Move over, Copilot: Microsoft is introducing a new category of agentic AI called “Autopilot,” starting with Scout, its first agent. And it doesn’t take much guessing to understand how Microsoft expects these things to operate: By constantly watching your every move and taking action in the background to ostensibly streamline your workday. Microsoft announced Autopilot, and the first Autopilot agent, Scout, at Microsoft Build on Tuesday, describing it and other future Autopilots as “always-on agents that work autonomously,” stay active in the background to “understand how work gets done across your apps and systems,” and can “take action without needing to be prompted each time.” Scout, for example, can be interacted with in Teams when one feels the need, but outside of instances when users need to query it directly, it’s always there. “It operates across cloud, desktop, and web, connecting to Teams, Outlook, OneDrive, and SharePoint, and to the data that powers your day, including chats, email, calendar, and contacts,” Omar Shahine, corporate VP of Microsoft Scout, wrote in the announcement. Autopilot agents supposedly have their own identities, according to Shahine, and are able to act autonomously within the constraints organizations set on their activities (access controls can be set by organizations). Per Shahine, letting Autopilots operate on autopilot “creates a more durable way to keep work in motion so it continues even when your attention is elsewhere.” Say, for example, you need to schedule a meeting: Scout can handle scheduling on its own while accounting for time zones; it can flag meetings it considers particularly important for its users and generate materials it believes users need to prepare before the scheduled time. Scout can also identify looming deadlines and block off time on a user’s calendar so that they can work on a particular project, “spot risks, like stalled decisions,” and basically act like a work nanny that schedules your day by being hyper-aware of every single little thing that needs to get done. Hopefully, your new Microsoft nannybot is more reliable than its Copilot predecessor, whose outputs Microsoft itself warns may not always be accurate. Get ready for a Claw-shaped hole in your environment “Microsoft Scout is built with enterprise-grade security and controls so it can be trusted in your organization from day one,” Shahine noted in the release, followed immediately by noting that it’s powered by OpenClaw, not exactly a platform with a stellar security reputation or record of not making bad decisions on behalf of users. Microsoft claims that Scout and whatever future Autopilot agents it releases are bound to an Entra identity that allows their activity within an enterprise environment to be attributed to a particular person’s Scout agent, and notes that it acts within the confines of access controls set by the organization, but it’s not clear what other protections against common AI exploits are included. As we’ve noted before, it’s often surprisingly easy to manipulate AI agents into behaving in ways their operators never intended, and malicious webpages can inject prompts that trick them into leaking sensitive information; in both cases, those sorts of attacks can be launched without any direct user interaction. We asked Microsoft for more details on the security aspect of Autopilots and Scout, but didn’t hear back before the deadline. It’s also worth noting that Microsoft Scout is in very limited access, with only a “select group of customers” getting access to the preview, along with organizations participating in the Frontier program, which grants them early access to Copilot and other Microsoft AI features. One more caveat, too: Frontier enrollees can only get access to the Scout preview if they’re GitHub Copilot subscribers. GitHub Copilot recently shifted to a usage-based billing model that has seen bills skyrocket, so expect those Microsoft bills to rise if you choose to give it a shot, too. ®

Intel is keen to reassure investors that its troubles with the 18A manufacturing process were a one-off, and that it is better positioned to capitalize on what it expects will be growing demand for CPUs used in AI inference workloads. Speaking at the Bank of America 2026 Global Technology Conference in San Francisco, Chipzilla’s chief financial officer David Zinsner claimed that the firm simply bit off more than it could chew in trying to move too fast with the new process node. “I would say it this way, I don’t know, early last year, I think the challenge around 18A was two things. One, we tried to do too much at once. And it took a while to get that settled. And I think second is, we were trying to play performance and yield and trying to improve both at the same time. It was like trying to fly the plane and fix the wing at the same time, basically,” he said. Intel 18A – its angstrom-era process, marketed as a 1.8 nm-class node – was initially expected to be production-ready by late 2024 and ramp toward volume manufacturing in 2025. However, the technology ran into delays, with the first products built on 18A not arriving until Intel unveiled its Core Ultra Series 3 CPUs back in January this year. Zinsner said that after Pat Gelsinger’s departure, when he and Michelle Johnston Holthaus took over as interim co-CEOs, he put Intel global operations chief Naga Chandrasekaran on the case, “and then they really just focused on first, stabilizing performance. And so they stabilize performance. Then once you’ve got your performance stabilized, then all you do is you work yield every month,” he explained. “The second thing that we did when Lip-Bu joined is we really opened up our data to our vendors to really help us learn things that we could do to improve yield and that made a dramatic difference,” Zinsner added. This meant overcoming some cultural resistance to sharing data, he claimed, but then “Once we fixed that, we really started to get some feedback into what we could do to improve. And then it was just our team just grinding it out every month.” Intel’s goal is now to get to yields that generate great margins, and the firm is now ahead of its schedule to get there by the end of 2027, he claimed. And when it comes to the next-generation 14A process, the one that Intel hopes will allow it to set up its foundry division as a contract manufacturing business as well as making its own chips, Zinsner was keen to stress that the program remains on track. “Now I would just say we have a more aggressive plan for 14A than 18A. When you look at kind of yield and performance measures at this point in time and maturity of 14A compared to that same moment in time for 18A, we’re ahead,” he claimed. “All the stuff that I said that we bit off more than we can chew on 18A, and it really took some time. Now it’s just a little bit of a rinse and repeat. I mean it will be a lot easier to do 14A because it’s just using a lot of the gate-all-around and backside power and so forth that we implemented in 18A,” Zinsner explained. As Intel chief Lip-Bu Tan explained a couple of weeks ago, the firm is now anticipating increasing demand for CPUs as the focus of the AI craze turns from training to inferencing work. Zinsner said that it is hard to judge exactly how big the growth in CPU demand would get, but “I think it’s going to be a big market.” “If you just stamped something and called it a CPU right now, it probably would sell. So in the near term, it’s all about supply,” he claimed. “I mean we’ve got enough demand out there that if we can do a good job executing on the ramping of supply, we should have no issue with growing our revenue meaningfully in the datacenter space,” he added. Zinsner also said that Intel was looking to draw up more long-term agreements with customers in the future. “So we’re locking in a price, for sure. We’re locking in a volume commitment. And then that enables us to do a better job of planning out our capacity and making sure when we’re investing in capacity, we’re going to see customers take that supply when it comes off the line,” he said. Intel this week unveiled its Clearwater Forest Xeon chips, along with more details of its upcoming Diamond Rapids Xeons, at the Computex trade show in Taiwan. ®

A body that represents mobile operators wants the migration to 6G networks to be as smooth as possible, learning lessons from the fractious 5G introduction that has left countries like the UK with a less than satisfactory service. The Next Generation Mobile Networks Alliance (NGMN) says that 6G requires a different standardization approach in order to prevent complexity and market confusion, alongside a smooth and cost-effective migration path for its members. What exactly defines 6G is still being thrashed out, but it is expected to be ready by the end of the decade. According to telecoms supplier Ericsson, 6G networks are likely to offer data rates of several hundred gigabits per second (Gbps) with sub-millisecond (ms) end-to-end latency, and usher in new use cases. But NGMN sees it as an opportunity to simplify network architectures, reduce long‑term costs and operational complexity, and ensure a smooth and scalable migration path. As it points out, deploying a new technology requires significant investment, and this needs to be justified by confidence it will deliver a sustainable return for the operators. The org has pushed out two reports ahead of a plenary meeting of the 3GPP standards consortium in Singapore this month. One looks at 6G architecture and migration options, while the other considers the timing of 6G’s introduction from an operator’s perspective. What the NGMN wants to see is consensus on a primary approach to 6G migration and reduction in complexity across user equipment (UE), the radio access network (RAN) and core networks. It also wants to see the required 6G specifications, including those for RAN and core network, delivered in a single drop of 3GPP Release 21 rather than pushed out piecemeal. This is to enable operators to perform a complete network rollout without multiple phases that result in unnecessary complexity and market confusion, it says. In its first document, the NGMN advocates for the use of Multi-RAT Spectrum sharing (MRSS) as a migration option, where RAT means radio access technology. This enables the simultaneous use of the same frequency band by more than one generation of cellular network, such as 5G and 6G. This will 6G with flexible access to 5G spectrum so that “competitive user throughput and performance” can be achieved, even in locations where a large amount of new spectrum (e.g. spectrum around 7 GHz) is unavailable or too costly to deploy, it claims. However, the 3GPP should give also consideration to alternatives such as Dual Connectivity and Dual Stack, in case MRSS is found to significantly reduce 5G performance or increase network costs. As for the operators’ expectations of 6G, the second document says that a key motivation is to evolve network core technology to deliver greater operational efficiency. This extends to more efficient use of new spectrum bands (6-7 GHz considered possible), network automation, AI as a service, energy efficiency, and delivering ubiquitous coverage. The value to end users and the cost of network deployment are driven for a significant part by the design choices made in standardization, and this is why a single drop of specifications is key. With 5G, the full promise of the technology could not be delivered with initial deployments, and multiple rollouts and device generations have been needed. In the UK, for example, network operators were forced to bolt 5G radios onto the existing infrastructure built for 4G, which meant early users did not perceive much improvement in service, as The Register wrote last year. This led to the impression that it wasn’t worth paying extra for, which sapped the networks of funding needed to invest in upgrades later. However, the ability to decouple investment in software from simultaneous investment in hardware for 6G is a key operator expectation, according to the NGMN. If they can deploy 6G by means of software upgrades in the 5G legacy frequency bands, it will limit the required 6G investment, and will facilitate faster 6G rollouts. Conversely, 6G deployments may be subject to major delays if operators have to face infrastructure renewals and software upgrades at the same time. Another factor is the availability of new spectrum. For 6G, this will be instrumental for new use cases requiring extra capacity. The GSMA said in a report last year that 6G networks will need up to three times the spectrum currently allocated for 5G, and was measuring up various mid-band frequencies, as well as some in the centimeter wave bands. Overall, it seems the NGMN wants the standards bodies to take their time and get it right, before any rollout of 6G technology is even considered. “It is critical to take the time necessary for producing standards ensuring the above requirements, learning the lessons of 5G-SA deployments, and not to rush into decisions having potentially detrimental impact on the industry,” the document states. Extending the completion date of 3GPP Release 21 should even be considered if such a risk is identified, the NGMN adds. “The transition to 6G will present significant opportunities, but only if the industry prioritizes migration paths that build on existing network assets, minimize operational complexity and deliver tangible benefits from the earliest deployment stages,” said NGMN Alliance board chairman and Orange Group CTO Laurent Leboucher. “Dedicating sufficient time to this process is crucial, otherwise risking unnecessary complexity and long-term challenges, limiting the value to operators and end users.” ®

COMPUTEX 2026 The sun is slowly but surely setting on copper interconnects, Marvell CEO Matt Murphy claimed in his Computex keynote this week. Within the next decade the IP house expects photons to take the place of electrons and change the way datacenters are built and run in the process. And, if Nvidia CEO Jensen Huang is right, the widespread transition to silicon photonics technologies could make Marvell the next trillion dollar company. With a market cap of $191 billion, Marvell still has a long way to go, not that stopped Wall Street investors from sending the company’s share price on a 30 percent rally on the proclamation. However, Huang’s prediction, made during Marvell’s Computex keynote this week, may be more than flattery. The large-scale deployment of AI infrastructure for training, inference, and agentic systems is already reshaping datacenter networks and pushing copper interconnects to the limit. “The distance a signal can travel over a copper cable is inversely proportional to the bandwidth, so every time you double the bandwidth, you have to cut the distance in half,” Murphy explained. Today the fastest network interconnects operate at 200 Gbps per lane, but at these speeds copper cables can only carry a signal about 2.5 meters, effectively limiting interconnects. With the launch of its next-gen NVSwitch silicon in its Vera Rubin platform, Nvidia will double this again to 400 Gbps, halving copper’s reach once again. There’s a reason the NVL72’s switches are located in the middle of the rack. “Going forward, even the connections within the rack will become optical,” Murphy said. “The whole industry knows this is coming. So, we’ve been preparing for this moment, not just Marvell, but the industry.” Optics offer much greater reach, but the tech isn’t without compromise. Pluggable optics are not only power hungry but they also fail. Power consumption is one of the reasons why Nvidia first revealed its NVL72 rack systems, Huang explained that using optics would have added another 20 kilowatts to the system’s then monstrous 120 kilowatt load. “You use optics wherever you must, you use copper wherever you can,” Huang said during Marvell’s keynote. While Huang expects copper interconnects to remain relevant for a while longer, Marvell is preparing for a future in which even PCB traces will be replaced by fiber optic cables. In 2020 Marvell acquired Inphi, which specialized in building optoelectrical interconnects, and more recently the company dropped billions to acquire Celestial AI’s silicon photonics interconnect tech. Then in March, Nvidia invested $2 billion in Marvell to, among other things, advance its silicon photonics interconnect tech. “We build optical modules that contain all the electronics needed to drive and modulate the laser and transmit data over long distances,” Murphy said. At copper’s end “Think about 10 years in the future and it’s a world where a lot of the copper connections are gone,” Murphy said. “This is a world where then distance doesn’t matter… that’s a profound change.” All modern datacenter infrastructure and software has been designed around the constraints of distances. “With optics, distance doesn’t matter. So now we can change the size of the scale up domain from 72 or 144 XPUs or GPUs to 1,000 or more, all optically interconnected,” he said. “The implications for workloads are enormous.” But it’s not just GPUs. Murphy explains that when everything from CPUs and GPUs to memory and storage are optically interconnected, they will no longer need to be in the same box. “Modern AI servers are composed of a certain number of CPUs, XPUs, memory, and network interfaces, and the reason they’re all on the same system is because of distance,“ he explained. “Imagine a completely disaggregated architecture, XPUs in one system, memory in another, agentic CPUs in another.” This means these resources can be reconfigured on the fly to achieve the ideal ratio of CPU to GPUs to system memory for a specific workload. Google is already doing this to a lesser extent with its TPU clouds. While the ratio of CPUs and memory to GPUs can’t be reconfigured on the fly, the use of optical circuit switches means the number and shape (topology) of Google’s TPUs can be adjusted to maximize inference or training performance. This also has implications beyond AI. Even if the bubble collapses and AI infrastructure demand evaporates, one can imagine AWS and other major cloud providers using silicon photonics or co-packaged optics to disaggregate compute resources and then reassemble them a la carte. Battling Broadzilla Marvell is a long way from a trillion dollar market cap and getting there assumes a certain other IP house doesn’t eat their lunch. Broadcom, whose market cap already surpasses $2 trillion, and whose customers include some of the largest hyperscalers in the world, including Google and Meta, has also been amassing a broad portfolio of silicon photonics and optics tech over the past several years. These technologies include co-packaged optics for switches and XPUs, as well as DSPs for high bandwidth pluggables. Much like Murphy, Broadcom’s CPU Hock Tan expects that photonics will replace most copper interconnects eventually, just not tomorrow. “I can see a point in time in the future when it matters as the only way to do it,” Tan told analysts late last year. “we are not quite there yet.” “The final, final, straw is when you can’t do it well in pluggable optics,” Tan said. “Then you go to silicon photonics.” ®