Mi&s Weekly Analyst Insights — Week Ending October 16, 2026
Welcome to this edition of our Weekly Analyst Insights roundup, which features key insights that our analysts have developed based on the past week’s events.
Issues of sovereignty in tech — whether we’re talking about digital sovereignty, data sovereignty, sovereign AI, or something else — are more prevalent than ever in the age of AI. And the growing presence of agentic AI only makes them more complicated. No surprise, then, that I and several of the other analysts in the firm have talked and written so much about sovereignty this year.
Mike Leone has been addressing this with an emphasis on the places where data and AI intersect, for example in a piece from May on the launch of IBM Sovereign Core. Last week, he wrote a Forbes article in which he shared some broader thoughts about how sovereignty comes to bear on IT decision-making — or how it should — when so many agents are making decisions of their own. Fair warning: If you take Mike’s advice to heart, it may mean some detailed homework inside your organization, and some candid conversations with your AI model suppliers. But you may sleep better at night once that work is done.
Last week, Anshel traveled to Atlanta for the Augmented Enterprise Summit, while Bill Curtis was in New Orleans for the Thread Group members meeting. This week, three of us are visiting different parts of the country: Bill is in Silicon Valley for a Synaptics analyst event, Mike is in Florida for the Couchbase Catalyst conference, and I am in D.C. for the Google Public Sector Summit (last year’s version of which I found highly interesting). Next week it’s even busier, with five of us traveling for events hosted by six different companies in five different states. (You can find our planned attendance at industry gatherings on the events section of our site.) As always, we continue to publish fresh content across different formats, so be sure to look for our insights and research — and be sure to follow us on X and LinkedIn.
Also last week, I appeared on Yahoo Finance to discuss a variety of developments in the AI market, including the latest on CoreWeave and Anthropic. During the week, Moor Insights & Strategy analysts’ perspectives appeared in business and technology outlets including Fortune, CIO, Reuters, CNBC, Fierce Network, TechTarget, The Motley Fool, Tech Times, Data Center Knowledge, and 24/7 Wall St. Topics included SpaceX’s spectrum deal and wireless ambitions; Apple’s new smart home devices; a proposed deal among among ByteDane, NVIDIA, and MediaTek to finance AI chips; Microsoft’s new Maia 300 AI Chip; AMD Helion; IonQ’s planned acquisition of Elisity; ServiceNow’s new agentic AI products; Bigeye’s Agent Trust Hub; and Workday’s new AI products.
Our MI&S team also published 14 deliverables — 1 Forbes Article, 2 Research Papers, 1 Research Note, 1 Analyst Insight, 7 Field Notes, and 2 Podcasts.
Check out this week’s Analyst Insights roundup for more from the MI&S team, including what’s top of mind for each analyst, our thoughts on vendor announcements, press quotes, and more.
Have a great week!
Patrick Moorhead
MI&S Analyst Insights
ServiceNow Stakes a Claim in Agentic Governance
ServiceNow has made some major announcements with respect to securing and managing agents that get it back into the agentic platform conversation. We keep hearing about more incidents where agents are doing unpredictable — and in some cases dangerous — things. And that is why we are also seeing vendors working hard to create adequate measures to prevent agentic incidents. Earlier this year, hyperscalers were leading the way, but now we are also seeing solid efforts from SaaS vendors. Both the SaaS players and the hyperscalers are also emphatically stating that these new solutions are cross-platform, to head off past criticisms of these types of tools. And while ServiceNow was quieter with its AI innovations earlier this year, it has since launched Flow, which was covered by my colleague Mike Leone, and now these governance improvements.
For security and policy enforcement, ServiceNow announced AI Passport, which is a runtime authority layer for AI agents and MCP servers. ServiceNow says it is built on open specifications so other governance products, identity providers, and enforcement points can issue and verify credentials, including for third-party agents, and so credentials are verifiable at every step of a multi-agent workflow.
On the governance side of things, the new Super Intelligence Control Tower adds shadow AI detection for unapproved agents, 400-plus out-of-box connectors for discovery across platforms and vendors, access mapping against real entitlements, and continuous observability of agent traces. A cost layer adds cost governance by user, department, and runtime; value intelligence that ties consumption to quality and realized value; AI Cost Management inside IT Asset Management (frontier model, embedded SaaS AI, hyperscaler, and GPU costs in one view); and graduated alert, throttle, and block policies. EmployeeWorks lets non-coding employees build productivity assistants within admin guardrails (HR Service Delivery first), schedule productivity routines that run in the background, and use ToolSmith, an agentic developer environment that builds tools from natural-language descriptions, with a Conversations API to embed all of it in other apps.
AWS Continues Its Steady Stream of New Kiro Developer Features
Last month we saw AWS make some interesting announcements. For example, it announced a developer on-ramp capability designed to help new developers get started with Kiro and AWS. This was meant to target users of AI-driven developer tools like Lovable and Cursor. AWS also announced a more robust developer workflow capability to help devs wire agents and processes together, geared towards more advanced developer toolsets and the orchestration capabilities we are seeing emerge from SaaS providers. Now the Kiro team is taking steps towards providing more secure local development and testing with its new sandbox capability. This stands to be very helpful because, when developers run agents locally (as many Kiro users do), there is a risk of an agent using local files or configs and creating havoc for the developer or the ops team when the code gets pushed into the cloud. In this case, the secure sandbox segregates the desktop from the agent, from where the code is built and tested. But what could be more interesting is the idea of being able to run multiple sandboxes locally, which would be a big help to developers in the services arena or developers in highly regulated environments. Right now, this is not a documented Kiro capability, but it appears to be possible, since other solutions like Docker CLI can do this.
Oracle Fusion Claw Aims to Improve Enterprise Agentic Execution
Oracle’s new Fusion Claw is another sign that the agent conversation is moving from assistants to harnesses built for long-running, autonomous work. Oracle describes Fusion Claw as a governed agentic execution runtime that pairs AI reasoning with deterministic enterprise computation, and it is launching with 25 Claw-powered agentic applications.
The design choice I find most interesting is the division of labor. A frontier model handles reasoning, planning and adaptation, then hands off to deterministic computation for high-volume execution, so expensive model calls are used only where intelligence is actually needed. Governance is equally central. An “Outcome Trust Harness” enforces each organization’s policies, permissions, and approval boundaries on every run, and an “Outcome Receipt” leaves an auditable record of the decisions made and transactions executed. Customers can set each process anywhere from light assistance to fully automated execution within delegated authority. This lines up with what I wrote in last week’s newsletter about Microsoft’s new Copilot client, which includes Autopilot, an always-on agent that runs long tasks under its own identity.
The Oracle product name may invite comparisons to OpenClaw, but the more important parallel is architectural. Vendors increasingly agree that the model is only part of the product. The harness around it, covering identity, policy, replanning, cost control and audit, is what makes multi-hour autonomous work something an enterprise will actually allow. The difference is in the starting point. Microsoft is building a horizontal harness for knowledge workers. Oracle is building one tied to the system of record, which narrows the scope but gives its agents the business context and transactional authority that a general-purpose agent has to borrow.
ServiceNow Stakes a Claim in Agentic Governance
ServiceNow has made some major announcements with respect to securing and managing agents that get it back into the agentic platform conversation. We keep hearing about more incidents where agents are doing unpredictable — and in some cases dangerous — things. And that is why we are also seeing vendors working hard to create adequate measures to prevent agentic incidents. Earlier this year, hyperscalers were leading the way, but now we are also seeing solid efforts from SaaS vendors. Both the SaaS players and the hyperscalers are also emphatically stating that these new solutions are cross-platform, to head off past criticisms of these types of tools. And while ServiceNow was quieter with its AI innovations earlier this year, it has since launched Flow, which was covered by my colleague Mike Leone, and now these governance improvements.
For security and policy enforcement, ServiceNow announced AI Passport, which is a runtime authority layer for AI agents and MCP servers. ServiceNow says it is built on open specifications so other governance products, identity providers, and enforcement points can issue and verify credentials, including for third-party agents, and so credentials are verifiable at every step of a multi-agent workflow.
On the governance side of things, the new Super Intelligence Control Tower adds shadow AI detection for unapproved agents, 400-plus out-of-box connectors for discovery across platforms and vendors, access mapping against real entitlements, and continuous observability of agent traces. A cost layer adds cost governance by user, department, and runtime; value intelligence that ties consumption to quality and realized value; AI Cost Management inside IT Asset Management (frontier model, embedded SaaS AI, hyperscaler, and GPU costs in one view); and graduated alert, throttle, and block policies. EmployeeWorks lets non-coding employees build productivity assistants within admin guardrails (HR Service Delivery first), schedule productivity routines that run in the background, and use ToolSmith, an agentic developer environment that builds tools from natural-language descriptions, with a Conversations API to embed all of it in other apps.
AWS Continues Its Steady Stream of New Kiro Developer Features
Last month we saw AWS make some interesting announcements. For example, it announced a developer on-ramp capability designed to help new developers get started with Kiro and AWS. This was meant to target users of AI-driven developer tools like Lovable and Cursor. AWS also announced a more robust developer workflow capability to help devs wire agents and processes together, geared towards more advanced developer toolsets and the orchestration capabilities we are seeing emerge from SaaS providers. Now the Kiro team is taking steps towards providing more secure local development and testing with its new sandbox capability. This stands to be very helpful because, when developers run agents locally (as many Kiro users do), there is a risk of an agent using local files or configs and creating havoc for the developer or the ops team when the code gets pushed into the cloud. In this case, the secure sandbox segregates the desktop from the agent, from where the code is built and tested. But what could be more interesting is the idea of being able to run multiple sandboxes locally, which would be a big help to developers in the services arena or developers in highly regulated environments. Right now, this is not a documented Kiro capability, but it appears to be possible, since other solutions like Docker CLI can do this.
Oracle Fusion Claw Aims to Improve Enterprise Agentic Execution
Oracle’s new Fusion Claw is another sign that the agent conversation is moving from assistants to harnesses built for long-running, autonomous work. Oracle describes Fusion Claw as a governed agentic execution runtime that pairs AI reasoning with deterministic enterprise computation, and it is launching with 25 Claw-powered agentic applications.
The design choice I find most interesting is the division of labor. A frontier model handles reasoning, planning and adaptation, then hands off to deterministic computation for high-volume execution, so expensive model calls are used only where intelligence is actually needed. Governance is equally central. An “Outcome Trust Harness” enforces each organization’s policies, permissions, and approval boundaries on every run, and an “Outcome Receipt” leaves an auditable record of the decisions made and transactions executed. Customers can set each process anywhere from light assistance to fully automated execution within delegated authority. This lines up with what I wrote in last week’s newsletter about Microsoft’s new Copilot client, which includes Autopilot, an always-on agent that runs long tasks under its own identity.
The Oracle product name may invite comparisons to OpenClaw, but the more important parallel is architectural. Vendors increasingly agree that the model is only part of the product. The harness around it, covering identity, policy, replanning, cost control and audit, is what makes multi-hour autonomous work something an enterprise will actually allow. The difference is in the starting point. Microsoft is building a horizontal harness for knowledge workers. Oracle is building one tied to the system of record, which narrows the scope but gives its agents the business context and transactional authority that a general-purpose agent has to borrow.
Will Super Intelligence Eventually Become Superintelligent?
President Donald Trump recently signed an executive order that directed the federal government to rebrand artificial intelligence as “super intelligence” (SI). It was a rejection of a term (AI) he considered “inaccurate and very ineloquent.” In the administration’s view, today’s technology is neither artificial in its impact nor ordinary in its capabilities. It is real, powerful, and the product of extraordinary human ingenuity. Its name, the president argued, should reflect that achievement.
President Trump’s change was also intended to reinforce U.S. dominance of one of the most consequential technological developments in modern history. By replacing “artificial” with “super,” the administration believes it projects confidence, technological superiority, and national leadership.
Yet despite the administration’s good intentions, the renaming introduced an important complication. Long before the administration became interested in renaming artificial intelligence, computer scientists, philosophers, and futurists were already using a remarkably similar word — superintelligence — to describe something fundamentally different.
Superintelligence is the projected future of today’s AI, but equipped with potentially dangerous capabilities that exceed human intellect across every domain. If such a system someday comes into existence and begins upgrading itself, it might quickly outstrip humanity’s ability to oversee or manage it.
Today’s artificial intelligence can perform tasks that once seemed exclusively human, but we have yet to achieve superintelligence. Even our most advanced AI systems have significant limitations with reliability, reasoning, autonomy, and judgment. Today’s systems are not the self-improving machine intellects envisioned for technological superintelligence.
It is possible for the general public to confuse the two concepts. When political language suggests that the era of “super intelligence” has already arrived, that risks creating a semantic blur between the two. This could lead the public to reasonably assume that the technological milestone of superintelligence has been achieved, which would be a misunderstanding.
We don’t need to obscure important scientific, ethical, economic, and existential questions associated with the possible emergence of machines whose intellectual capabilities exceed our own. Clarity and control are key elements necessary to ensure that superintelligence doesn’t outthink or outwit its creators.
Concern for how “super intelligence” and “superintelligence” are used doesn’t lessen President Trump’s desire to celebrate American innovation and technological leadership. Political branding and scientific terminology can coexist, provided the meanings remain distinct. After all, superintelligence and super intelligence differ by more than a space between two words. That distinction represents the difference between what we have accomplished and a necessary understanding of what may still lie ahead.
Referring to today’s technology as “super intelligence” does not bring us any closer to mastering the true dangers and advantages of superintelligence. However, it does give the starting line a more impressive name.
Most agent controls I see check an agent’s identity and what it can reach inside one company. Bigeye just added a check on the data itself, while four other vendors showed how an agent could check another company’s agent.
Whenever a team shows me an agent with tightly scoped permissions, I ask what happens when the table it reads broke that morning. Permissions don’t cover that. Bigeye’s new Policy Engine weighs the state of the data before each agent action. It can block an agent from a dataset with an open quality incident or one nobody approved for AI. All of these decisions land in one audit trail. I also like that enforcement runs through AI gateways that many customers already operate. As I told TechTarget about Bigeye’s Policy Engine, it takes customers from merely seeing what their agents do to controlling it. Next, I’d love to see blocked requests feed into how Bigeye ranks data issues, because the tables behind the most blocks are worth fixing first.
Bigeye’s checks cover a company’s own agents. I thought this next item was really cool, and easy to miss since it came out of GovWare, a cybersecurity conference in Singapore. Infoblox, GoDaddy, Google, and Cloudflare built a chain of trust with Singapore’s Infocomm Media Development Authority for agents doing business with agents from other companies. In a simulated flight purchase across AWS, Cloudflare, and Google, it stopped a lookalike seller and forged credentials. I love that Infoblox’s piece lets companies publish their agents in DNS. Another company’s agent then finds them the same way a browser finds a website. Most teams can’t confirm that the agent on the other side of a transaction is what it claims to be; building that check on DNS, which every company already runs, should make it far easier to adopt.
Sovereignty plans usually start with storage and the datacenter, and the AI running on top gets checked last. Scality wants to help companies run that AI on infrastructure they own.
When companies walk me through their sovereignty plans, I hear plenty about where data sits and very little about the AI. My Forbes piece from last week argues that AI is probably the weakest part of most sovereignty plans. Most companies rent their AI models, and a model maker can switch off a model either on its own or because a government ordered it to. Each tool an agent calls also receives data wherever that tool runs, sometimes with another provider or in another country. In situations where losing the model would stop the business process, I would keep a second model tested and ready. An open-weight model running on your own systems is one option.
Few teams want to build their own inference setup from scratch, though. To address this, Scality just launched AI Inference Factory, a supported software stack for running inference on infrastructure a company owns. It bundles validated open-weight models, a serving layer, and a control plane for access and metering, all on Scality’s storage. It runs on standard servers from Dell, HPE, Lenovo, and Supermicro, with no special hardware required. I like that the storage maintains each conversation’s context, so GPUs reload it instead of recomputing it. Scality also validates and maintains the whole stack, which I’d expect government agencies and regulated companies to value most.
AI keeps making it cheaper to make charts. But knowing what each number means, and why it moved, still requires people.
When anyone can generate a live dashboard by asking a question, the chart itself stops being what a BI vendor sells. Anthropic’s Claude Dashboards, now in beta, does exactly that. I think the value moves to the data underneath and the definitions that decide what each number means. BI vendors holding those definitions stand to gain from this. I’d also expect governance teams to be uneasy. Plenty of these dashboards, each with its own filters and source tables, will never reach a data team. I dig into what this means for buyers and vendors in my brand-new Forbes piece about how Claude Dashboards makes definitions expensive.
Dashboards can show that a metric dropped, but then someone needs to spend an afternoon digging for the cause. Salesforce just agreed to acquire Loops, whose technology explains why a metric changed and suggests what to do next. I’m a fan of bringing Loops into Tableau, where Pulse already watches metrics and flags changes. An explanation is only as good as the definitions under it, so I’d want Loops reasoning over the same Tableau Semantics definitions everyone else uses.
I also recently spent time in New York with Qlik’s Executive Advisory Board, a room full of data leaders from some of the biggest companies in the world. Context, governance, and cost kept coming up — the agentic AI themes the industry has tracked all year. Interestingly, consolidation within the data stack got more attention than I expected. It makes sense, since many businesses want a simpler data ecosystem. Shrinking the surface area that agents pull from would offset the operational and permission complexities they’re all dealing with today. Often that may not be possible, but it’s on these data leaders’ minds. More on this one in a longer write-up coming soon.
The Apple Home Collection Launch
Apple announced new Matter and Thread smart home products at a “Welcome Home” event in New York on October 13, 2026. The announcements introduced three Apple devices and the Apple Home Collection, a new set of smart home products jointly developed with LG Electronics and Schneider Electric. The Apple products are HomeView (a smart display), a new Apple TV 4K (with a new way to use Siri), and HomePod mini 2 (a smart speaker). Many of the products include support for relevant Matter, Thread, and Aliro standards, reinforcing Apple’s strong commitment to these initiatives.
The Apple Home Collection products carry LG and Schneider Home branding, and Apple intends to sell them through apple.com, the Apple Store app, and Apple Stores. This is a new kind of collaboration, where Apple plays a larger role in design, partnering with companies that already have mass-market manufacturing and distribution channels. HomePod mini 2 is slated to be available starting October 30, with HomeView and Apple TV 4K available starting November 13. More Apple Home Collection products should be available in December, with others starting in early 2027.
Apple products:
- HomeView ($299) is a smart display with an interesting seven-inch square screen, an A18 chip, and Thread networking.
- Apple TV 4K ($299) receives its first update in four years. Three far-field microphones can hear you from across the room, allowing viewers to talk to Siri hands-free, like with a HomePod. Conversations appear on the TV screen. Siri AI, in beta, shows search results as full-screen poster art and answers questions as a small overlay while the show keeps playing. Apple says that its A19 chip delivers up to 50% faster processor performance than the previous generation. Two pairs of HomePod speakers (mini 2 or second generation) can form a surround-sound system, and Thread functionality carries over from the previous models.
- HomePod mini 2 ($149) gets an upgrade with Apple’s S10 chip, and it’s a Thread border router.
Apple Home Collection products:
- LG Smart Deadbolt Lock ($349.95) uses Matter, ultra-wideband (for distance and direction sensing), and Apple Home Key. As of this writing, the product page doesn’t list Aliro and mentions Thread only in a footnote. The press release mentions both Aliro and Thread, but until it’s on the product page, I can only confirm that it supports Matter and Home Key.
- LG Smart Thermostat and Schneider Home Smart Thermostat ($199.95 each) are Matter devices.
- LG Temperature Sensor and Schneider Home Room Sensor ($39.95 each) support Matter over Thread.
- LG Video Doorbell Wired (249.95),FloodlightCam(399.95), Outdoor Cam Plug-In (299.95),andIndoorCamPlug-In(199.95): Apple has not specified Matter support for the cameras, but all of them list Thread as a supported connection.
My take: Apple has been instrumental in developing and supporting Matter and Thread, the standards that define the future of the smart home. I’m pleased to see new products support these standards, and I encourage Apple to be more specific when describing how individual products use them. I’m also happy to see a new way for Apple to collaborate with partners to rapidly expand its consumer product volume and velocity using established brands with global mass-market footprints. But the jury is still out on the products themselves. Are they up to Apple’s high product standards? We’ll see.
French Supermarket Freezer Attack
Your Internet-connected IoT systems might be secured like unlocked cars with keys on the front seats.
On October 11 and 12, 2026, an overnight attack on French supermarkets compromised refrigeration controllers from Danfoss, Copeland, and Dixell. Attackers changed settings on internet-connected controllers in hundreds of stores. Les Mousquetaires, the parent of Intermarché, said freezer units at about 150 of its stores heated to 50°C (122°F) for six hours, causing food losses and fires. Le Monde reported that defrost cycles were lengthened from one hour to eight, reportedly timed so alarms would not sound.
On October 12, France’s CERT-FR named the compromised controllers from all three vendors and said exposing management interfaces directly to the internet is contrary to best practice. Danfoss said most affected units were older models or ran outdated firmware. In August, Claroty presented research on the Danfoss AK-SM 800A at DEF CON and reported that 2,765 controllers were exposed. More than a day after the attack, thousands of Danfoss management interfaces in France were still reachable from the internet. No group has claimed responsibility, and investigators have not said how the attackers got in.
My take: This attack pattern is strikingly similar to the Minnesota water attacks I covered in this Forbes article: controllers reachable from the internet, settings changed remotely, no malware, and no clear owner of long-term security.
For more than a decade, internet connectivity has brought physical system operators remote monitoring and control, data collection, convenience, and efficiency. Connectivity was easy, and the risks looked small. Embedded controllers had obscure interfaces and odd data formats, and an attacker’s time was surely better spent hacking more valuable systems. That calculation assumed that attacker time and skill were scarce, however, and AI is eroding that assumption. Finding exposed, known-vulnerable devices and generating code to attack them is getting faster and cheaper, and attacks are becoming more frequent.
Who ensures that each connection remains permanently secure? In many cases, the answer is “nobody.” Even if equipment manufacturers provide secure systems with regular updates, someone has to properly install and configure each connected device and maintain it perpetually. Many organizations use contractors for initial installation, often without formal arrangements for long-term maintenance. Manufacturers say, “You didn’t install our patches”; contractors say, “You paid us to install this equipment, not maintain it”; and operations folks say, “We never had to worry about this before, nor do we have the budget to deal with it now.” All three might be correct, but AI raises the stakes by enabling attackers to execute large-scale attacks easily. AI-savvy hackers don’t need to crack encryption one key at a time or install complicated malware. Unpatched firmware, default passwords, and back-door connections are more like an unlocked car with the key on the seat.
The time has come to elevate physical device security conversations to management and executive levels. Bosses have to break the blame loop, ask tough questions, and insist on formal answers. Here are some suggestions:
- List all physical devices with an internet connection, direct or through contractor systems.
- Does each device follow the manufacturer’s recommendations for internet connectivity?
- Does the device actually need to be connected for good business reasons, or is it just a nice-to-have convenience? (In other words, minimize your attack surface.)
- Confirm the update status of each device. Is it compliant with manufacturer recommendations?
- Who is accountable for protecting every path into all connected devices?
- When would you find out if a device is compromised? What might the potential damage be, and who would take action?
- What’s the isolation and recovery plan?
AI simplifies the exploitation of device vulnerabilities and makes exploits much more scalable, sophisticated, and commonplace. For years, IoT cheerleaders have yelled “Let’s connect everything.” It’s time to change that yell to “Let’s make sure all those connections are secure.” If you can’t do that, your only option is to disconnect all insecure devices or move them to air-gapped or properly firewalled networks.
The risk-reward math for device connectivity has changed, and that shift requires executive-level oversight.
iPhone Duo pre-orders opened last week, and to some people’s surprise, they were immediately sold out through the end of January. Yes, you may have to wait until almost the Super Bowl to get your new foldable iPhone.This fits my own belief — since well before pre-orders began — that the Duo would be a status symbol due to its price and limited availability. The new iPhone is the first new form factor from Apple since the iPhone’s inception, plus there are many people who have been waiting for an iOS foldable for a long time. Basically, since I’ve been using foldables, plenty of loyal iOS users have been saying they’d wait for a device like the Duo. That’s why I felt sure demand would immediately outstrip supply, especially since Apple could not forecast how many people had been waiting for a Duo and how many would be willing to pay $2,000 in this complex economic climate. Smartphone shipments are down 20% overall right now, so it’s perfectly rational for Apple not to have expected as much demand as it actually encountered.
The carriers have some deals for the Duo in place, but none are particularly aggressive given current market dynamics, especially with SpaceX hammering carriers’ share prices. I think the Duo will likely surpass everyone’s sales expectations, and that many users will be hooked from day one and never go back to a traditional candy-bar phone. That said, I also believe there will be people who will take advantage of the Handoff feature and keep their iPhone 17 Pros because the camera on the iPhone Duo simply isn’t up to the task. The only foldable that has everything I want in a foldable — including a great camera — is the Motorola Razr Fold, and it’s not an iOS phone.
I expect the Duo will have a second wave of demand once people start receiving their Duos and showing them to friends and family; if I’m right about that, we will likely see pre-orders extend even past the end of January. That said, there have been rumors that Apple has already expanded manufacturing capacity for the Duo by 30%, but I’m still not sure that can keep up with demand. Most likely, these phones will not be readily available until the second quarter of next year, at best. I’m excited to get my Duo in hand to review and compare to the rest of the foldables I’ve been testing, including the Samsung Galaxy Z Fold8, the Google Pixel Pro 11 Fold, and the Razr Fold.
Ericsson and Nokia’s pullback from private wireless could spell trouble for others like Celona — or cede the market entirely. For a long time, private wireless was seen as a potential growth area for 5G, with both Ericsson and Nokia pursuing it in 4G and accelerating it into 5G. Unfortunately, it seems that private 5G ended up being a lot like IoT, in that the scale doesn’t necessarily make a one-size-fits-all solution possible. Many of these deployments tend to be smaller with fewer endpoints and more bespoke to the enterprise’s needs. We’ve seen companies like John Deere go out and build their own private 5G solutions in-house, which in many ways reduces deployment costs and management overhead and makes infrastructure vendors less relevant. It is unclear what will happen to the rest of the market now that Ericsson and Nokia are both out, but I suspect there will still be lots of smaller-scale opportunities as we head towards 6G.
NVIDIA and its partners have launched new RTX Spark systems to an enthusiastic response in the market. These are the first real competitors in the Arm space since Qualcomm’s entry with the Snapdragon X Elite family. I believe that the RTX Spark family introduces a new tier of performance for on-device AI that simply didn’t exist before. NVIDIA claims up to 1 petaflop of AI compute via FP4 with a 20-core Arm CPU and an RTX 5070-class mobile GPU. All of this compute is combined into a single Arm-based device running on Windows, and it gives Microsoft a new chance at competing with Apple in local AI. Frankly, up until now, Microsoft and Windows have struggled to compete with Linux and macOS in AI development and local AI, and the new RTX Spark laptops and desktops could help turn that tide. I have already published my review of the RTX Spark platform and some of its early growing pains regarding bugs and compatibility, but I do believe most of those issues will be resolved quickly. The only thing that I believe holds the platform back is the CPU architecture being two years old; it would be more competitive if it used something like a C2 Ultra or C2 Pro from Arm.
The most recent chapter in the ongoing legal conflict between Arm and Qualcomm has drawn to a close, with a jury verdict finding that Arm breached its architecture license agreement (ALA) and interfered with Qualcomm’s ability to develop certain chips but did not breach its technology license agreement (TLA). The jury awarded Qualcomm $2 in nominal damages for the two different breaches, but it remains to be seen what the actual damages will be. The remedies that Qualcomm seeks include not paying Arm license fees for a period of five years, which could amount to billions of dollars. There are still many details of this case that the federal judge must resolve, and it’s unlikely that we will have a full resolution, even once those details are sorted out, until all the appeals have been exhausted. What is fundamental is that the jury agreed that Arm interfered in Qualcomm’s relationship with Meta and withheld support and tools that Qualcomm needed to build its chips.
Research Papers
- Making Data Usable Where It Lives (Mike Leone, Patrick Moorhead)
- Where the Chip Really Begins (Matt Kimball, Patrick Moorhead)
Research Note
Field Notes
- Where Intel’s GPU D270 Fits In Enterprise AI (Matt Kimball)
- NVIDIA RTX Spark Elevates Windows on Arm, But Its CPU And Software Hold It Back (Anshel Sag)
- Microsoft Commits To Maia At Gigawatt Scale And As Its Future, Primary Inference Platform (Patrick Moorhead)
- Applied Materials Launches EPIC – What Does This Mean For AI Infrastructure? (Matt Kimball)
- Cisco at OCP 2026: Silicon One, NVIDIA Optics and the Network Behind AI (Matt Kimball)
- AMD Helion: Building The Rack Around The Agent (Matt Kimball)
- Microsoft Azure Maia 300: From Custom Chip To Hyperscale Platform (Matt Kimball)
Podcasts
The 6G Podcast (Anshel Sag)
Six Five (Patrick Moorhead)
Don’t miss future MI&S podcast episodes! Subscribe to our YouTube Channel here.
Press Citations
AMD / GPU / Patrick Moorhead / Tech Times
AMD Helion OCP Blueprint Adds Model Router: Operators Scale CPUs Without Buying More GPUs
Apple / HomeView Home Devices & Siri AI / Anshel Sag / Fortune
Apple’s new home devices are a bet on its upgraded Siri AI
Bigeye / Agent Trust Hub / Mike Leone / TechTarget
Bigeye focuses on enforcing AI governance with Policy Engine
ByteDane, NVIDIA, MediaTek / Chips / Matt Kimball / Tech Times
ByteDance ASIC Deal: Nvidia Funds Chips US Export Controls Meant to Block
IonQ / Elisity Acquisition / Paul Smith-Goodson / CSO Online
IonQ’s $1.2 billion acquisition of Elisity is trying to position it for quantum cybersecurity
Microsoft / Maia 300 AI Chip / Matt Kimball / Data Center Knowledge
Microsoft Pushes Custom AI Chips with Azure Maia 300
Microsoft / Maia 300 AI Chip / Matt Kimball / Fierce Network
Microsoft’s Maia 300 isn’t a Nvidia killer – it’s a power play
Microsoft / Maia 300 AI Chip / Matt Kimball / TechTarget
Microsoft bets on Azure Maia chips to stretch AI data center power
Microsoft / Maia 300 AI Chip / Patrick Moorhead / 24/7 Wall St.
Microsoft Just Took Its Biggest Step Toward Independence From Nvidia
Microsoft / Maia 300 AI Chip / Patrick Moorhead / CNBC
Microsoft to rely more on homegrown AI chips, reducing dependence on Nvidia
Microsoft / Maia 300 AI Chip / Patrick Moorhead / The Motley Fool
Microsoft Just Unveiled a Brand-New AI Chip. Could It Be the “Nvidia Killer” Everyone’s Been Looking For?
ServiceNow / AI Agent Building & Agent Identities / Melody Brue / CIO
ServiceNow makes it easier to build AI agents, gives them verifiable identities
SpaceX / Spectrum Deal & Wireless Ambitions / Anshel Sag / Reuters (via Investing.com)
SpaceX’s spectrum deal still leaves it far from rivaling telecom giants
Workday / UI / Melody Brue / CIO
Workday launches Sana-based UI, autopilot agents, as it deepens its focus on AI
Press Citations
AI Stocks / AI Stock Selloff / Patrick Moorhead / Yahoo Finance
This AI rout is just another scare for investors
New Gear or Software We Are Using and Testing (New)
- Alienware 39” OLED Gaming Monitor (Anshel Sag)
- Microsoft Surface Laptop Ultra (Anshel Sag)
- Apple Watch Series 12 (Anshel Sag)
- Unannounced AR Glasses (Anshel Sag)
- Even Realities G2 & R1 Ring (Anshel Sag)
- Unannounced Smart Glasses (Anshel Sag)
- Google Pixel 11 Pro Fold (Anshel Sag)
- Google Pixel Watch 5 (Anshel Sag)
- Galaxy Watch Ultra2 (Anshel Sag)
- Galaxy Watch 9 (Anshel Sag)
- Samsung Galaxy Z Fold8 (Anshel Sag)
- Galaxy Z Fold8 Ultra (Anshel Sag)
- Samsara Tracking Label (Bill Curtis)
- Panasonic Toughbook 56 (Anshel Sag)
- X by Xreal Headset (Anshel Sag)
- Amazon Quick (Jason Andersen)
- MSI Claw 8 EX AI+ Gaming Handheld (Anshel Sag)
- XREAL R1 AR Gaming Headset (Anshel Sag)
- Fitbit Air (Anshel Sag)
- Anker Nano Power Strip (10-in-1, 70W, Clamp) (Anshel Sag)
- Anker Mag Go Prime Wireless Charging Station (Anshel Sag)
- Anker Nano Charger (Anshel Sag)
- Lenovo Legion Go 2 (Anshel Sag)
- OnePlus 15 (Anshel Sag)
- Oppo Find X9 Pro (Anshel Sag)
- Apple M5 MacBook Pro (Anshel Sag)
- Naya Create Modular Keyboard (Anshel Sag)
- 2.0 Antec Flux Pro Case (Anshel Sag)
Events MI&S Plans on Attending, in Person or Virtually (New)
Unless otherwise noted, our analysts will be attending the following events in person.
- Zoomtopia, October 20, Virtual (Jason Andersen)
- Synaptics Analyst event, October 20, San Jose (Bill Curtis)
- Couchbase Catalyst, October 20-22, Ft. Lauderdale (Mike Leone)
- IBM TechXchange, October 25-28, Atlanta (Jason Andersen)
- Lenovo Analyst Summit, October 26-27, Raleigh (Matt Kimball)
- Oracle AI World + SuiteWorld 2026, October 26-29, Las Vegas (Matt Kimball, Mike Leone)
- Qualcomm Dragonwing IoT Day, October 28, San Diego (Bill Curtis)
- Raven Resonance Launch, October 28, San Francisco (Anshel Sag)
- Zoomtopia, October 20, Virtual (Jason Andersen)
- Synaptics Analyst event, October 20, San Jose (Bill Curtis)
- Couchbase Catalyst, October 20-22, Ft. Lauderdale (Mike Leone)
- IBM TechXchange, October 25-28, Atlanta (Jason Andersen)
- Lenovo Analyst Summit, October 26-27, Raleigh (Matt Kimball)
- Oracle AI World + SuiteWorld 2026, October 26-29, Las Vegas (Matt Kimball, Mike Leone)
- Qualcomm Dragonwing IoT Day, October 28, San Diego (Bill Curtis)
- Raven Resonance Launch, October 28, San Francisco (Anshel Sag)
- Dell Analyst Summit, November 2-4, Austin (Matt Kimball, Mike Leone)
- Red Hat Analyst Day, November 5, Boston (Jason Andersen)
- Adobe MAX, November 9-12, Miami (Melody Brue)
- Mechanical Orchard Analyst Summit, November 10, Boston (Matt Kimball)
- Commvault Shift, November 10-12, Nashville (Mike Leone)
- AWS re:Invent, November 30-December 4, Las Vegas (Patrick Moorhead, Matt Kimball, Jason Andersen, Mike Leone)
- AWS re:Invent, November 30-December 4, Las Vegas (Patrick Moorhead, Matt Kimball, Jason Andersen, Mike Leone)
Subscribe
Want to talk to the team? Get in touch here!
The post MI&S Weekly Analyst Insights — Week Ending October 16, 2026 appeared first on Moor Insights & Strategy.
Popular Products
-
Electronic String Tension Calibrator ...$41.56$20.78 -
Pickleball Paddle Case Hard Shell Rac...$27.56$13.78 -
Beach Tennis Racket Head Tape Protect...$59.56$29.78 -
Glow-in-the-Dark Outdoor Pickleball B...$49.56$24.78 -
Tennis Racket Lead Tape - 20Pcs$51.56$25.78