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.
Last week was one of our busiest of the year, with several of us attending multiple events across many time zones. Anshel Sag and Matt Kimball traveled the farthest to attend Computex in Taipei. (Take a look at Matt’s “Datacenter — Silicon” section in this week’s roundup, and keep an eye out for multiple Computex-related Research Notes from Anshel in the coming days.) Meanwhile, both Bill Curtis and I followed the Computex presentations remotely. Bill used this to write a long entry about physical agentic computing in his “IoT and Edge” section of this week’s roundup, and I posted my own thoughts about NVIDIA, Qualcomm, AMD, and Microsoft on LinkedIn.
Satya Nadella presents at Microsoft Build 2026. (Credit: Jason Andersen)
Meanwhile, Jason Andersen attended Microsoft Build 2026 in San Francisco. You can see his takeaways under “Apps, Agents, and Automation” in this week’s roundup and on his LinkedIn post. I also attended Cisco Live, and I shared notes on CEO Chuck Robbins’ keynote and president / chief product officer Jeetu Patel’s keynote. You’ll find even more event analysis in other sections of this week’s roundup by Anshel, Melody Brue, and Mike Leone. I wouldn’t want every single week to be so packed, but it was exciting to know what’s front and center for so many of tech’s big players.
This week, Matt and Mike are at the NetApp Analyst Summit in San Jose. From there, Matt travels to the Broadcom Mainframe Analyst Summit in Boston. For news and hot takes from these events, be sure to follow us on X and LinkedIn.
The spring season of client and vendor events continues in full swing, and we’re on the road a lot over the coming months, including at the AWS Analyst Summit, Databricks Data + AI Summit, HPE Discover, Pure Accelerate, and Connectivity Standards Alliance Unify — all in June. If you’ll be attending any of the same events or see that we’ll be in your city, please reach out.
Last week, Moor Insights & Strategy analysts’ perspectives appeared across leading business and technology outlets, including Wired, TechTarget, VentureBeat, CNBC, InfoWorld, Data Center Knowledge, CIO, CX Today, CX Foundation, and The Wall Street Journal. Coverage highlighted Apple’s exploration of camera‑equipped AirPods, as well as Cisco Live announcements around Cloud Control and AgenticOps for unifying IT infrastructure management. Analysts also examined why context layers are becoming the next production challenge for enterprise AI, weighed in on HPE’s biggest earnings beat since 2018, and assessed Microsoft Fabric and Web IQ as foundations for better enterprise agents. Other commentary focused on NVIDIA’s Vera Rubin and Vera CPUs and the new DSX OS for running large‑scale AI factories, Snowflake’s Horizon Context and CoWork efforts to improve AI agents, and Zoom’s ZoomMate AI agent and productivity suite.
Our MI&S team also published 10 deliverables — 3 Research Notes, 2 Analyst Insights, and 5 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
Microsoft Build
This week I was one of only eight analysts invited to attend Microsoft Build. That is a very small number of analysts at what was a very different kind of Microsoft event. Build took a different approach this year and moved back to San Francisco to the Fort Mason Center (a personal favorite) for a more stripped-down event. So, including the location, it was very similar to what we’d see at a GitHub Universe or a previous HashiConf (which is now merged with IBM TechXchange, alas). However, despite a slimmer analyst list, Microsoft took a major-event approach to the content, with a three-hour keynote and more than 80 announcements on day one. Here are my takeaways.
The day-one keynote ran long but landed as unusually broad for a developer event — more than 80 announcements anchored by a heavier-than-usual emphasis on infrastructure. The clearest throughline was edge AI moving from concept to real investment: From models such as Ion Instruct and Ion Plan that run full agentic loops locally on Windows with no cloud dependency, to new developer systems such as the Surface RTX Spark dev box (1 petaflop, 128GB unified memory), it read as a direct answer to Apple’s developer-hardware lead. However, I was more interested in Project Solara, which introduces agent-first form factors (desk units, badge devices) aimed at field service, healthcare, and industrial settings. Underpinning all of it was a strategic framing move signaling that architects and developers should be watching the infrastructure layer more closely than most currently do. In some ways that notion is a throwback to the pre-Windows era in which hardware more heavily influenced software design.
The second theme was how the developer engagement and platform layers are shifting in ways that matter for enterprise adoption. Microsoft is looking to continue its lead with GitHub and is repositioning it as the control plane for agentic workflows rather than just version control. If successful, that could have a profound impact on its competitors, who are trying to get lift with their own agentic control planes but lack GitHub’s huge install base. In addition to evolving its core, there is a new GitHub Copilot app merging CLI, canvas, and multimodel support to handle the rising session-management load as coding volume climbs. (Agentic systems have already driven a reported 3x increase in commits.) Think of it as Claude Desktop for high-volume use.
Finally, there was a lot of improvement at the platform layer. The most notable was seeing tighter integration within Microsoft IQ, which ties Foundry, Fabric, and Microsoft 365 into a more unified intelligence layer so developers don’t have to hand-wire context sources, and it will enable a common set of platform services (security, governance, APIs, etc.) for agents to use.
Day two reinforced the client-edge and model narratives. The Surface Laptop Ultra with onboard NVIDIA RTX and the developer-specific Windows configurations (some available now, the rest landing in the fall) give IT a genuinely stronger case for standardizing the client-edge footprint on Windows/Microsoft 365 — especially with safety features like Microsoft Execution Containers (MXC) — even if prying Macs out of the hands of devs and creative teams remains a tall order.
I also had the opportunity for some deeper conversations about the new Microsoft-built MAI models and Foundry IQ, which left a positive impression. While some may ask why Microsoft would invest in new AI models when it has relationships and support across most of the leaders, it’s not just the models. When paired with Frontier Tuning, the MAI path supports the thesis that a smaller fine-tuned model can outperform a much larger general-purpose one in terms of latency, accuracy, and cost, making the ROI for a bespoke organization-specific model real in the right circumstances. The big question will be the cost and complexity of creating and maintaining the models, which we don’t yet have good insight into.
Microsoft Build
This week I was one of only eight analysts invited to attend Microsoft Build. That is a very small number of analysts at what was a very different kind of Microsoft event. Build took a different approach this year and moved back to San Francisco to the Fort Mason Center (a personal favorite) for a more stripped-down event. So, including the location, it was very similar to what we’d see at a GitHub Universe or a previous HashiConf (which is now merged with IBM TechXchange, alas). However, despite a slimmer analyst list, Microsoft took a major-event approach to the content, with a three-hour keynote and more than 80 announcements on day one. Here are my takeaways.
The day-one keynote ran long but landed as unusually broad for a developer event — more than 80 announcements anchored by a heavier-than-usual emphasis on infrastructure. The clearest throughline was edge AI moving from concept to real investment: From models such as Ion Instruct and Ion Plan that run full agentic loops locally on Windows with no cloud dependency, to new developer systems such as the Surface RTX Spark dev box (1 petaflop, 128GB unified memory), it read as a direct answer to Apple’s developer-hardware lead. However, I was more interested in Project Solara, which introduces agent-first form factors (desk units, badge devices) aimed at field service, healthcare, and industrial settings. Underpinning all of it was a strategic framing move signaling that architects and developers should be watching the infrastructure layer more closely than most currently do. In some ways that notion is a throwback to the pre-Windows era in which hardware more heavily influenced software design.
The second theme was how the developer engagement and platform layers are shifting in ways that matter for enterprise adoption. Microsoft is looking to continue its lead with GitHub and is repositioning it as the control plane for agentic workflows rather than just version control. If successful, that could have a profound impact on its competitors, who are trying to get lift with their own agentic control planes but lack GitHub’s huge install base. In addition to evolving its core, there is a new GitHub Copilot app merging CLI, canvas, and multimodel support to handle the rising session-management load as coding volume climbs. (Agentic systems have already driven a reported 3x increase in commits.) Think of it as Claude Desktop for high-volume use.
Finally, there was a lot of improvement at the platform layer. The most notable was seeing tighter integration within Microsoft IQ, which ties Foundry, Fabric, and Microsoft 365 into a more unified intelligence layer so developers don’t have to hand-wire context sources, and it will enable a common set of platform services (security, governance, APIs, etc.) for agents to use.
Day two reinforced the client-edge and model narratives. The Surface Laptop Ultra with onboard NVIDIA RTX and the developer-specific Windows configurations (some available now, the rest landing in the fall) give IT a genuinely stronger case for standardizing the client-edge footprint on Windows/Microsoft 365 — especially with safety features like Microsoft Execution Containers (MXC) — even if prying Macs out of the hands of devs and creative teams remains a tall order.
I also had the opportunity for some deeper conversations about the new Microsoft-built MAI models and Foundry IQ, which left a positive impression. While some may ask why Microsoft would invest in new AI models when it has relationships and support across most of the leaders, it’s not just the models. When paired with Frontier Tuning, the MAI path supports the thesis that a smaller fine-tuned model can outperform a much larger general-purpose one in terms of latency, accuracy, and cost, making the ROI for a bespoke organization-specific model real in the right circumstances. The big question will be the cost and complexity of creating and maintaining the models, which we don’t yet have good insight into.
Microsoft has announced Project Solara, which it calls a chip-to-cloud platform to enable agent-first experiences on any kind of device you might want. Microsoft says this “operating system” is not beholden to a single device but can move from device to device and follow you wherever you need it. Microsoft says it should enable multiple agents to live in a multi-agent world that also respects people’s privacy, security, and personalization. I understand this vision, but I believe it will be interesting to see how it’s implemented in practice and whether this spawns a new class of devices or simply modifies the way we use our existing devices. Qualcomm was prominently featured in the mobile-phone-like form factor, and MediaTek in the smart-display form factor, leveraging the expertise of each chip maker.
Cisco Live U.S. 2026 sent a clear message to the market: Cisco is no longer positioning itself as a networking vendor. It is positioning itself as the AI-native control plane for critical enterprise infrastructure. The centerpiece of this year’s show was agentic operations, with AI “coworkers” governed across Cisco Cloud Control, Cisco IQ, and Splunk to deliver cross-domain visibility, trusted delegation, and digital resilience at scale. What stood out most was how Cisco is collapsing historically siloed domains (networking, security, observability, and collaboration) into a single unified architecture.
Webex and contact center, often treated as a side story in Cisco’s portfolio narrative, stepped firmly into the spotlight as first-class citizens within the same Cloud Control, security, and AI fabric. That is a meaningful signal: collaboration is no longer a standalone business line, but a fully integrated layer within Cisco’s broader AI platform.
I’ll be digging into this further in an upcoming research note with my full observations and analysis from the show.
The fight in data platforms has mostly moved off the file format. The table-format question is largely settled around Iceberg, and the open question now is who governs access once a bunch of engines are all reading the same tables. Cloudera made a sensible move in this regard last week: It adopted Apache Polaris, the open Iceberg catalog, and contributed a Ranger plugin so its governance engine can enforce policy inside that catalog. The idea is that policy travels with the data instead of getting re-implemented per engine. That’s the governed interoperability story that regulated buyers keep asking for. It fits Cloudera’s open-source roots and chips away at the lock-in objection the proprietary catalogs must face. But I wouldn’t oversell it. The Ranger plugin is still in beta, and contributing a plugin to an open catalog is a direction more than a finished capability. Still, it gives us a useful read on where the open-catalog question is heading.
I spent the week at Snowflake Summit, and the partner activity around it was worth a look. Plenty of vendors show up in force at a flagship event like this, so the turnout on its own doesn’t mean much. They were all plugging into the same place. Collibra and Immuta wired access governance into the platform, Anomalo plugged in data-quality monitoring, and ThoughtSpot and AtScale lined up their semantic models with Snowflake’s. For an enterprise buyer, the upside is you mostly don’t rip and replace, since the tools you already run increasingly resolve to the same governed definitions instead of each carrying its own version of the truth. The fair question with any of these is how good the governed context underneath actually is, because a partner tool is only as trustworthy as what it reads. Horizon Context is Snowflake’s answer to that, though a lot of it is still early, so the partners are plugging into a foundation that’s improving but not finished. The usual worry with leaning this hard on one platform is lock-in. Snowflake has done real work to ease that concern. Open table formats, the shared catalog, governance that runs inside other vendors’ tools — all of it means you can use other engines without losing the governance. You still want to watch how much of your data work ends up running through Snowflake, but it has left the exits open in a way a lot of platforms don’t.
Cisco Live was the pure infrastructure event of last week. Agentic AI is forcing networking and security to be rebuilt around shared identity, telemetry, and policy, and Cisco is laying bets on that approach from the silicon up. Cloud Control is the piece that ties it together — one management plane across networking, security, observability, and AI infrastructure. The clever move underneath it is how Cisco recast Silicon One. The chip that used to get pitched as an answer to InfiniBand for big training jobs is now the secure networking silicon that feeds the whole platform, emitting the telemetry and absorbing the policy that the shared control layer depends on. Nexus One does the same thing higher up, pushing workload identity into switching so a shared GPU cluster knows which job is which. For most enterprises running agents in production, the hard part now is operating all of it, and that’s exactly the layer Cisco is selling into. It’s the most coherent platform story Cisco has assembled in years, and I still think the company is underrated as a full-stack agentic AI player.
An agent is a new kind of user. It carries credentials and takes actions on its own, and most security tooling was never built to police something like that. Cisco went straight at it. Agent Gateway in Secure Access treats agents as their own identities, with action-level policy over what the agents can call across models, tools, and APIs. AI Defense paired with AppOmni goes after agents embedded in SaaS apps like Copilot and ServiceNow, where a lot of real agent activity is going to live. Put next to what Cisco already offered in model security and runtime protection, the agent-security stack is starting to look unusually complete for one vendor. For a security team staring at agent sprawl with no clear way to see or police it, that breadth is the appeal. The honest caution flag is on timing, though, because a chunk of this is still when-and-if-available and some of the more ambitious pieces slipped to later in 2026. To put it another way, the story fully arrived at the show while the shipping is still partial. The real test is whether it all gets genuinely built and integrated by the time agents are everywhere. Announcing the platform was the easy part.
Last week was undoubtedly the week of Computex coverage for me. While I wouldn’t normally place Computex as a priority conference, AI has changed everything. And the discussion has shifted to how AI is delivered — from the chips to the systems to the networks and out to the edge, whether that edge is an AI PC, a wearable, or some other kind of device.
Need proof of this? Consider the following: Marvell CEO Matt Murphy delivered one of the keynotes. Marvell — the company that designs and builds custom chips that power the AI stack at the lowest levels — is not typically one we would see at a consumer show. Not only was Murphy delivering a keynote, he was joined onstage by NVIDIA CEO Jensen Huang, who proclaimed that Marvell would be the next “trillion-dollar company.” (It’s worth noting that NVIDIA and Marvell are tight partners, with the former recently investing $2 billion in the latter.)
That said, below are some of my top-level takeaways from Taipei.
AI Infrastructure Is Fragmenting, Not Consolidating
When we saw what I would call the first wave of the AI cycle — training — the dominant player was NVIDIA. And every systems vendor rushed to partner with and design around it. In particular, we saw a number of “AI factories” that were effectively indistinguishable from one another.
As the inference wave begins, we are seeing a fragmentation of that silicon space, with familiar challengers like AMD gaining momentum and newcomers such as Cerebras and Tenstorrent entering the scene. Further, we see more players (e.g., FuriosaAI) emerging. (FuriosaAI just signed a partnership agreement with Broadcom to jointly develop its next-generation accelerator.) Likewise, the cloud providers have leaned heavily into their respective custom accelerators to deliver at-scale inference as they recognize the opportunity in front of them.
The reasons for this are quite simple. Inference is a different kind of workload (actually multiple workloads) with distinct requirements for compute, memory, and connectivity. And organizations are bound by power and cost. Finally, inference is not bound by CUDA. As a result, we are seeing greater diversity in this market. And I suspect we will see even more before any kind of consolidation takes place.
When Is a CPU not a CPU?
At Computex, NVIDIA announced that its AI CPU, Vera, was seeing broader adoption in the market, with several large-scale customers such as Oracle Cloud Infrastructure adopting it. Additionally, Arm CEO Rene Haas held a keynote showcasing the progress of Arm across the entire AI ecosystem, with the company’s AGI CPU in focus.
Joining the Arm CPU market was Qualcomm, which unveiled the Dragonfly CPU (for which details are being held back until the company’s investor day in late June).
This all leads me to a question I’ve been pondering for a while. Are these CPUs truly CPUs? I think I may be looking at this the wrong way, but when I think of a CPU, the C (central) is critical. And I think of it supporting a wide range of workloads. In fact, NVIDIA reinforces my thinking here by the way it positions Vera as a CPU built for agents — or a purpose-built AI CPU.
I don’t know that this point matters. Maybe I’m splitting hairs. But it feels like the definition of a CPU has become quite broad.
Hey x86 — Arm Is Coming for You
Arm’s CPU datacenter market share has doubled in the last year. Three of the major chip announcements at Computex involved Arm. NVIDIA is delivering an Arm client CPU, with design wins already secured. And Intel announced it is deepening and formalizing its custom silicon business, surely in response to this momentum.
Those who have been around this market for a while can remember when big iron and UNIX dominated the datacenter, until a low-powered client processor started handling workloads like file and print. And folks scoffed, saying that such a processor could never handle real tasks like serving databases. By a few years later, UNIX was a dinosaur, and Intel owned the datacenter.
With that in mind, do not discount what is happening with Arm. Not in the datacenter, not on the client. It is likely to continue taking share. Not because of power efficiency. But because of optionality. Going back to my previous point about CPUs becoming more specialized … the Arm ecosystem is building right-sized chips for the right workloads in the right power envelopes, from the smallest devices to the largest server farms.
One note on the client front. Huang, maybe poking a little fun at Qualcomm, talked about how his company got the client chip right with its RTX Spark. But Qualcomm has gotten its client work right, too. And Qualcomm did all of the down-and-dirty work in hardening Arm for Windows. Make no mistake about it. NVIDIA seems sure to be the company that moves the market, but Qualcomm did all the hard stuff.
Intel Is Well on Its Way Back
Intel certainly did not deliver the most dramatic announcements at Computex. And in many ways, I think that may be exactly what the company needed.
The launch of Xeon 6+ on Intel 18A, scaling to 288 cores (up to 576 cores in 1U), represents the company leaning into a combination of its CPU, transistor, and manufacturing acumen to deliver cutting-edge parts to the datacenter space. This is a claim no other company can make.
But maybe more interesting is what Intel didn’t do.
The company appears to be taking a measured approach to AI. Rather than making aggressive claims about GPUs or directly challenging NVIDIA on every front, Intel seems focused on rebuilding credibility and targeting areas where it can create differentiated value. The Rack Scale Architecture announcement fits within that framework. The broader concept of the “Intelligence Center” provides an interesting reframing of how enterprises may think about future datacenter deployments.
I am especially interested in the Crescent Island GPU and Intel’s partnership with SambaNova. The strategy is intriguing, though questions remain about target workloads, customer adoption, and how Intel ultimately differentiates its approach from competing AI infrastructure ecosystems. That said, I am OK with ambiguity instead of claims that have to be retracted later.
Funny, but the demo I found most compelling may not have even been on the show floor. Intel gave a demo of its Superclaw technology, where a workstation with four B70 GPUs was serving agentic AI at the department level. For all of the cool and exciting AI demos I’ve witnessed over the past several years, this was the first time I saw a company demonstrate how AI can be served to a real enterprise customer — with affordable infrastructure and realistic use cases where tasks are executed locally first and expanded outward as necessary.
Practical AI, delivered.
If this is where Intel is headed, it may be further ahead than many give it credit for. It just isn’t chasing headlines.
Microsoft Build had a lot going on across Azure data last week. Take its Rayfin launch. Microsoft is turning Fabric from a place you send data for analytics into a governed backend for the apps that agents now build. Coding agents made generating an app’s frontend almost free, but standing up a secure production backend behind it hasn’t gotten easier, and that’s where agent-built apps stall. Rayfin lets a developer (or their agent) define the whole backend in code and deploy it onto Fabric, so the app inherits Microsoft’s identity, governance, and data controls the moment it ships. HorizonDB, a new managed Postgres, and Fabric IQ reaching GA as the semantic layer round out the estate. The bet underneath this slate is that whoever owns the governed data estate owns the apps built on top. For an enterprise drowning in quickly built AI apps, that’s a real draw, since each app lands inside controls you already trust instead of becoming another thing security has to chase. The catch comes with developers: Governance is what a CIO wants, but to the person writing the code, it can feel like lock-in. The question is whether the apps that matter get built on Fabric or just hosted there.
Snowflake spent its Summit event answering the question that has followed it for years — whether it’s genuinely open — with a corollary: whether its governance reaches the agents now acting on the data. After spending a few days on the show floor, I’ve come to think it answered more than I expected, and with shipped product instead of positioning. Snowflake splits its agents by who uses them, with builders on one surface and business users on another, and both grounded in the same governed data. It also put OpenAI and Anthropic models inside its own control plane as governed endpoints, so a model reaches into the data under the same rules rather than the data leaving to reach the model. I’m not handing it a clean win. Snowflake CoWork still has to beat Copilot for daily attention, a lot of the sharper pieces are in preview, and aiming this at every knowledge worker is a stretch. But the distance between Snowflake’s open story and its actual product is the smallest it’s been, and that’s what I’d carry out of the week.
Computex 2026: The Dawn of Physical Agentic Computing
Computex 2026 signaled a significant industry milestone for agentic AI operating natively in the physical world. Four distinct ecosystems emerged — three vertical hardware-plus-software platforms and one horizontal control plane.
Platform 1: The Windows-NVIDIA Agent Computer: Positioned for premium client endpoints, the “Winvidia” platform (you heard it here first) integrates the NVIDIA RTX Spark architecture (Grace + Blackwell) with Windows-native Microsoft Execution Containers (MXC) and NVIDIA OpenShell to host autonomous agents as primary users. It’s the agent PC you may soon buy.
Platform 2: The NVIDIA Linux Robotics Platform: Built for high-throughput industrial edge systems needing autonomy without Windows dependencies, this tier extends the Blackwell architecture, CUDA, and OpenShell down to a Jetson Thor substrate, running NVIDIA’s native Isaac stack and Cosmos world models.
Platform 3: The Non-NVIDIA Robotics Platform: A highly diverse, low-power multi-vendor ecosystem featuring Qualcomm, Intel, MediaTek, NXP, and a growing list of other chip companies. It prioritizes tightly quantized vision-language-action models, leveraging novel architectures like NXP’s “Neural Axis” to distribute intelligence down to low-latency local reflexes.
The Microsoft Governance Overlay: This is the horizontal wild card spanning all three platforms. Microsoft recognized the need for any platform running an enterprise agent to operate within Microsoft’s established administration infrastructure. The result is a hardware-independent agent control plane that uses Agent 365, Entra ID, and Intune to establish uniform security contexts, compliance boundaries, and audit registries across disparate edge fleets.
The Strategic Outlook
Looking through a longer lens, the edge agent landscape is transitioning from a hardware architecture race to a software lifecycle war. While the market splits by compute tiers — with the open ecosystem scaling across cost-sensitive, lower-power industrial footprints and NVIDIA capturing premium, heavy multimodal workloads — silicon capability alone will not secure long-term design wins. The permanent moats belong to developer gravity and toolchain completeness; cost-effective chips cannot win the floor if NVIDIA entirely owns the simulation-to-deployment pipeline. Success will require significant software investment.
However, the ultimate governor of rapid market scale-up isn’t the silicon — it is the unmapped terrain above it. As universal regulatory frameworks come into effect, the industry faces a significant long-term challenge in fleet maintenance. The path from an edge prototype to a secure, compliant product, maintained over a decade or more, remains extremely complicated, highly fragmented, and riddled with technical debt. Scaling edge AI requires a comprehensive solution.
The critical battleground over the next decade will be the consolidation of this unclaimed deployment layer. Whichever ecosystem successfully standardizes the long-term deployment and maintenance pipeline stands to unleash the growth engine of the physical edge. This will take a while, but I already see signs of consolidation and a willingness among some companies to reduce undifferentiated friction, even though that means giving up some proprietary revenue opportunities in favor of accelerating industry-wide deployment. I’ll be writing a lot more about this in the near future.
Samsara’s $2 Billion ARR
Many successful IoT and edge companies share a common business strategy: Apply horizontal technologies to vertical markets. Use high tech, but sell high ROI. The biggest names in AI should take a page from this playbook.
Samsara is the paradigm example. The company ingests common technical denominators from the physical economy — sensor readings, image data, material movement, schedules, workflows, and compliance — into its Operations Cloud and productizes that rich vertical context into applications that drop directly into enterprise operations.
This ROI-focused strategy explains why Samsara’s just-reported Q1 FY2027 annual recurring revenue (ARR) is nearly $2 billion, heavily anchored by a 37% growth acceleration in its large enterprise ($100,000 or more) customer base. Regular readers of these pages recognize the stairstep shape of the company’s ARR bar chart, which has remained consistent for years and reflects the cumulative benefits of sharp customer focus.
This quarter, the best evidence of large enterprise customer growth is Samsara’s biggest deal for its Connected Asset Maintenance offering to date — a software-only deployment across Hertz’s North American rental fleet. Samsara’s Operations Cloud pulls in data from existing vehicle management systems and OEM telemetry streams, translating maintenance logs, operational records, and diagnostic data into automated workflows that optimize decisions about vehicle service, reliability, cost, and resale value.
I’m attending Samsara’s annual Beyond conference and analyst day later this month, and I’ll have a lot more to say about the company’s strategy after spending quality time with the executive team.
At Computex last week, I got an opportunity to demo many of the experiences built for NVIDIA’s RTX Spark and found that NVIDIA and Microsoft worked together with some of the world’s leading independent software vendors (ISVs) to enable either native or highly optimized emulation experiences. Adobe is already on board, partnering with NVIDIA to bring Photoshop and Premiere Pro to RTX Spark with native support for Arm. Epic Games and Blender both demonstrated how beneficial the shared memory architecture can be when paired with powerful CPUs and GPUs. NVIDIA also showed gaming and coding applications of RTX Spark to demonstrate its versatility and its fit for many markets with AI-accelerated workloads.
At the Microsoft Build conference, which also happened last week, Microsoft announced many significant changes to Windows to enable agentic AI experiences on Windows PCs, with a focus on enabling the RTX Spark. Microsoft also said it would bring all the Windows quality improvements it has been working on this year to RTX Spark when it launches this fall. I think this makes a lot of sense because both NVIDIA and Microsoft want to put their best foot forward with a new, highly anticipated platform like this one. Microsoft also introduced the MXC (Microsoft Execution Containers) SDK along with native OpenClaw support and OpenShell via MXC.
IonQ’s “Walking Cat” Is a Blueprint for Fault-Tolerant Quantum Computing
Because fault-tolerant quantum computing is within sight and also because quantum technology is moving from theory to the market, funding in the sector has greatly increased. The U.S. government has been pouring billions of dollars into quantum, with more billions coming from private investment.
IonQ is considered a leader for its “Walking Cat” architecture, an end-to-end blueprint for a buildable, fault-tolerant quantum computer (FTQC). The blueprint’s significance is that it is comparable to John von Neumann’s 1945 EDVAC report, which laid the groundwork for modern classical computing. Walking Cat covers the design of the compiler, the logical architecture, and the microarchitecture. Theoretically, it can support millions of gates across hundreds of logical qubits. It is designed to be the foundation for IonQ’s 2030 goal of 2 million physical qubits and 80,000 logical qubits. (I went into more detail in my Research Brief on IonQ published earlier this year.)
While the architecture has an odd name, it reflects two core physical mechanisms. “Cat states” are specialized quantum resource states that probe logical qubits for errors without destroying their fragile quantum state, as in Schrödinger’s famous thought experiment and Peter Shor’s 1996 error correction work.
The term “Walking ions” describes the physical movement of ions across a quantum charge-coupled device (QCCD) chip. Ions travel between gate zones for operations and optical zones for measurement. That enables ion traps’ important any-to-any qubit connectivity.
HMRS Framework
Instead of mixing incompatible error-correcting codes, IonQ has designed four engineering principles called HMRS:
Real-World Validation
IonQ’s logical architecture consists of five components: memory blocks, magic factories, cat factories, Bell factories, and a qubit factory, which handles ion replacement, provides full chip connectivity, and supports parallel execution. Shor’s algorithm, plus a Heisenberg Hamiltonian-based materials science simulation, was used to validate the blueprint.
The architecture for a 10,000-physical-qubit system is projected to solve classically intractable materials science problems within a month. If that happens, it would signal the official arrival of fault-tolerant quantum computing.
At Computex 2026, NVIDIA unveiled the new RTX Spark processor, specifically intended to extend NVIDIA’s AI leadership further into client devices. While NVIDIA already has dominant market share in gaming and professional graphics, all those products are GPUs paired with AMD or Intel CPUs. They are also beholden to those chip companies’ platforms and software to deliver AI. With RTX Spark, NVIDIA delivers a complete SoC with shared memory to enable AI delivery much closer to what NVIDIA’s datacenter products can achieve. While RTX Spark is very similar to DGX Spark, there are some differences, namely the lower power limit (DGX Spark’s is 240W) and the thin-and-light laptop form factor, which also limits cooling. There is also now an NPU (neural processing unit) in the RTX Spark, making it Windows Copilot+-compatible. NVIDIA says it will ship RTX Spark systems this fall, but so far there’s no clear pricing from anyone. RTX Spark is clearly NVIDIA’s approach to compete with Apple, and it will be interesting to see how competitive it is with the likes of the Snapdragon X2 Elite Extreme, Apple M5 Max, and other high-performance agentic AI SoCs.
At Computex 2026, AMD also announced yet another re-release of its popular 5800X3D (we’re on 9950X3D now), which runs on its last-generation AM4 socket. This offers people with PCs that might feel a bit dated an opportunity to upgrade the CPU to improve performance without replacing anything else. Some people have balked at the $349 price, especially since the new 7700X3D is priced at $329. (The 7700X3D is essentially a downclocked 7800X3D, i.e. not really a new chip but rather a new SKU of an older one.) These are all welcome additions to AMD’s lineup for gamers struggling to afford building new PCs. But the reality is that gamers want new products that build on the Ryzen 9000 series’ success and not just refreshes of older products. On that front, AMD didn’t really launch any new CPUs so much as it committed to continuing to use the current AM5 socket through 2029, while also celebrating 10 years of AM4 support with the two new chips. On the GPU front, AMD also announced the 9070 GRE, a more accessible version of the 9070 XT, starting at $549. On top of all that, AMD announced a new memory spec called AMD EXPO ULL, designed to deliver ultra-low latency for the EXPO standard, with new memory kits coming out this month. AMD says these kits increase frames per second (FPS) by an average of 4% — likely in low-resolution, high-FPS games that are mostly CPU-bound.
Six Five (Patrick Moorhead)
Unless otherwise noted, our analysts will be attending the following events in person.
August events coming soon.
September events coming soon.
December events coming soon.
Founder, CEO and Chief Analyst | + posts
Patrick Moorhead is the founder, CEO, and chief analyst of Moor Insights & Strategy. His big-picture view of technology is grounded in more than 20 years as an executive leading strategy, product management, product marketing, and corporate marketing functions at NCR, AT&T, Compaq, and AMD. He has shared his expertise in areas from silicon to infrastructure to enterprise SaaS and everything in-between in thousands of national broadcast appearances (CNBC, Yahoo Finance), articles (Forbes, CIO), research-based analyses, and podcast episodes. Today, he has 100+ CXO-level advisory clients and is often ranked the #1 technology industry analyst by ARInsights.
Anshel Sag is Moor Insights & Strategy’s in-house millennial with over 18 years of experience in the IT industry. Anshel has had extensive experience working with consumers and enterprises while interfacing with both B2B and B2C relationships, gaining empathy and understanding of what users really want. Some of his earliest experience goes back as far as his childhood when he started PC gaming at the ripe of old age of 5, building his first PC at 11, and learning his first programming languages at 13.
Mike Leone is a principal analyst at Moor Insights & Strategy covering data platforms and analytics, data infrastructure and storage, and data governance and enterprise data strategy. He brings 15 years of analyst experience from his work at Enterprise Strategy Group, where he rose to practice director for data management, analytics, and AI. Mike's work is grounded in a strong technical and strategic foundation, including early roles in software and hardware engineering.
Mel Brue is vice president and principal analyst covering modern work and financial services. Mel has more than 25 years of real tech industry experience in marketing, business development, and communications across various disciplines, both in-house and at agencies, with companies ranging from start-ups to global brands. She has built a unique specialty working in technology and highly regulated spaces, such as mobile payments and finance, gaming, automotive, wine and spirits, and mobile content, ensuring initiatives address the needs of customers, employees, lobbyists and legislators, as well as shareholders.
Matt Kimball is a Moor Insights & Strategy senior datacenter analyst covering servers and storage. Matt’s 25 plus years of real-world experience in high tech spans from hardware to software as a product manager, product marketer, engineer and enterprise IT practitioner. This experience has led to a firm conviction that the success of an offering lies, of course, in a profitable, unique and targeted offering, but most importantly in the ability to position and communicate it effectively to the target audience.
Senior Analyst-in-Residence | + posts
Bill Curtis is the Moor Insights & Strategy Analyst in Residence for large-scale Internet of Things systems. Bill helps enterprises design distributed solutions that integrate the full end-to-end IoT stack from real-world devices to analytics.
Jason Andersen is vice president and principal analyst covering application development platforms, technologies, and services. Jason brings over 25 years of experience in product management, product marketing, corporate strategy, sales, and business development at Red Hat, IBM, and Stratus to his work for MI&S and its advisory clients. Working both in the field and in the headquarters of some of the most innovative technology companies, Jason has a wealth of experience in building great products and driving their adoption across a broad spectrum of industries and use cases.
Paul Smith-Goodson is the Moor Insights & Strategy Vice President and Principal Analyst for quantum computing and artificial intelligence. His early interest in quantum began while working on a joint AT&T and Bell Labs project and, during 360 overviews of Murray Hill advanced projects, Peter Shor provided an overview of his ground-breaking research in quantum error correction.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。