Our analysts have done a lot of work to contextualize the developments coming out of Computex in Taipei earlier this month.
(Photo by Timo Volz on Unsplash)
No rest for the weary just yet, because the team is very busy from coast to coast this week before we enjoy a bit of a travel respite. Matt and I are attending HPE Discover in Las Vegas, where Matt will also attend Pure Accelerate, and Mel will be on-site at InfoComm. Jason will be in New York for the AWS Analyst Summit; Anshel will attend AWE 2026 in Long Beach, California; Mike will be in San Francisco for the Databricks Data + AI Summit; and Bill will attend the Connectivity Standards Alliance Unify 2026 event in Austin. For news and hot takes from these events — as well as fresh insights and research throughout the summer — be sure to follow us on X and LinkedIn.
Last week, Moor Insights & Strategy analysts’ perspectives appeared across major business and technology outlets, including Yahoo Finance, Network World, Business Insider, TechNewsWorld, Reuters, Fierce Network, CIO, Barron’s, Gizmodo, IEEE Spectrum, and InsiderPH. Coverage touched on software and chip earnings reactions at Adobe and Intel, AWS’s datacenter efficiency claims, Anthropic’s latest AI models, Apple’s new AI features and Siri strategy, Apple and Google’s private AI infrastructure collaboration, emerging AI‑native document formats, Meta’s smart glasses trajectory, NVIDIA’s RTX Spark AI PCs, and Zoom’s continued push into conversation‑driven AI for productivity.
Our MI&S team also published 11 deliverables — 1 Forbes Article, 1 Research Paper, 4 Research Notes, 2 Analyst Insights, 2 Field Notes, and 3 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
New Research on SAP
A new piece of MI&S research is available now on the Moor Insights & Strategy website. At SAP Sapphire 2026 in Orlando, SAP unveiled a more aggressive AI strategy than expected, built around its vision of the “Autonomous Enterprise” and a new three-layer architecture — the SAP Business AI Platform as the foundation, the SAP Autonomous Suite in the middle, and Joule Work at the user-engagement layer.
Our take is that SAP is positioning itself as a post-AI software company, not bolting AI onto existing apps but restructuring its architecture, go-to-market, and R&D around AI in ways most SaaS peers haven’t attempted. Three proof points stand out:
The open questions are whether SAP can earn the developer and ISV mindshare that platform competition demands, and how it manages a partner ecosystem whose application-layer economics AI seems sure to compress. Read the full analysis now.
New Video Short on Microsoft’s Approach to Agentic Evaluations
Over the spring conference season, I attended a lot of events, and my output of short videos went down. But I am happy to report that I have restarted making short videos for the Six Five YouTube channel. Last week I posted a recap about Microsoft’s take on evaluations becoming a more vital part of agentic apps. It’s an interesting approach, given that we have seen the rise of a number of new techniques to improve context such as memory and skills. But evaluations happen after the inference as a means to check the output versus create it. And while evaluations will invariably increase token spend, they could lead to less rework and better results overall. Check out the video here.
Emerging Standards for Better AI Context
Sticking with the theme of improving context for AI and agents, last week also saw news about new cross-vendor standards to improve AI results. We also are seeing some debate as to the value of HTML as an input, as opposed to markdown files. Both vectors are important in terms of how we interact with AI and achieving better answers from it. However, I raised the point in this Computerworld piece written by Paul Barker that these standards also need to be delivered on the fly and enable humans to use their own natural language — and not introduce a lot of heavy syntax or rules.
Claude Fable
Finally, as a big user of Claude Desktop and Claude Cowork, I have been following the news about Claude Fable and some of the tests people are performing on it, such as head-to-head coding challenges and differences in costing. (There’s a lot of that out on YouTube and the like.) But I have two big concerns. The first is that the data privacy terms are different, as Pat Moorhead pointed out on his socials. That should be enough to keep business users from using it for anything but testing. But to be honest, I am still wondering more about a second point — whether the guardrails put in place will be sufficient and, if someone were able to use Fable in an unexpected and calamitous way, who would be responsible? So, while it’s a good thing that the Anthropic team gave us “Mythos lite,” I’d be very cautious about using it in production.
New Research on SAP
A new piece of MI&S research is available now on the Moor Insights & Strategy website. At SAP Sapphire 2026 in Orlando, SAP unveiled a more aggressive AI strategy than expected, built around its vision of the “Autonomous Enterprise” and a new three-layer architecture — the SAP Business AI Platform as the foundation, the SAP Autonomous Suite in the middle, and Joule Work at the user-engagement layer.
Our take is that SAP is positioning itself as a post-AI software company, not bolting AI onto existing apps but restructuring its architecture, go-to-market, and R&D around AI in ways most SaaS peers haven’t attempted. Three proof points stand out:
The open questions are whether SAP can earn the developer and ISV mindshare that platform competition demands, and how it manages a partner ecosystem whose application-layer economics AI seems sure to compress. Read the full analysis now.
New Video Short on Microsoft’s Approach to Agentic Evaluations
Over the spring conference season, I attended a lot of events, and my output of short videos went down. But I am happy to report that I have restarted making short videos for the Six Five YouTube channel. Last week I posted a recap about Microsoft’s take on evaluations becoming a more vital part of agentic apps. It’s an interesting approach, given that we have seen the rise of a number of new techniques to improve context such as memory and skills. But evaluations happen after the inference as a means to check the output versus create it. And while evaluations will invariably increase token spend, they could lead to less rework and better results overall. Check out the video here.
Emerging Standards for Better AI Context
Sticking with the theme of improving context for AI and agents, last week also saw news about new cross-vendor standards to improve AI results. We also are seeing some debate as to the value of HTML as an input, as opposed to markdown files. Both vectors are important in terms of how we interact with AI and achieving better answers from it. However, I raised the point in this Computerworld piece written by Paul Barker that these standards also need to be delivered on the fly and enable humans to use their own natural language — and not introduce a lot of heavy syntax or rules.
Claude Fable
Finally, as a big user of Claude Desktop and Claude Cowork, I have been following the news about Claude Fable and some of the tests people are performing on it, such as head-to-head coding challenges and differences in costing. (There’s a lot of that out on YouTube and the like.) But I have two big concerns. The first is that the data privacy terms are different, as Pat Moorhead pointed out on his socials. That should be enough to keep business users from using it for anything but testing. But to be honest, I am still wondering more about a second point — whether the guardrails put in place will be sufficient and, if someone were able to use Fable in an unexpected and calamitous way, who would be responsible? So, while it’s a good thing that the Anthropic team gave us “Mythos lite,” I’d be very cautious about using it in production.
At its WWDC event, Apple launched the new Apple software universe featuring new Gemini-enhanced foundational models and Siri AI. The interesting part is that there are different tiers of on-device AI, with the best model recommended only for the latest phones. Apple is really leaning into the on-device-first AI strategy with the new Siri, calling on Private Cloud Compute only when it needs to. While I do think this helps Apple catch up to the competition, many of these features were things that Apple promised years ago — and that should have come out before now. In any case, Apple still has lots of room for improvement with features like Spatial Reframing, which borrows Gaussian splat technology from visionOS to help users reframe and regenerate images.
The June 2026 IBM article titled “AI understands you? Yeah, right” explores the problem of artificial intelligence accurately recognizing sarcasm. That is a problem especially associated with human-facing technologies such as agentic customer service bots.
The Core Problem with AI and Sarcasm
It seems odd to me that LLMs are good at analyzing sentiment and emotions, but they miss the mark by a wide margin when detecting sarcasm. The theory is that machine learning works by relying on the probability of the output based on prior wording. Sarcasm operates differently because it operates on intentional contradictions. The actual meaning is completely opposite to what the literal text implies. However, text alone can’t be used to infer sarcasm.
To address this problem, IBM Research and others have begun to use multimodal datasets that include text, as well as audio with tone, pitch, and speech rate. The dataset also needs a video for facial expression analysis.
Unfortunately, there are additional problems associated with creating these datasets. The Multimodal Sarcasm Detection Dataset (MUStARD) is a prominent and high-quality multimodal sarcasm dataset composed of curated sitcom clips from shows such as Friends. Despite its quality, it has only about 700 clips. To increase its size, IBM collaborated to double the number of clips and relabel it as MUStARD++. That effort also added another valuable feature of emotional intensity.
IBM isn’t the only company performing modifications to the dataset. Other researchers are also scaling the data using deep-learning back-translation to generate synthetic audio voices. They are also mining real-world conversations using podcasts such as PodSarc.
Researchers have picked up on additional clues that may signal sarcasm. For instance, they have begun tracking eye-gaze behavior. They have learned that humans pause and skip backward more often when deciphering written sarcasm. Hopefully this kind of analysis can be applied to videos as well as text.
Sarcasm is an important issue that must be addressed as we move agentic AI into more applications. Ultimately, solving the “sarcasm conundrum” is essential for the future of intuitive, human-like AI, ensuring that a customer saying “I love being kept on hold” is met with a troubleshooting solution rather than a robotic “Thank you!”
For a long time, the term “sharing” meant handing someone a dataset. In the buildup to Databricks’ Data + AI Summit being held this week, the company announced OpenSharing. The idea is that now you can share the AI itself, a model, or an agent skill, and do it through one open protocol with governance baked in so the rules travel along with whatever you send. This means you hand a partner a model without dragging them onto your platform, and nobody needs to rebuild the access rules once it lands with them. Databricks kept it open, too, putting the protocol under the Linux Foundation and getting OpenAI behind it. The company wants this to be the default for how companies share AI as agents take over, the way Delta Sharing became a default for sharing data.
ClickHouse and Datadog have also linked up to keep full-fidelity observability data in play. Run agents in production and they kick out a flood of telemetry — every trace, every evaluation, every move an agent makes — and the usual instinct is to sample that down to keep the bill sane. Problem is, the thinned-out version leaves you blind right when an agent goes sideways. ClickHouse has quietly become the place where a ton of that data lives, so Datadog leaning on it to hold the whole record rather than a slice tells you where this is headed. Once agents start making their own calls, being able to go back and replay exactly what one of them did is how you keep them honest.
The hardest problem in enterprise AI used to be getting your hands on compute — the scramble over who had the GPUs. Now it’s getting your own data ready for models to use, and that’s the problem NetApp built its recent analyst summit around. I came away thinking this is a company that’s done with being boxed in as a storage vendor. Its pitch is to make your data discoverable, governed, and safe right where it already lives, so the AI inherits the controls you trust instead of you needing to copy everything into some new stack. No rival really matches that hybrid reach, with ONTAP running on-prem and sold as a first-party service inside all three big clouds. The catalog and data engine NetApp has layered on top are the newer pieces, and it still has to prove they can reach across a messy, multi-vendor estate at scale.
The piece I think gets underrated is cyber resilience. NetApp bakes ransomware detection and recovery into the storage layer itself and backs it with a recovery guarantee, and that matters more the moment that same storage starts feeding your most sensitive data into AI. You can’t bolt protection on after the data is already flowing into models, so it has to live in the layer the data sits in. A wave of newer vendors has shown up with storage built for AI speed, and the performance is real. But keeping enterprise data safe and recoverable for years is a different discipline, and it takes years to earn that kind of trust. Speed in feeding the model matters, sure. But whether you’d trust the platform with the data in the first place is what actually closes the deal.
For years Oracle got written off as the “other” cloud, or just a legacy database company. The company’s latest financial results make that hard to keep saying. Its cloud infrastructure business nearly doubled from a year ago, and it’s forecasting even faster growth next year, which tells you more than the giant backlog number everyone quoted. For anyone building a data and AI stack, Oracle has earned a real spot on the shortlist, especially now that its database runs natively inside AWS, Azure, and Google Cloud, wherever your workloads already live. I’d stay clear-eyed about the bet, though. A big chunk of that huge backlog is prepaid AI capacity rather than steady software revenue, and the buildout is burning cash, which means Oracle has to keep converting it into real revenue. So far, it is.
Moving workloads onto any of these clouds is its own fight, and AWS just removed one of the oldest reasons not to. Paying twice for the same SQL Server license kept a lot of those workloads parked on-prem for years, but letting customers reuse licenses they already own on RDS finally kills that penalty. I weighed in on it for InfoWorld. The cost savings grab the headline, but what most shops really want is control over when and how they move. Getting that data closer to AWS analytics and AI is where plenty of teams want to be, but sitting next to those services doesn’t make the data ready for agents. The modeling and governance work is the same as it ever was; dropping the license barrier just gets you to the starting line. The hard part was never moving the data — it’s getting it ready to be used.
Microsoft says that it is completely rethinking the Xbox business model as the gaming division is expected to report a meager 3% “accountability margin” this fiscal year. (This is an internally defined metric describing a business unit’s contribution to the overall performance of the company.) Xbox and Microsoft, it would seem, have been quite poor in monetizing the tens of billions of dollars Microsoft has spent on gaming studios.
I believe that this underperformance is a consequence of the increased cost of Game Pass, a lack of first-party exclusives, losing the console war to Sony, and falling back on PC gaming. Microsoft has a long uphill battle here, and it looks like the company will need to lay off even more workers and reprioritize growth and profits. Ultimately, if Xbox and Microsoft start building games with new and interesting IPs, the business could see some growth. I believe that way too many studios have milked old IP until there was nothing left, and I think a lot of that has happened in the case of Xbox as well.
Siemens announced Intelligence Center X at Realize LIVE Detroit on June 1, 2026. It’s an industrial AI orchestration platform that combines three existing Siemens products — the Mendix low-code platform, plus Graph Studio and AI Studio from the RapidMiner portfolio acquired with Altair in 2024 — with a new layer of pre-built industrial ontologies that supply domain context to AI agents. Intelligence Center X connects enterprise data (e.g., SAP, Snowflake) with operational data (Siemens Industrial Edge) under shared governance. The vendor-cited results — an 85% reduction in production-issue resolution time, 6,000 hours of manual work recaptured, customer complaint resolution compressed from five days to less than one — come from a single named customer, Brazilian flat-glass manufacturer Vivix Vidros Planos, which deployed roughly 30 Mendix applications connecting SAP S/4HANA, Siemens Industrial Edge, and Snowflake.
This is a name-and-claim play. The pilot-to-scale gap is now a canonical industrial AI problem, and Siemens has named it. However, that is not the same as solving it. Multiple incumbents — Rockwell, ABB, Honeywell, Emerson, GE, Hitachi Digital — sit at the same intersection of operational software and customer relationships, and none dominate the category. The vendor-claimed metrics need independent validation in production conditions before claiming victory. What Intelligence Center X delivers, regardless of competitive outcome, is validation of the problem framing: Industrial AI orchestration is now an explicit software category. The race is on, and there might not be a single winner.
Advantech released WISE-Edge Developer Architecture (WEDA) at the Advantech World Partner Conference held in parallel with Computex. WEDA is a multi-silicon abstraction layer for industrial physical AI: hardware-accelerated containers with direct GPU/NPU/DSP passthrough across AMD, Intel, NVIDIA, NXP, Qualcomm, and Rockchip platforms. The architecture exposes unified APIs, Model Context Protocol endpoints, and packaged AI skills, and it supports OpenUSD and DTDL for digital twin and asset modeling.
Industrial operations are inherently heterogeneous, with multiple silicon vendors, software frameworks, and form factors. Deployment cycles span decades, and operations — not IT — governs procurement. WEDA accepts this extreme heterogeneity as the starting condition rather than an obstacle to avoid. Vertical full-stack systems like Jetson are great for greenfield robotics, where a customer standardizes on one supplier from the start. However, vertical stacks do not serve brownfield factories running Intel, NXP, and Qualcomm platforms. WEDA targets the manufacturing world as it exists today. The MCP citation in the specification is the quiet tell: Industrial orchestration is adopting agent-protocol language, with or without semantic harmonization underneath. It’s a “data as-is” situation, with heterogeneity at the core. Advantech wants to be the runtime layer for multi-silicon physical AI.
Neura Robotics announced a Series C of “up to” $1.4 billion at a $7 billion valuation on June 10, led by Tether, with co-investors including NVIDIA, Qualcomm Ventures, Amazon, Bosch, Schaeffler, the European Investment Bank, and imec.xpand. The headline number is “up to,” contingent on performance milestones. The investor list maps onto the physical AI supply chain: NVIDIA Cosmos for training and simulation; Qualcomm Dragonwing IQ10 as the on-robot compute platform; AWS as the cloud and distribution backbone for Neuraverse; Bosch and Schaeffler positioned as Tier 1 industrial customers and manufacturing partners; and EIB for European sovereign capital. The Neura story is one of partner diversity, and the company has assembled a financial coalition to match.
The unanswered question is architectural. Each strategic partner brings a distinct software ecosystem — Cosmos, Isaac, and GR00T from NVIDIA; the Dragonwing stack from Qualcomm; AWS services from Amazon; industrial automation toolchains from Bosch and Schaeffler. Neuraverse is the integrating layer, but the company has not described how it remains coherent as independent partner stacks evolve faster than any single integrator can absorb. Integration stories sound great on paper, so attracting capital is the easy part. But history has not been kind to large-scale integration plays in operational (industrial) situations. We’ll be tracking Neura to see how Neuraverse evolves.
As the World Cup has gotten underway over the past few days, we have started to see some of the newest tech enabling better user and fan experiences. Opensignal visited 11 World Cup host cities and tested each market’s network speeds; based on its methodology, it crowned T-Mobile the fastest and best 5G network for World Cup fans. This especially makes sense when you consider that the stadiums hosting World Cup matches tend to be larger and in bigger cities, and that’s really where T-Mobile’s 5G network thrives against AT&T and Verizon.
As the World Cup games began to be played, we got glimpses of some new technical capabilities. While Lenovo’s presence was felt across the stadiums’ banners and feeds, the company was also behind a new video angle — coming from a 5G camera worn by each referee and powered by Qualcomm technology. This new angle shows a field-level shot that automatically stabilized, helping you see the action as if you were the referee on the field. While details about this body-worn referee camera are scant, Qualcomm did take credit online for its 5G mmWave capabilities. The 5G mmWave is likely being supported by the official FIFA 5G partner, Verizon, which has ample mmWave coverage, especially around stadiums.
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.
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.
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.
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.
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.
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.
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.
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.
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。