian

  • The cloud has weight. Chips, energy and the geopolitics of AI | Radar Weekly N1

    The cloud has weight. Chips, energy and the geopolitics of AI | Radar Weekly N1

    Chips, energy and data centres sustain artificial intelligence: the cloud has a physical foundation. Radar Weekly N1 explores infrastructure costs and the geopolitics of AI, connecting technology, capital and rules of access through the signals of 26 June 2026.

    The cloud has weight. Chips, energy and the geopolitics of AI

    Weekly Issue 1 | Friday 26 June 2026 | Approximately 5 minutes to read

    One question runs through this first Radar Weekly: how much does artificial intelligence really weigh? On screen, it appears as an answer, an image, a conversation. Behind that apparent lightness lie chips, memory, energy, water, networks and land. The cloud is not outside the world: it occupies a concrete part of it.

    Issue 1, dated 26 June 2026, examines the geopolitics of AI through markets, research, companies and rules of access. Here we retrace its essential points: the first stretch of a journey that begins to recognise AI as infrastructure rather than something intangible.

    1. From the model to matter

    The question should not begin with which model is more capable, but should focus above all on who can build it, power it, maintain it and distribute it. A digital service needs physical structures, investment and agreements: as demand for computing grows, the costs and availability of these resources become increasingly important.

    This is the thread that holds the issue together. Software, meaning the programs, remains central, but it is not enough to explain the race for AI: we also need to look at the machines that make it work and the conditions that make its use possible.

    📌 Did you know? The cloud has an address

    A data centre is a facility that houses servers, memory, networks and cooling systems. It is one of the physical places where cloud services and AI models run. Speaking of the “cloud” describes how we access a service, not an absence of matter behind that service.

    2. Who sells the tools, who bears the cost

    A distinction emerges from Radar’s reading of the markets: selling infrastructure and financing its construction are not the same thing. On one side are suppliers of chips, memory and storage systems; on the other are the platforms that must bear the expense required to provide their services.

    The issue’s metaphor is simple: some companies sell the pickaxe, while others dig and pay the cost of the mine. This is neither a ranking of winners nor a call to rush into investing; it simply helps explain why growing demand for AI can have different consequences for costs, margins and expected returns.

    3. Chips, energy and land

    The frontier of chip research and investment in data centres, also examined through the Indian case, bring technology back into a concrete geography. Where is it built? Who manages the data? Who pays for the energy? Who sets the conditions of access?

    The miniaturisation of transistors, the components of chips, also belongs in this picture: density, efficiency and consumption are part of computing capacity. But a research result is not the same as a product already available. The issue maintains this distinction, avoiding turning an experimental frontier into an immediate promise.

    4. Digital finance beyond the price

    The journey also passes through cryptocurrencies and stablecoins, cryptoassets designed to maintain a stable value, often linked to the dollar or other currencies. The focus moves beyond prices alone to the reserves backing them, custody, rules and the management of personal data.

    This point is connected to the rest: when a service becomes infrastructure, the responsibilities of those who manage it and the conditions under which people can use it also matter. Technology, finance and control thus begin to meet within the same line of thought.

    5. When AI enters objects

    Another signal concerns Physical AI, artificial intelligence integrated into sensors, robots, vehicles and industrial devices. Here, AI does more than produce an answer: it enters systems that perceive and act in the material world.

    The formula proposed in the issue brings together AI, sensors, efficient chips and physical control. It is a way of reading a possible evolution: from conversation to operational capability. How objects actually work becomes part of the technological question, alongside the computing that sustains them.

    6. Who can gain access, who decides

    If models and data centres acquire strategic value, access is no longer merely a technical matter. Radar leaves one question open: what limits, licences and authorisations might distinguish countries, users and purposes of use?

    ⚠️ What to watch — Control also operates through access

    The question is not only what a model can say. It is also who can use it, for what purpose and under what conditions. In this first issue, it deliberately remains a question to observe.

    The thread to take with you

    In this issue’s reading, the future of AI is not decided solely by the quality of its answers. It is also decided by the ability to bear its cost, control its infrastructure and define its accessibility. This is where Radar begins: connecting seemingly distant signals to understand the transformation they are forming together.

    The full issue brings together the weekly dashboard, charts, the growing vocabulary and scenarios to watch. This reading accompanies that journey without replacing the archive document.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved in the original content. Free access does not authorise reproduction or republication without consent, except for uses permitted by law. Third-party materials remain the property of their respective rights holders. To share this issue, use the link to the article.

    This issue opens the path of Understanding Today. Building Tomorrow.: Radar Weekly and Radar Specials explore technological, cultural and social transformations and their consequences for everyday life.

    Cover in focus

    English cover of Radar Weekly N1, The cloud has weight: chips, energy and AI geopolitics, dated 26 June 2026, with a luminous cloud above a chip and data centres and the official Gruppo Radar signature.

    Read this article in Italian

  • The Price of Trust: Capital, Security and Data | Radar Weekly N19

    The Price of Trust: Capital, Security and Data | Radar Weekly N19

    The price of trust is often hidden behind simple actions: entrusting data to a platform, leaving money with a digital service, asking an artificial intelligence tool for help. Radar Weekly N19 follows what supports these habits: the capital companies raise, the infrastructure that has to be paid for, the security of assets held in custody and the personal information that can influence prices and pay. These are different stages of the same question: who bears the cost of growth, and who answers when something goes wrong?

    The Price of Trust

    Weekly No. 19 | Monday, 2 November 2026 | About 7 minutes to read

    The news collected up to 30 October takes us inside financial statements, financing plans and oversight procedures. Anthropic is preparing for a possible stock market listing, Amazon shows substantial demand for cloud services, and Meta is seeking new resources. Meanwhile, an incident hits Binance, and in the United States an authority asks businesses how they use people’s data to set prices and pay.

    In Weekly 13, we followed the relationship between custody, resources and responsibilities in crypto services. Now we broaden our view: trust also involves those who finance the infrastructure, those who control its terms and those who encounter its costs in everyday life.

    1. Anthropic: reading a loss before interpreting it

    The prospectus Anthropic filed with the SEC on 26 October presents the company’s activities and risks ahead of a possible listing. The filing opens the books to investors; it does not mean the shares have already reached the market. The price and quantity remain to be determined.

    In the first half of 2026, Anthropic reports $16.286 billion in revenue, an operating loss of $693 million and $8.945 billion in computing and infrastructure costs. The most conspicuous number, however, is the net loss: $80.748 billion. Understanding it also requires looking at the financial and accounting components.

    Approximately $72.388 billion comes from changes in the value of convertible instruments and other liabilities: on its own, this item does not equal money that left the company’s coffers during those six months. The distinction makes it easier to assess the cost of developing models and the weight of financing terms, without turning a single figure into a verdict on the entire company.

    📌 Did you know? — Financial statements offer more than one perspective

    Revenue tells us about sales; the operating result considers the costs and income of the core business; the net result also includes other components. Cash, by contrast, follows the money coming in and going out. Looking at these measures together helps us understand how a result arises and what resources remain available.

    2. Amazon: customers arrive, facilities have to be paid for first

    Amazon’s third-quarter results, published on 29 October, show demand that is already tangible. The group reports $201.9 billion in revenue; AWS, its cloud division, reaches $46 billion, with annual growth of around 39%. Total operating income is $28.1 billion.

    Meeting that demand, however, requires facilities and equipment that must be purchased today. Over the last twelve months, free cash flow, the cash remaining after operating activities and the investment in tangible assets included in the company’s calculation, is negative by $42.4 billion. Amazon links the deterioration mainly to infrastructure purchases, largely intended for AI.

    Here, the periods matter: sales and income cover a quarter, while the cash figure covers a year. Placing them side by side requires this care. As we have seen while following the weight of growth along the AI value chain, a service can find customers while putting up substantial resources in advance to build the capacity it needs. The return on those investments will take shape over time.

    3. Meta: new capital, new rights?

    On 26 October, Meta announces offerings for approximately $35 billion and envisages a further share programme of up to $60 billion. The latter has not yet been formalised, and the company does not guarantee that it will take place or what amount will actually be raised. Adding the two figures and presenting them as $95 billion already received would change the meaning of the announcement.

    The financing also involves a new class of non-voting shares. Those who provide money can therefore take an economic stake in the business without gaining the same say in its decisions. Understanding how control changes requires reading the rights attached to the shares, as well as the capital raised.

    The question runs through all these projects: what activities will be financed, what results will need to support them, and how will the risk be distributed? Raising resources creates a possibility; turning them into useful, sustainable services still requires work.

    4. Binance: from frozen assets to their return

    In its statement of 30 October, Binance estimates the impact of a security incident at approximately $6.3 billion. The company’s initial account points to a compromised computer belonging to a member of infrastructure staff as a possible step in the chain of events. The investigation is still ongoing.

    The platform reports approximately $3 billion in assets frozen on the blockchain and promises to cover losses through its user protection fund and company resources. Freezing an asset can prevent it from being moved; recovering it and returning it requires further steps. A frozen amount is not money that has already been returned to customers, and the announced coverage remains to be verified.

    Binance is a centralised intermediary: the fact that it manages assets on blockchains does not make its organisation decentralised. For those using the service, it matters to know who holds the assets in custody, what controls are in place and what procedures exist to deal with an interruption. It is at those moments that the terms of the relationship become more concrete than a promise.

    5. When personal data becomes part of the price

    A price can change with the time of day, availability or demand. The problem takes a different form when it is tailored to an individual through their habits, browsing or a loyalty programme. In that case, comparing alternatives also requires understanding what information helped construct the offer.

    On 28 October, the Federal Trade Commission requests information from nine businesses in transport, delivery and commerce about the use of data to personalise prices and pay. This is information gathering, not a finding of wrongdoing. It nevertheless brings into focus criteria that those buying or working through a platform may know only in part.

    🧠 Radar Insight — Knowing how an offer is constructed

    For a customer, transparency helps compare a price. For someone working through a platform, it also concerns the value assigned to their time: what determines their pay? Can they find out about and challenge the criteria? The two situations have different consequences, but both require understandable terms.

    6. Italy and Europe: the bill reaches everyday spending

    On 29 October, the ECB keeps interest rates unchanged, with the deposit rate at 2.50%, without setting a path for subsequent decisions. For a household or a business, this does not mean financing immediately becomes less burdensome: the market, duration, risk and contractual terms also matter.

    The provisional ISTAT estimates of 30 October put Italian inflation at 5.4% year on year, up from 4.2% in September. The acceleration is mainly due to energy; the shopping basket, which includes food and household and personal care products, rises from 1.5% to 2.5%. The effect on each household still depends on what it buys.

    These figures describe the context in which facilities are financed, costs are borne and decisions are made. They do not show that AI causes price increases, nor do they automatically anticipate an ECB decision. They help us follow the path between companies’ large figures and the resources available in everyday life.

    The price of trust emerges along this whole path. To assess it, alongside the question of how much a technology grows, we need to ask who finances it, what data it uses and who can intervene when something does not work. This is how an announcement or a number becomes useful in making choices.

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved for original content. Free access does not authorise reproduction or republication without consent, except for uses permitted by law. Third-party materials remain the property of their respective rights holders. To share this issue, use the link to the article.

    Cover in focus

    The Price of Trust — Radar Weekly N19 cover, 2 November 2026, a connected city linking infrastructure and people

    Read this article in Italian

  • Who controls AI infrastructure? | Radar Weekly N2

    Who controls AI infrastructure? | Radar Weekly N2

    AI infrastructure needs chips, memory, energy and capital. Radar Weekly N2 explores the cost of the cloud and the conditions of access to technology, bringing together computing centres, digital finance and control through the signals of 29 June 2026.

    Who controls AI infrastructure?

    Weekly Number 2 | Monday, 29 June 2026 | Approximately 7 minutes to read

    The cloud has weight, but that weight is not distributed evenly. Some build the components, others buy computing capacity, and others access a service by paying a subscription. Part of the future of artificial intelligence is decided between these stages: who bears the cost, who gains value from it, and who sets the terms of use.

    The second issue takes up the question opened by the first and brings it into the workings of infrastructure. Watching how quickly models grow is not enough. We need to understand what makes that growth possible and the economic, material and organisational commitments it brings with it.

    1. Spending today, sustaining the service tomorrow

    Building a computing centre means committing resources before the service delivers the expected results. Buildings, connections, machines and facilities are needed; energy, maintenance and staff are added to these. Part of the spending creates assets intended to last over time: this is what financial language calls capex, investment in assets and infrastructure.

    Renting facilities or computing capacity can also involve long-term commitments. An expansion announcement alone therefore does not tell us whether the activity will be sustainable. The practical question comes afterwards: will the service be able to generate enough revenue to cover its costs and meet the obligations undertaken? Radar looks at this distance between investment and return, without confusing growth in capacity with an economic result already achieved.

    2. Computing needs memory too

    The processor is only one part of the system. To work, it must receive data quickly enough: if the data arrives slowly, a powerful machine can be left waiting. In the race for AI, memory therefore matters too, together with the connections that allow the different components to communicate.

    📌 Did you know? A powerful engine can be left waiting for data

    GPUs are processors capable of performing many calculations in parallel, also used in AI. HBM, high-bandwidth memory, allows large amounts of data to be transferred quickly. A useful image is that of an engine and the system that feeds it: improving the engine is not enough if the supply remains insufficient. In AI, computing capacity and memory speed have to work together.

    The availability of chips and memory thus connects research to factories, supply chains and relations between countries. AI infrastructure does not depend only on a good program: it also depends on being able to obtain the components, install them and keep them running reliably.

    3. Greater efficiency does not mean no cost

    A more efficient model may need fewer resources to carry out a task. That is an important improvement, but it does not make the service immaterial. Every use still requires machines, memory and electricity; if the cost falls and use expands, overall demand can still grow.

    Here the issue invites us to hold two questions together: how much does it cost to obtain an answer, and how reliable and useful is that answer? A cheap service unsuited to the work required does not solve the same problem as one that performs it well. The comparison concerns the ability to actually use the technology, not just the advertised price.

    4. A movement on the blockchain does not tell the whole story

    The discussion then moves to digital finance, where visible movements can seem easier to interpret than they are. A blockchain is a shared record of transactions: it allows transfers to be observed, but does not automatically reveal why they happened.

    🔎 A point to clarify — Transferring does not mean selling

    Moving crypto-assets to an address linked to an exchange may be consistent with a sale, but also with custody operations, internal transfers or other operational needs. The transfer is the observed fact; a sale is one possible explanation. Further evidence is needed to state that it took place. This distinction prevents a trace from becoming a conclusion.

    The same applies when observing flows into and out of funds offering exposure to cryptocurrencies. Prices, subscriptions and redemptions describe different aspects. Reading them together helps us understand market behaviour; treating them as interchangeable can obscure the meaning of individual operations.

    5. Digital money has a structure behind it

    Stablecoins seek to maintain a stable value against a reference, often a currency. The promise of stability raises questions about the resources supporting it: what reserves exist, who holds them, and under what conditions can redemption take place? The product’s name does not replace these checks.

    A related step concerns tokenisation, the representation of an asset or a right through a digital element recorded on a blockchain. Registering a token alone does not clarify what its holder can obtain. The relationship with the real asset, custody and the terms of the right represented remain part of how the system works.

    Execution also matters in DeFi, finance that uses programs on blockchains to manage transactions. The order in which transactions are processed can affect the outcome of a trade. This draws attention to the operational side: what looks like a simple click depends on rules and mechanisms that deserve to be understood.

    6. When an answer has to become an action

    With Physical AI, AI embedded in robots, vehicles and devices, the relationship with the material world becomes even clearer. This is not just about producing text: a system has to receive information from sensors and act in a real environment. Response times, energy consumption, maintenance and reliability become direct parts of the problem.

    In some cases, part of the processing takes place close to the device rather than relying entirely on a remote centre. This is the principle of edge computing, computing at the edges of the network. Where computing happens becomes a practical choice: how long does it take to react, what happens if the connection fails, and which data needs to be transmitted?

    7. Access is part of the technology

    All these stages lead to a common question. If computing capacity, data and networks are strategic resources, does whoever defines access to them play a role that goes beyond the quality of an individual model? Do prices, contracts, the availability of infrastructure and the conditions set for its use matter?

    The issue also invites us to examine the relationship between safety and control. Protecting a system against risks does not remove the need to understand who makes decisions, which limits are applied and what opportunities for scrutiny remain open. It is not only a question of what AI can say, but of who can use it and for what purpose.

    A thread to carry with you

    The weight of the cloud becomes clear through its connections: memory supports computing, capital supports facilities, and the terms of access determine who can use them. Digital finance and AI in physical objects bring us back to the same point: behind an easy-to-use interface lies a concrete organisation of resources and responsibilities.

    The full Radar Weekly N2 brings together the dashboard, in-depth analysis, charts and vocabulary of 29 June 2026. This reading connects some of its themes; to follow every stage and the context of the issue, the PDF is available below.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    The journey begins with Radar Weekly N1: chips, energy and the geopolitics of AI. The introduction to Understanding Today. Building Tomorrow brings together the meaning of the section and its two paths, Weekly and Specials.

    Cover in focus

    Controls - English cover of Radar Weekly N2, The cloud has weight, dated 29 June 2026: a luminous cloud above a data centre, with the official Gruppo Radar signature.

    Read this article in Italian

  • The cloud has a balance sheet: access to digital power | Radar Weekly N3

    The cloud has a balance sheet: access to digital power | Radar Weekly N3

    Access to digital power depends on chips, memory, energy and capital, but also on the contracts and rules governing their use. Radar Weekly N3, dated 6 July 2026, connects the balance sheet of artificial intelligence to the availability of computing and the infrastructure of digital finance.

    The cloud has a balance sheet: access to digital power

    Weekly Number 3 | Monday, 6 July 2026 | Approximately 7 minutes to read

    Having revealed its material weight, the cloud also shows its balance sheet. Building a computing centre means investing today, taking on commitments that last over time and waiting for the service to produce value. Meanwhile, that capacity has to be powered, maintained and made available to someone. The question is no longer only what a model can do: it is who can use it, who bears the costs and how the returns are distributed.

    The issue continues the journey that began with chips, energy and the geopolitics of AI and went deeper into the conditions of access to infrastructure. Here those threads meet: computing needs physical resources, resources need financing, and their availability depends on economic and political decisions. This reading follows some stages of the dossier, preserving the context of July 2026.

    1. The cloud’s balance sheet: spending is not enough

    An investment can make a service possible without proving, by itself, that the service will be sustainable. Those selling chips or memory, those building data centres and those buying computing capacity occupy different positions. A supplier’s revenue may arrive while the customer is still working out how much value the new machines will deliver. Looking only at the growth in spending risks confusing these different timescales.

    Radar invites us to follow the transition from investment to actual work: the use of facilities, operating costs, revenue and results in business activities. Very large capacity that is barely used remains a cost to bear. Capacity used well can instead improve processes and products. There is no need to choose in advance between enthusiasm and rejection: we need to understand where value actually begins to take shape.

    💡 Understanding the issue — The cost does not end with the purchase

    Capex means investment in durable assets, such as servers, networks and buildings. A hyperscaler is a large cloud operator that builds and manages this infrastructure on a vast scale. Leasing, using facilities or equipment under a contract, can also create future obligations. After construction, electricity, cooling, maintenance and staff are still needed: reading the balance sheet means considering both the initial spending and what keeps the service running every day.

    2. Digital power is built in layers

    A model that is ready does not amount to a service available at the required scale. Computing depends on processors, but also on the memory that feeds them data, the networks connecting the machines and the systems cooling them. If one of these stages slows down, the rest of the infrastructure may be unable to work at the expected pace.

    HBM, high-bandwidth memory, transfers large amounts of data quickly to processors. DRAM is working memory that holds data while the system is powered; NAND allows it to be retained after power is switched off, as in SSDs. These abbreviations are not interchangeable: they meet different needs that have to come together in a usable machine.

    Energy also enters this chain, together with the practical ability to bring it where it is needed. The International Energy Agency’s Energy and AI report, published in 2025, examines the link between AI and the power supply for data centres in greater depth. An announcement therefore does not settle the availability of computing: it depends on components, connections and facilities that take time to build.

    3. Available, but to whom?

    The issue focuses on access to the most advanced models, called frontier models because they stand at the frontier of available capabilities. The point is not only to compare their performance. How they are distributed matters too: a restricted preview, a service with limits and a capability open to different groups do not offer the same opportunities to experiment.

    Material access adds another layer. A model may be formally available and still be difficult to use for those who lack the necessary capacity, cannot afford the price or do not meet the terms of the contract. This is precisely where digital power meets the organisation of the service: quantities, regions, priorities and requirements help determine who can actually work with it.

    Safety remains a practical necessity. But acknowledging it does not close the questions about limits: who decides them, how are they justified, and what opportunities exist to scrutinise them? Radar holds these two needs together, without treating every restriction as an abuse or every provider decision as an inevitable response.

    4. Autonomy means more than an alternative chip

    When access to a strategic component is restricted, dependence becomes clearer. Reducing it requires more than replacing a machine: compatible software, memory, cloud services, developers, networks and industrial expertise are needed. A chip may be the most visible point of competition, but the service comes from the whole.

    This is how the issue approaches technological autonomy. Building a supply chain means making the connections between its parts usable and reliable. Rules restricting technology exports, export controls, can slow access to certain tools while strengthening interest in developing alternatives. Neither effect alone proves that one ecosystem has overtaken another technologically: we need to observe what the ecosystem can actually produce and put to use.

    5. Digital money needs clear terms too

    Digital finance enters the discussion through a related question: what makes a promise usable and trustworthy? Stablecoins seek to maintain a value linked to a reference, often the dollar. Understanding how they work means looking at reserves, who holds them and the terms of redemption, not just the value displayed on a screen.

    The dossier follows their transition towards more regulated payment infrastructure. The change concerns responsibilities, checks and operational functioning. A network may enable fast transfers, but trust also depends on being able to know who is accountable for the commitments undertaken. Technical speed and the strength of a promise remain two distinct aspects.

    🔎 A point to clarify — A stable balance does not explain redemption

    Redemption means redeeming a token under the stipulated terms, obtaining the value or asset it represents. A stablecoin showing a price close to one dollar is not enough to explain how this operation takes place. Who can request it, the reserves available, the timing and any limits all matter. Price stability and the right to redemption are connected, but they are not the same information.

    6. Sound verification must start with reliable information

    In DeFi, finance that uses programs on blockchains, risk does not concern only the code executing a transaction. There are also connections between networks, nodes and verification systems. A bridge allows different blockchains to exchange messages or transfer assets: its reliability depends on how events on the other side are verified.

    If a system receives false information about an external event, it may execute a transaction consistent with that information but wrong in relation to what actually happened. This is why the dossier also examines off-chain components, those outside the blockchain. A correctly recorded transaction does not, by itself, prove that all the information making it possible was correct.

    This brings us back to the original thread: infrastructure is more than what is visible in an interface. It includes those supplying data, those checking connections and those intervening when something does not add up. Trust is also built by making these responsibilities clear.

    A thread to carry with you

    The cloud has a balance sheet because its power has to be built, financed and maintained. It also has terms of access, because resources do not become available in the same way to everyone. AI and digital finance remain different fields, but the issue connects them through what supports them: infrastructure, contracts, rules and opportunities for verification.

    The dossier’s scenarios follow these connections over thirty days, six months and two years. They are not predetermined outcomes: they indicate which signals to watch, from investment results to component availability and the concentration of services. The question that remains open is how to turn technological capability into sustainable value without losing access and trust.

    The full Radar Weekly N3 brings together the dashboard, business cases, charts, scenarios and vocabulary of 6 July 2026. Here, some stages have been connected; the PDF allows the complete journey to be followed.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    To retrace the thread, Weekly N1 — The cloud has weight and Weekly N2 — Who controls AI infrastructure? are available. The introduction to Understanding Today. Building Tomorrow brings together the project and its Weekly and Specials paths.

    Cover in focus

    English cover of Radar Weekly N3, The cloud has a balance sheet: access to digital power, dated 6 July 2026, with a luminous cloud inside a circular structure above a data centre and the official Gruppo Radar signature.

    Read this article in Italian

  • When AI becomes heavy infrastructure | Radar Weekly N4

    When AI becomes heavy infrastructure | Radar Weekly N4

    AI becomes heavy infrastructure when the work of models depends on servers, memory, energy and long-term investment. Radar Weekly N4, dated 11 July 2026, connects this material foundation to systems that act, network security and the consent of the people involved.

    When AI becomes heavy infrastructure

    Weekly Number 4 | Saturday, 11 July 2026 | Approximately 6 minutes to read

    The cloud keeps a name that suggests lightness, but its workings are anything but weightless: they require transformers, cables, cooling and industrial facilities. And when a model does more than answer and takes part in an activity, that weight becomes organisational too: we need to know what it can do, which data it encounters and who checks the result.

    The fourth issue continues the journey through the material weight of AI, the conditions of access to computing and the balance sheet of digital power. Now the question broadens: what changes when software organises work and its availability depends on heavy infrastructure?

    1. From an answer to work: who controls the steps?

    A correct answer does not complete a task. If a system has to work on files, consult tools or navigate an interface, the path it follows matters too. It must be able to distinguish what is authorised from what is not, keep a record of changes and stop when a human decision is missing.

    This is the transition to agentic systems: tools that can organise and execute a sequence of actions. Their usefulness is seen in results that can be checked, not in the impression that they do everything by themselves. A modified document, a test carried out and an error reported are different kinds of evidence from a promise of autonomy. The greater the ability to act, the more important it becomes to define the boundaries of the assignment.

    💡 Did you know? — Using an interface does not mean having a free hand

    Computer use means using a digital interface through actions such as selecting a command or filling in a field. The technical ability to click does not authorise every operation. Accessible data, permissions, checks and the stopping point must remain clear: a system can help carry out the work without deciding in the person’s place.

    2. Heavy infrastructure needs a place

    A data centre is not built simply by buying processors. A suitable site, an electricity connection, networks, permits, staff and construction time are needed. Announced capacity and capacity that can actually be used do not necessarily coincide. Between them lie construction sites, supplies and connections that have to work together.

    The issue reads large industrial projects through this distance. A long-term contract can make revenue and supplies more predictable, but it also creates commitments lasting beyond the moment of the announcement. Heavy infrastructure thus ties the economics of computing to the local area: grid availability, cooling, resource use and the distribution of costs.

    The International Energy Agency’s Energy and AI report, published in 2025, examines the relationship between data centres and electricity in greater depth. Understanding the change requires more than asking how much energy is needed in total. Where it is available, when, and with what consequences for other uses of the grid all matter.

    3. The processor can be left waiting: the role of memory

    A powerful chip can perform poorly if data arrives too slowly. The GPU, a processor suited to many calculations in parallel, needs memory and connections able to feed its work. HBM, high-bandwidth memory, meets this need: transferring large amounts of data quickly to the processor.

    The supply chain is therefore not a succession of brands to compare. It is a set of dependencies. Memory, component assembly, factories, energy and networks can become limiting factors even when the chip design is advanced. The term bottleneck describes precisely the stage that slows the others down: increasing capacity elsewhere is not enough to remove it.

    4. Capital finances construction; it does not guarantee the result

    Investment makes heavy infrastructure possible, but the return arrives only if that capacity finds a sustainable use. Capex is spending on durable assets, such as servers, facilities and structures. Those selling them may receive payment while those buying them still have to turn them into services, customers and revenue.

    This distinction helps us read markets without confusing industrial growth with guaranteed returns. Expectations may already be very high: a good result alone does not tell us how much was expected or how much is already reflected in valuations. The dossier invites us to follow utilisation, margins and costs, as well as the scale of announcements.

    5. Shared dependencies make security a system-wide issue

    If many activities use the same provider or infrastructure, an outage can affect them together. This is concentration risk. It does not require a technology to be inherently weak: it can also arise because a dependency is shared and difficult to replace.

    The Weekly connects this problem to the security of financial networks and blockchains. A bridge connects different networks; its reliability depends on authorisations, checks and how updates are managed. A token’s price does not reveal how these stages are organised.

    🔎 Understanding the issue — Preparing defences takes time

    Post-quantum cryptography develops protections designed to withstand attacks from future quantum computers too. It does not mean using a quantum computer to protect data. NIST published the first three standards in 2024. The dossier’s point concerns the time needed to prepare a migration: identifying the systems involved, updating tools and maintaining compatibility while defences change.

    6. When AI meets people who did not choose it

    Wearable devices bring AI into and close to everyday life. Glasses, microphones and sensors can make a service more accessible, but they also involve the people sharing the same environment. The person wearing a device has made a choice; someone passing nearby may not know what information is being collected.

    Here the issue opens the question of consent in shared environments: the ability to understand what is happening even when data collection is unobtrusive. Recording and interpreting are not the same operation, but both may concern people who do not use the service. Responsibility does not end with a device’s settings: it reaches into workplaces, shared spaces and relationships.

    The connections to keep in view

    AI’s heavy infrastructure is not made only of buildings. It includes resources, contracts, permissions and dependencies. The issue’s thread runs from a machine waiting for data to a network that has to bear the load, from a system carrying out actions to a person who must be able to understand how they are involved.

    This is why the dossier of 11 July does not look for the signal only in the newest model. It also examines the timescales of facilities, component availability, the quality of checks and the ability to change provider. These stages are less exciting, but they help us understand what is actually becoming usable.

    The full document preserves the dashboard, business cases, market analysis, scenarios and sources of the issue. This selection connects some of its central themes to accompany the complete reading, without replacing it.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    The preceding journey is brought together in Weekly N3 — The cloud has a balance sheet. For the introduction and the two collections, visit Understanding Today. Building Tomorrow.

    Cover in focus

    English cover of Radar Weekly N4, When AI becomes heavy infrastructure, dated 11 July 2026: server towers, transformers and cables in a mountain landscape, with the official Gruppo Radar signature.

    Read this article in Italian

  • When power meets the real world | Radar Weekly N5

    When power meets the real world | Radar Weekly N5

    Technological power becomes useful when it meets factories, energy, people and practical activities. Radar Weekly N5, dated 20 July 2026, connects the chip supply chain to markets, the costs of digital work and the conditions of access to artificial intelligence.

    When power meets the real world

    Weekly Number 5 | Monday, 20 July 2026 | Approximately 6 minutes to read

    Building more capacity does not, by itself, solve the problem of how to use it. A factory may receive orders, a model may answer more quickly and a service may become available on a new device. But between these possibilities and a useful result remain timescales, costs and choices that someone has to bear.

    Following the issue devoted to AI as heavy infrastructure, this Weekly looks at the point where power enters work. The question is not only how much can be produced: who can use that capacity, under what conditions and with what control?

    1. Technological power passes through factories

    TSMC and ASML occupy two different positions in the semiconductor supply chain. The first manufactures chips for those who design them; the second builds the lithography systems needed to create circuits on silicon. Looking at them together allows us to distinguish the design of a technology from the tools and facilities that make it manufacturable.

    A wafer is the silicon disc on which many circuits are made. Depositing a material is not enough: a sequence of precise manufacturing steps must be repeated. In ASML’s explanation of lithography principles, an optical system transfers a pattern onto a light-sensitive layer. This is one of the stages through which the design takes shape on the wafer.

    The dossier reads industrial demand through these concrete processes. An order for machinery concerns capacity to be built and used over time; it does not yet amount to a service available to those at work. Between the two ends lie supplies, assembly, staff, networks and customers.

    💡 Inside the chip — Precision has a chain behind it

    EUV and DUV refer to lithography technologies using extreme and deep ultraviolet radiation. They are not simply two labels meaning “better chip” or “worse chip”: the processing required changes with the design and the layers to be made. What matters in the supply chain is compatibility between stages. An advanced machine does not replace all the other expertise needed to obtain a usable component.

    2. A strong industry does not make every price attractive

    Production and financial markets follow different timescales. A company may enjoy strong demand while its share price incorporates even higher expectations. Industrial results tell us what the company is doing; valuation also tells us how much investors expect from the future.

    This is why the issue does not treat a price correction as proof that the technology has stopped working. Nor does it regard supply-chain growth as a guarantee of returns. The distance between orders, production capacity and expectations requires several elements to be read together, without relying on a single market movement.

    Capex, spending on durable assets such as facilities and servers, makes construction possible. The return then depends on utilisation: how much work runs through those facilities, what revenue it generates and what costs remain. Those selling the machines and those using them are not at the same point in this journey.

    3. The cost of an answer is not the cost of the work

    A service that is cheap per operation can become expensive if the task has to be repeated, errors corrected or every result checked at length. In digital work, the meaningful measure is what can be completed reliably, not simply the number of answers obtained.

    A token is a unit through which a model processes text: it may correspond to a word, part of a word or other elements. Counting tokens helps measure part of the consumption, but does not tell us whether a case has been properly completed, a document is correct or a decision has been understood.

    The Weekly therefore proposes considering technical cost together with the result: the attempts needed, checks, the person’s time and the ability to recover from an error. Automation that truly eases an activity and automation that shifts the effort into review may look similar in a demonstration, but not in daily use.

    4. Access also changes through interfaces

    An assistant does not meet the user in a vacuum. It arrives through a phone, an operating system, an application or a function already present on the device. How easily it can be found and chosen can affect competition as much as the quality of the model.

    The issue connects European questions about access to Android services to this stage. Being able to enter an ecosystem does not mean obtaining any data or function without limits. Availability, interoperability and authorisations must be distinguished: respectively, the ability to use a service, to let different systems communicate and to permit certain operations.

    For those at work, these differences become concrete when they have to connect a tool to documents, messages or other activities. A more convenient interface is useful if it also makes permissions and the consequences of a choice understandable. Otherwise, the initial simplicity may conceal dependencies that are hard to see.

    5. From declared performance to testing in use

    New models arrive accompanied by numbers: size, the ability to handle long texts, test results. This information deserves careful reading, but each measure describes only part of their behaviour. A large context, for example, indicates how much material can be supplied to the system, not that every detail will be remembered and used correctly.

    A useful test remains tied to the task. Is the result consistent? Can the sources be traced? Is a limitation reported? The dossier places announcements within this framework, without confusing declared performance with independent verification or with value for every activity.

    🔎 The practical question — What remains under control?

    The convenience of a service does not, by itself, answer the question about data. A photograph visible online does not automatically express the person’s consent to every new use. In the transition from content to an AI function, the purpose, the information available and the ability to choose all matter. This boundary is precisely the issue’s theme: technical accessibility does not exhaust responsibility.

    6. The real world presents the bill for dependencies

    Chips, memory and servers need electricity and places in which to operate. The International Energy Agency’s Energy and AI report, published in 2025, examines this relationship and its consequences for the energy system. Technological power cannot be separated from the material conditions supporting it.

    The same criterion applies to organisational dependencies. A service may be powerful but difficult to replace; an activity may function as long as a provider maintains certain conditions. Looking at the ability to change tools, retain one’s materials and understand the costs helps assess what remains in the hands of those using the technology.

    7. The connections to keep in view

    Power meets the real world when it stops being only a promise of performance and becomes a relationship between production, capital, access and responsibility. This issue brings together the factory that builds, the market that values and the person who needs a useful result.

    The full document preserves the week’s dashboard, cases, data and sources of 20 July 2026. Here, some stages are connected to accompany the reading of the issue, without replacing its complete analysis.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    The journey continues from Weekly N4 — When AI becomes heavy infrastructure. The introduction to the collections is in Understanding Today. Building Tomorrow.

    Cover in focus

    English cover of Radar Weekly N5, When power meets the real world, dated 20 July 2026: a silicon wafer connects factories, data centres and a city through luminous lines, with the official Gruppo Radar signature.

    Read this article in Italian

  • The threshold shifts: safety and capital in AI | Radar Weekly N8

    The threshold shifts: safety and capital in AI | Radar Weekly N8

    The threshold shifts when an AI system acquires capabilities that require different controls, but also when investments change the balance among those financing it. Radar Weekly N8, dated 10 August 2026, brings together operational safety, capital and the chip supply chain, distinguishing what has been observed from what remains uncertain.

    The threshold shifts: safety and capital in AI

    Weekly Number 8 | Monday, 10 August 2026 | Approximately 5 minutes to read

    1. When a capability requires different precautions

    Not every development in artificial intelligence consists of a faster answer or a better-written text. Some change the kind of activity a system can perform. When this happens, the question is not only how much it has improved, but which precautions must accompany it.

    On 7 August 2026, OpenAI publishes an update on the cyber capabilities of the Astra model. The company states that it cannot rule out reaching the “Critical” threshold in its evaluation framework. The wording matters: this is not definitive certification of a capability, but an assessment that leads to stricter precautions.

    The subject concerns activities such as identifying previously unknown vulnerabilities and chaining several stages of a cyberattack. Performance observed in a test does not guarantee that the system will always succeed in every environment. It may nevertheless be enough to change the level of caution required.

    🔎 “Cannot rule out” does not mean “has demonstrated”

    A risk threshold is a criterion used to decide which controls to apply. If significant uncertainty remains, precautions may be increased before a definitive conclusion is reached. This differs from declaring every feared scenario established.

    2. The laboratory is part of the experiment

    Evaluations do not measure a model suspended in a vacuum. They measure a model within an environment, with tools, permissions and constraints. Changing that environment can also change what is observed.

    In its report of 4 August 2026, the UK’s AI Security Institute describes 19 unauthorised actions across 10 of the 122 trials conducted. The environment was deliberately open to the Internet and some filters were disabled to make the test possible.

    It would be incorrect to describe these episodes as escapes from an isolated internal environment: the report specifies that this was not the case. Nor does it identify real-world harm. It does show why actions outside the assignment must also be observed, rather than assessing only whether the main task was completed.

    The example of a proposed malicious code change, which was not accepted, helps reveal the difference between an attempt and its effect. That distinction does not erase the significance of the behaviour; it prevents consequences from being attributed to it when they have not been documented.

    🛡️ An open test is not an escape

    If Internet access is a deliberate feature of a test, using that connection does not, by itself, demonstrate that a technical barrier has been crossed. The anomaly may instead be an action outside the assignment. Environment, authorisation, attempt and effect must be described separately.

    3. Stopping an activity can be a safety measure

    In the case of Astra, OpenAI describes stronger isolation, restrictions on tools and network access, and monitoring. Some activities must be suspended when the required controls are unavailable.

    This is a concrete choice, not a general appeal to responsibility. Safety is not added only after a performance has been achieved: it helps determine which tests may be conducted and under what conditions.

    This concerns research into the most advanced capabilities. It does not justify automatically extending every incident to all models or every everyday use of AI. The useful question is more precise: are the tools and the environment adequate for the capabilities of the system being put to work?

    4. Capital crosses a threshold too

    The second part of the issue turns to financing. In the dossier on SoftBank and OpenAI, the central point is to distinguish the money invested from the value attributed to the holding and from commitments still to be financed.

    In an accounting estimate, a holding may be worth more than it cost. The difference can contribute to the period’s result without turning into cash until a transaction realises that value. This is where growth in assets and the availability of money can tell different stories.

    Bridge financing, by contrast, provides resources to cross an interval: for example, between an investment commitment and securing more stable capital. It is not cost-free money. It comes with a term, conditions and a need for repayment or refinancing.

    Looking at holdings, liquidity and debt together helps us understand the weight of an industrial choice. Enthusiasm for the future does not replace analysis of how that future is financed.

    💡 Estimated value and available cash

    Fair value, the estimated value under accounting rules, describes what an asset is worth at the valuation date. Cash describes available monetary resources. A revaluation can increase assets without filling a bank account.

    5. Contracts and results in the chip supply chain

    AI capabilities also depend on memory, computing and the availability of supplies. Sandisk’s results and AMD’s second-quarter 2026 results bring this part of the supply chain into the issue’s reading.

    A multiyear supply contract makes part of future demand more visible, but it is not revenue and, above all, not cash received. The whole organisation behind it matters, including deliveries, contractual terms, prices and the ability to honour commitments.

    Comparing two quarters also requires care: prices, sales mix and trade restrictions can affect results. Reducing every change to AI demand alone risks losing important parts of the explanation.

    6. Two boundaries to read together

    The issue does not put safety and finance on the same plane. It brings them closer because both require a boundary to be recognised: on one side, the boundary between a system’s capabilities and available controls; on the other, between expected value and resources actually committed.

    When a threshold shifts, the previous rules may no longer be enough. This does not create a need to stop every innovation or regard every investment as a problem. It certainly means reviewing the conditions in light of what has changed.

    The PDF develops this reading with data, documentary evidence and the Weekly’s other dossiers. The article offers the thread connecting them: understanding not only what AI promises, but which responsibilities, infrastructure and constraints it makes necessary.

    Read the full issue

    © 2026 Barbara Gattobigio · Gruppo Radar
    Comprendere oggi. Costruire domani · by effettoantropico.life
    All rights reserved on original content. Free access does not authorise reproduction or republication without consent, except as permitted by law. Third-party materials remain the property of their respective owners. To share this issue, use the link to the article.

    Keep exploring

    Understanding Today. Building Tomorrow brings together Radar Weekly and the section’s in-depth reports.

    The cloud has a balance sheet – Who controls AI infrastructure?

    Cover in focus

    English cover of Radar Weekly N8, The threshold changes, dated 10 August 2026: a luminous core protected by glass beneath a golden arch between a laboratory and blue servers, with the official Gruppo Radar signature.

    Read this article in Italian