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.
One of the biggest tech challenges facing enterprises today is moving from limited pilots to full production at scale with AI. Sure, advances in AI models themselves continue to be important, but in practical terms it’s often the subtler operational stuff that matters more — things like data governance, auditability, data quality, and scalability. That’s where the rubber meets the road for implementing enterprise AI in a way that matters for business operations and results.
As discussed in Melody Brue’s article, Box Extract applies AI to automatically capture key fields as structured metadata, turning unstructured content into searchable, reusable data. (Credit: Box)
This practical aspect is something our software-focused analysts have been covering for a while now. And we saw fresh examples of it last week when Melody Brue published a day-of-release Research Note about Box Extract and then a related Forbes piece looking more broadly at how different vendors (Box, Hyland, Microsoft, Salesforce, Google Cloud, OpenText, and Workiva) are approaching the challenge of extracting unstructured content, governing it, and parsing it for agentic AI. At the same time, Robert Kramer published a Forbes piece analyzing the findings of Salesforce’s recent State of Data and Analytics Report — again with a key emphasis on harnessing unstructured enterprise data to drive AI. Meanwhile, Jason Andersen is wrapping up a detailed treatment on how enterprises can better scale AI agents — both so they function better and to control costs. Look for that one to be published soon.
We’re three years in on the generative AI revolution that started when OpenAI released ChatGPT. Today, the practical necessities of applying AI to real-world business problems — in ways that perform reliably, cost-effectively, and on cue, just like any other area of enterprise IT — are more important than ever. I know our analysts, in hardware and software alike, will continue to stay on top of these issues.
This week, I am in Davos, Switzerland, at the World Economic Forum, meeting with global executives and filming “The View from Davos” Six Five videos. If you or any of your executives are in Davos, please reach out. Jason is in New York for Microsoft’s AI Tour (and Robert will join virtually). Anshel is attending the MIT Reality Hack in Cambridge, Massachusetts.
We already have a number of events on the calendar for 2026. As you look ahead, please let us know if you’ll be attending any of the same events so we can say hello face-to-face.
Over the last week, MI&S analysts were quoted in multiple international publications, including Interactive Crypto, InfoWorld, Business World, Fierce Network, TechTarget, CSO, Android Central, Network World, IGN, The Verge, CRN, and Business Insider. The media sought our insights on AMD, Anthropic, Apple and Google, IBM Sovereign AI, Iran’s internet shutdown, Meta VR, Microsoft’s community-first infrastructure, PC gaming, RAM, XREAL, Intel, and NVIDIA.
Our MI&S team also published 13 deliverables — 4 Forbes Articles, 6 Research Notes, 1 Analyst Insight, 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
Last week I got the chance to attend the MongoDB.local event in San Francisco. Keeping true to its developer roots, Mongo hosts 20 smaller, more technically oriented events per year, as compared to the one or two massive events that many of its peers and larger companies host. There are upsides and downsides to this approach. The degree of intimacy, quality conversations, and exposure to the technical and executive leadership of the company is unparalleled. But sometimes it also means that, from a marketing perspective, bigger announcements or statements of direction may not get the amplification they deserve.
The San Francisco event was important because it was the first MongoDB.local to be attended by new CEO CJ Desai, and his keynote kicked off with the declaration that MongoDB is a data platform. This declaration was based upon the unification and integration of its search and embedding capabilities (acquired last year) into the Atlas platform. This is significant in MongoDB’s evolution from its origin as an open source database loved by developers into a SaaS database platform (Atlas) and now a more diverse data platform targeted at AI and agents. Desai made a strong case for the technical aspects of this pivot. But, of course, the real question will be whether customers will come along on the journey and expand their usage to include these new services. My sense is that it is still a work in progress. That said, there were also naysayers when MongoDB migrated away from the local open source database model, and it then grew its revenues by 10x with Atlas, which I discussed in the piece I wrote a few months ago after MongoDB.local in New York.
There were multiple announcements last week about AI sovereignty, which is becoming an increasingly hot topic. While there were announcements in this vein from Oracle and AWS, the most interesting announcement for me was from IBM with its preview of IBM Sovereign Core. True to its hybrid roots, IBM has chosen to abstract the functionality so sovereignty and compliance can be delivered across multiple cloud and on-premise infrastructures. This is a big difference from the cloud-specific offerings from Oracle Cloud Infrastructure, Google Cloud, and AWS. Another unique aspect of this is that IBM is targeting the managed service provider ecosystem with this solution. That’s interesting because, if you are an MSP and you want to provide sovereign services to customers today, you are kind of stuck with using one cloud, which may be too limiting for your customers. So, the positioning is quite interesting. But what I am finding most appealing are the direct cost benefits of a solution like this. Regulatory and compliance issues are a big cost problem for IT now and will surely only get worse with generative AI since it creates new data on the fly in real time. That is something we all have not dealt with yet from a regulatory perspective.
As IBM’s new offering is still a preview, there are still many technical details being worked out, but from what we see so far, it appears that Sovereign Core will end up being a composite of different solutions under the IBM umbrella (notably Red Hat OpenShift, HashiCorp Vault, and IBM Concert). We expect to see a full launch in the springtime, and I expect that CIOs will be quite interested to see how this take on a big challenge will work out. For more information, you can check out this great article from Beth Pariseau from TechTarget.
Now that we have passed the holiday season, many cloud, AI, and software vendors are gearing up for what could be a very interesting 2026. They are also gearing up for springtime events (though it seems like we just finished the fall events!). As one example, while I was in San Francisco to attend the MongoDB event, I got to drop by Salesforce Tower to meet with some of the Salesforce team. I am not going to spill any tea on the specifics of the conversation, but I will say that we had interesting discussions about how some AI features and capabilities are a bit ahead of the curve — which is leading to two new challenges. The first is a lack of broad user understanding of AI because the pace of AI innovation has been so relentless that people simply cannot keep up. The second is how some features may need more maturation to help the most users.
It was a frank and very encouraging conversation. Disruptive technology is always a major challenge for product teams because it is so hard to separate the signal from the noise when it comes to market feedback. But I got a good sense that Salesforce will continue to keep pushing its agent vision forward. I am looking forward to hearing more from them at the Salesforce TDX developer conference in April.
Last week I got the chance to attend the MongoDB.local event in San Francisco. Keeping true to its developer roots, Mongo hosts 20 smaller, more technically oriented events per year, as compared to the one or two massive events that many of its peers and larger companies host. There are upsides and downsides to this approach. The degree of intimacy, quality conversations, and exposure to the technical and executive leadership of the company is unparalleled. But sometimes it also means that, from a marketing perspective, bigger announcements or statements of direction may not get the amplification they deserve.
The San Francisco event was important because it was the first MongoDB.local to be attended by new CEO CJ Desai, and his keynote kicked off with the declaration that MongoDB is a data platform. This declaration was based upon the unification and integration of its search and embedding capabilities (acquired last year) into the Atlas platform. This is significant in MongoDB’s evolution from its origin as an open source database loved by developers into a SaaS database platform (Atlas) and now a more diverse data platform targeted at AI and agents. Desai made a strong case for the technical aspects of this pivot. But, of course, the real question will be whether customers will come along on the journey and expand their usage to include these new services. My sense is that it is still a work in progress. That said, there were also naysayers when MongoDB migrated away from the local open source database model, and it then grew its revenues by 10x with Atlas, which I discussed in the piece I wrote a few months ago after MongoDB.local in New York.
There were multiple announcements last week about AI sovereignty, which is becoming an increasingly hot topic. While there were announcements in this vein from Oracle and AWS, the most interesting announcement for me was from IBM with its preview of IBM Sovereign Core. True to its hybrid roots, IBM has chosen to abstract the functionality so sovereignty and compliance can be delivered across multiple cloud and on-premise infrastructures. This is a big difference from the cloud-specific offerings from Oracle Cloud Infrastructure, Google Cloud, and AWS. Another unique aspect of this is that IBM is targeting the managed service provider ecosystem with this solution. That’s interesting because, if you are an MSP and you want to provide sovereign services to customers today, you are kind of stuck with using one cloud, which may be too limiting for your customers. So, the positioning is quite interesting. But what I am finding most appealing are the direct cost benefits of a solution like this. Regulatory and compliance issues are a big cost problem for IT now and will surely only get worse with generative AI since it creates new data on the fly in real time. That is something we all have not dealt with yet from a regulatory perspective.
As IBM’s new offering is still a preview, there are still many technical details being worked out, but from what we see so far, it appears that Sovereign Core will end up being a composite of different solutions under the IBM umbrella (notably Red Hat OpenShift, HashiCorp Vault, and IBM Concert). We expect to see a full launch in the springtime, and I expect that CIOs will be quite interested to see how this take on a big challenge will work out. For more information, you can check out this great article from Beth Pariseau from TechTarget.
Now that we have passed the holiday season, many cloud, AI, and software vendors are gearing up for what could be a very interesting 2026. They are also gearing up for springtime events (though it seems like we just finished the fall events!). As one example, while I was in San Francisco to attend the MongoDB event, I got to drop by Salesforce Tower to meet with some of the Salesforce team. I am not going to spill any tea on the specifics of the conversation, but I will say that we had interesting discussions about how some AI features and capabilities are a bit ahead of the curve — which is leading to two new challenges. The first is a lack of broad user understanding of AI because the pace of AI innovation has been so relentless that people simply cannot keep up. The second is how some features may need more maturation to help the most users.
It was a frank and very encouraging conversation. Disruptive technology is always a major challenge for product teams because it is so hard to separate the signal from the noise when it comes to market feedback. But I got a good sense that Salesforce will continue to keep pushing its agent vision forward. I am looking forward to hearing more from them at the Salesforce TDX developer conference in April.
OpenAI signed a deal for 750 megawatts of Cerebras compute through 2028, which amounts to about $10 billion worth of orders. This reflects OpenAI’s continued diversification of its compute capabilities across many different architectures and AI compute platforms. Cerebras has long been one of the leaders in inference performance and has pivoted toward a service-based business model that utilizes its compute to deliver low-latency and high-performance inference.
OpenAI is also planning the launch of an ad-supported free version of ChatGPT, which Sam Altman once said would be the last resort for monetization. In addition to that, the lawsuit between Elon Musk and OpenAI is expected to move forward — possibly revealing juicy details about the business and its relationship with Musk. Additionally, some companies with prominent OpenAI alums (e.g. Thinking Machines) are starting to see those people return to OpenAI for one reason or another.
Apple and Google signed a deal for Google to provide Gemini models for Apple to train and customize for its own purposes, to the tune of at least $1 billion a year. Apple will run these models on its Private Cloud Compute infrastructure, maintaining the private computing model that Apple created for its customers. These models are planned to power Apple Intelligence and empower the upgrade to Siri that everyone has been waiting for. The structure of the agreement is more of a cloud-computing contract than a licensing deal, which means that if Apple exceeds the usage it agreed to, it could potentially pay more — as much as $5 billion. The improvements from the new Gemini-based model are expected in March or April 2026, which significantly shortens Apple’s timeframe to enhance Apple Intelligence — and will likely improve model quality as well.
There have been a number of papers published lately promoting a common narrative about malicious AI behavior. Anthropic, for instance, has published about some very interesting instances where AI has tried to gain leverage through blackmail or other coercive behavior. A new paper from the Institute of Risk Analysis, Prediction, and Management at the Southern University of Science and Technology in Shenzhen, China — “Why AI Alignment Failure Is Structural: Learned Human Interaction Structures and AGI as an Endogenous Evolutionary Shock” — challenges the notion that malicious AI behavior comes from rogue or failing models. Instead, this paper presents the case that negative behaviors are reflections of the full documented history and record of human social interactions.
I agree with the researchers’ contention that LLMs do not have moral reasoning. Instead, they postulate that models internalize human interactions that range across a spectrum from fair market trades to coercive power plays that are all found in human data. According to the paper, “alignment failure” is actually a model generalizing behaviors that humans use in asymmetric power situations. Using relational models theory (RMT), the paper suggests that blackmail and market pricing exist on the same continuum of human logic, and that AI is simply navigating that entire map of human behavior.
Other key Insights:
- Anthropomorphic View: We mistakenly believe that AI will naturally lean toward “good” behavior. That assumption fails to account for the fact that the majority of human data is probably pluralistic, messy, and filled with “bad-versus-good” coercive behavior.
- Structural versus Accidental Failure: I’m on the fence about the researchers’ belief that many of the publicized alignment issues aren’t the result of a glitch or an evil intention. However, I believe it is true that these issues are reflections and generalizations of how humans negotiate and exert their power.
- Artificial General Intelligence as a Frictionless Amplifier: The primary risk of AGI isn’t that it will turn on us (at least not now); rather, the biggest problem is that it removes cognitive frictions such as delays, errors, and human limits that usually prevent social contradictions from collapsing.
Oracle Cloud Infrastructure rolled out its OCI Generative AI service this week. The feature set is largely expected. The positioning is not.
At a basic level, OCI Generative AI is a fully managed service embedded directly into OCI, supporting chat, summarization, embeddings, code generation, and text generation. It’s accessible via API, CLI, and an interactive playground. None of that is surprising. That’s the baseline for entry.
What matters is where Oracle is drawing the line.
OCI Generative AI is designed to run inside Oracle’s commercial, sovereign, and government cloud regions, including U.S. Classified and Top Secret environments. That immediately narrows the audience — and that appears to be intentional. Oracle is not chasing the broadest GenAI adoption curve. It is targeting organizations that have largely been excluded from the first wave of generative AI because they cannot put sensitive data on shared public AI services.
This creates a different competitive posture than AWS (Bedrock), Azure (OpenAI), and Google Cloud (Vertex AI). While those platforms offer isolation controls, they are still fundamentally delivered on shared infrastructure. Oracle is leaning into dedicated, sovereign deployments as a default design choice, not an edge case. For defense, public sector, healthcare, and regulated financial services, that distinction is material.
The announcement also reinforces a broader signal from Oracle. AI is no longer a side narrative for OCI. It is becoming central to the platform’s growth strategy, tightly aligned with Oracle’s database-centric view of enterprise AI and its expanding multicloud relationships. Oracle’s rapidly growing cloud footprint gives that strategy credibility.
My take: Oracle isn’t trying to win generative AI on breadth or hype. It is letting its competitors fight that out. Rather, Oracle is positioning OCI as the AI cloud for organizations that require control, sovereignty, and trust first. For regulated industries, that may matter more than raw model choice.
Dell launched an updated PowerStore and associated PowerStoreOS 4.3 to the market last week — a release that some may see as incremental. However, it does a lot to deliver for the enterprise IT storage professional.
PowerStoreOS 4.3 is a reminder that the hardest problems in enterprise storage today aren’t theoretical — they’re operational. Power, space, staffing, resilience, and risk tolerance are all tightening as data growth and AI-adjacent workloads shift from episodic to continuous. Storage platforms don’t get credit for ambition anymore; they get judged on whether they remove friction.
The headline change — support for high-capacity QLC NVMe drives — is all about capacity and datacenter space. Packing multi-petabyte effective capacity into a 2U system directly addresses the physical constraints many enterprises are already up against. This is about doing more within fixed-power envelopes and limited floor space, not chasing peak-performance curves that matter only in ideal conditions.
I believe that while the impact of this is significant to the enterprise, support for QLC NVMe drives is a little less impressive, considering the real benefit is delivered from the increased capacity of the QLC. The real hero of this release should be, in my opinion, all of the work the company did within PowerStoreOS.
Resilience is another clear priority. Expanded native replication options, including metro-level designs and improved automation, reflect the fact that failure scenarios are no longer edge cases. Ransomware, regional outages, and operational mistakes are assumed conditions. Features like multiparty authorization reinforce that reality by treating governance and insider risk as part of the storage conversation, not an overlay.
File services also received practical attention. Enhancements around NFS support, server-side copy, and workload visibility tools speak to the unglamorous truth that file remains foundational for many enterprises. Reducing unnecessary data movement and giving administrators clearer insight into noisy or misbehaving workloads helps prevent performance issues from turning into outages.
What stands out the most for me in PowerStoreOS 4.3 is what Dell didn’t do. This isn’t a radical architectural reinvention, and it doesn’t try to reposition PowerStore as something it’s not. Instead, it tightens execution around density, resilience, and operational clarity — areas where enterprises actually feel pain.
My take: PowerStoreOS 4.3 is storage acting like infrastructure again. It prioritizes predictability over novelty and assumes modernization happens under constraint, not on greenfield budgets. That kind of pragmatism doesn’t generate hype, but it’s exactly what keeps storage from becoming the bottleneck as AI, cyber risk, and operational pressure continue to rise.
Commvault recently announced its Unified Data Vault to align with how organizations actually use AWS today for application data, backups, logs, and AI workloads. Amazon S3 has become the default place to store this data, but it is often spread across buckets with inconsistent protection and retention.
Unified Data Vault provides a Commvault-managed, S3-compatible endpoint that applications and AI workloads can write to directly. As data is written, Commvault can apply immutability, encryption, deduplication, and policy-based controls without relying on agents or custom scripts. If implemented and governed well, this approach could reduce operational effort and help limit the risk of data being stored without adequate protection, while still allowing teams to keep their existing S3 workflows.
There are also some tradeoffs to consider. Since protection and retention policies are applied as data is written, teams need to agree on them earlier rather than fixing gaps later. Cost and data lifecycle management remain important as data volumes grow, even with efficiency features in place. But if managed effectively, Unified Data Vault could enhance the reliability and recoverability of S3-based data.
Salesforce’s latest State of Data and Analytics report lines up with what I see in the field. A lot of enterprises talk about being data-driven and ready for AI, yet the foundations often aren’t there. When 84% of data leaders say their data strategy needs a reset for AI to work, that’s not a tooling problem, it’s a structure and trust problem. Fragmented systems, inconsistent governance, and questionable data quality are showing up directly as inaccurate or unreliable AI outputs.
My thought is that AI isn’t failing — data is. Most organizations are still running hundreds of applications that barely talk to each other, with valuable context stuck in silos or unstructured sources. The companies making real progress aren’t chasing more pilots or models. They’re fixing integration gaps, assigning clear ownership, improving data quality, and treating governance as ongoing operational work. I like Salesforce’s message that the speed of AI development depends on the willingness of enterprises to clean and trust their data. Read more of my analysis of the Salesforce report in my latest Forbes article.
Good news for the ERP industry as Tenneco has gone live with QAD Adaptive at its first manufacturing site. This isn’t just about upgrading ERP systems; it emphasizes that cloud-native, adaptable ERP platforms can handle live production and make real-time decisions on a large scale.
The QAD press release features a quote from me explaining why this is important for the ERP industry. I’m seeing a transition from transaction systems to platforms designed for execution, resilience, and continuous adaptation.
For more on this, check out my discussion with my cohost Melody Brue on the latest episode of the Enterprise Applications Podcast.
IFS’s acquisition of Softeon is a meaningful one as it addresses a gap between enterprise systems and warehouse execution. Softeon offers WMS and WES capabilities for high-variability areas like labor allocation, task sequencing, and automation coordination. This acquisition should enhance IFS’s capability to support end-to-end execution across manufacturing, service, and fulfillment, rather than limiting itself to planning or transactional control. Customers can benefit from enhancing the feedback loop between what they plan and what actually happens in the warehouse. Instead of having to wait until after the fact to report, this should allow customers to use AI to make smart decisions in the moment.
The challenge will be execution. Customers will pay close attention to how Softeon fits into the IFS platform, how results show up in day-to-day operations, and how well AI recommendations are managed and trusted on the warehouse floor. Change management processes will be critical, especially in environments where supervisors and frontline teams are being asked to rely on AI-driven guidance instead of longstanding manual processes. Since many enterprises already run established WMS platforms, coexistence, migration paths, and clear cost justification will matter just as much as new functionality. If IFS can show measurable gains in warehouse flow, labor productivity, and how quickly problems are identified and resolved — without adding complexity — the acquisition should deliver real value. If not, there could be a risk of this becoming an add-on with limited day-to-day impact.
Last week, GlobalFoundries announced an agreement to acquire the Synopsys ARC Processor IP Solutions business, expanding its portfolio of deeply embedded compute cores and strengthening its physical AI roadmap with ARC’s DSP, NPU, and ASIP technologies.
The timing is strategic. As embedded processors shift from bare-metal chips toward AI-native compute platforms, the ARC (classic ISA) and ARC-V (RISC-V ISA) processor families target the “deeply embedded” tier — low-power controllers, domain-specific accelerators, and tightly coupled compute blocks that run physical devices and increasingly handle small-footprint inference.
By folding ARC into MIPS, which has already transitioned to RISC-V, GlobalFoundries is building a portfolio of specialized embedded processors while keeping its foundry and system IP businesses ISA-neutral. Customers get a unified, scalable, process-tuned, and largely open-source portfolio of embedded processors for low-power products, including inference-enabled applications. Housing ARC, ARC-V, and MIPS under one roof gives GlobalFoundries pre-integrated building blocks and tools that cut development time for custom chips while maintaining a neutral manufacturing posture for Arm and other processor options.
For Synopsys, the deal sharpens focus. Divesting the ARC/processor IP business lets the company concentrate resources on security, interface, and foundation IP — areas where it holds clear leadership — while freeing up investment for EDA tools and higher‑margin opportunities across its growing silicon‑to‑systems portfolio. I expect Synopsys to double down on differentiated IP, advanced design automation, digital-twin technology, and Ansys‑powered physical system modelling and emulation.
For GlobalFoundries, the strategic upside is clear: capture more of the embedded value chain, especially in high-growth, AI-centric edge and device applications, without compromising its role as an open, broadly compatible foundry.
Arm Moves Physical and Edge AI From Concept to Platform
CES 2026 marked a clear inflection point for edge AI. The industry narrative has expanded decisively from intelligence in the cloud to encompass intelligence embedded in the physical world — systems that sense, decide, and act in real time. Robots, autonomous vehicles, and industrial equipment are emerging as first-class compute platforms, and across CES, one pattern was unmistakable: edge systems overwhelmingly use Arm processors.
This physical phase of AI is fundamentally constrained by the real world. Edge systems operate under strict power, thermal, latency, safety, and reliability limits, where predictability and continuous operation matter more than model size or peak TOPS performance. Critically, operational environments do not want or need datacenter-class systems. Edge AI in the physical world needs a different kind of compute platform designed for deployment, inference, and real-world interfaces, not model training in datacenters. Intelligence needs to be integrated quickly and pervasively into robots, vehicles, factories, and existing infrastructure. That reality favors proven, energy-efficient architectures built from the outset to host AI-native applications in real-world, safety-critical physical systems.
Arm’s CES messaging reflects this shift. Rather than positioning AI as a single workload or vertical, Arm frames physical AI as a platform transition — one that spans robotics, automotive, edge infrastructure, AI PCs, and cloud development environments. What ties these domains together is not a specific chip, but a shared compute foundation and a deeply mature software ecosystem that enables portability from the cloud to the edge to physical embodiments.
The whole silicon industry is validating this approach, not just Arm. NVIDIA’s physical AI narrative — “The ChatGPT moment for physical AI is here,” where “manufacturing plants are going to be essentially giant robots” — underscores how real-world, embodied intelligence is becoming mainstream rather than futuristic.
The result is a scalable, safety-ready foundation for AI that can move, adapt, and operate in the real world. Arm’s direction is clear: physical AI has arrived, and Arm is positioning itself as its underlying compute platform.
Last week’s Verizon outage was its second nationwide outage in the last year. Neither AT&T nor T-Mobile have had nationwide outages in the last year, which can only make it even more difficult for Verizon to get back the customers it has lost to both of those carriers over the past five years. It also hurts the image of Verizon as a dependable carrier that enterprises can depend on, especially since the last time AT&T had an outage was 2024; for T-Mobile, it was 2023. Verizon continues to try to rehabilitate its image, but outages are not helping. I think that Verizon will have to continue to do more work to build trust with both consumers and enterprises as it tries to catch up with T-Mobile and AT&T.
IBM has announced the availability of two new quantum processors that it introduced at last month’s IBM Quantum Developer Conference (which I wrote about here). Premium and Flex Plan customers have early access to ibm_miami, the first deployment of the IBM Quantum Nighthawk architecture. Nighthawk is a 120-qubit superconducting QPU implementing a square-lattice connectivity graph with 218 tunable couplers (compared to 176 in Heron), enabling increased circuit depth and support for more complex, connectivity-intensive workloads. IBM reports that since the initial preview at QDC, continued system-level optimization has resulted in the highest coherence performance across the IBM Quantum fleet to date, with a median T1 of 350 µs — surpassing previously reported QDC measurements.
ibm_miami is currently designated as an exploratory QPU and is undergoing active characterization and calibration as part of IBM Quantum’s iterative hardware learning cycles. Accordingly, the system is subject to the following temporary operational constraints:
- The default repetition time has been increased from 250 µs to 4 ms, increasing wall-clock QPU execution time for submitted workloads.
- Dynamic circuits are not yet supported, and mid-circuit measurement fidelity is currently limited.
In parallel, customers for IBM’s Premium, Flex, and Pay-As-You-Go Plans now have access to ibm_boston, the most performant IBM Quantum Heron r3 system released to date. At 100 qubits, ibm_boston demonstrates an effective Pauli error per layered gate (EPLG) of 2.15 × 10⁻³, with 57 of 176 available two-qubit gates achieving error rates below 1 × 10⁻³.
Microsoft’s new Community-First AI Infrastructure initiative, launched last week, is an early manifestation of my recent prediction that “in 2026, the sustainability conversation around datacenters and AI will broaden from isolated concerns, such as residential electricity costs and water consumption, to systemic questions about whether infrastructure and efficiency improvements can keep pace with the exponential demand growth of AI.” The initiative directly addresses one of those isolated concerns that residential electricity costs are driven by datacenter demand. In markets like Texas, grid strain from industrial power consumption has already begun flowing through to consumer bills. Microsoft’s explicit commitment to absorb the full costs of datacenter electricity so they don’t pass through to residents represents a deliberate departure from the historical pattern in which infrastructure gains accrue to operators while costs are distributed to communities.
The company is not just expanding AI infrastructure; it is also committing itself (1) to shield residential customers from datacenter-driven electricity cost increases, (2) to move to water-positive operations through leak-detection partnerships and wetland restoration, and (3) to provide granular, region-level transparency into water use and grid impact. This level of specificity — with published data by region, funded grid upgrades, and early utility collaboration on power requirements — signals that Microsoft views community acceptance as operationally critical, not optionally nice-to-have.
But the deeper strategic importance lies in the broader context: as AI demand grows exponentially, the physical infrastructure required to support it grows with it. The critical question emerging in 2026 is whether infrastructure and efficiency improvements can actually keep pace with that exponential demand growth, or if instead the tension between expanding AI compute requirements and finite local resources creates a breaking point. Microsoft’s move suggests the company has concluded that being seen as a responsible infrastructure partner, managing both environmental impact and community economic benefit, is necessary to maintain social license to build at the scale AI demand requires.
Beyond environmental metrics alone, Microsoft is tying infrastructure expansion to measurable local economic benefit through funded grid upgrades, partnerships with building trades unions and community colleges to create workforce pipelines, and continued nonprofit support via employee giving programs that directed $229 million to community organizations in 2024. The strategic bet is explicit: in a world where AI capacity demands massive physical infrastructure, long-term viability depends on communities perceiving a genuine net gain in both economic opportunity and quality of life — not just more server farms in their region.
This is important because it signals that the conversation about AI infrastructure is no longer primarily technical or financial. It’s becoming much more political and social. Communities will increasingly ask: What do we gain from hosting AI infrastructure? What do we lose? Microsoft’s early positioning suggests that companies betting on large-scale AI buildout will need answers to those questions, backed by data and commitment, not just assurances. That shift alone may reshape how infrastructure investment decisions get made across the industry in 2026.
Forbes Articles
- Semiconductor Makers Shared the Spotlight at CES 2026 (Anshel Sag)
- Using Unstructured Content for Agentic AI: A Big Enterprise Bottleneck (Melody Brue)
- Samsung Expands Its Odyssey Gaming Monitor Lineup at CES 2026 (Anshel Sag)
- Weak Data Management Hinders Enterprise AI, Salesforce Research Shows (Robert Kramer)
Research Notes
- Lenovo at CES 2026: Enterprise AI Is Becoming an Operations Problem (Matt Kimball)
- Logitech MX Master 4 Mouse Is Built for Productivity, Comfort, and AI (Anshel Sag)
- Box Extract Turns Unstructured Content into AI-Ready Data (Melody Brue)
- CES 2026 Part 2 — AMD Instinct and the YottaFLOP (Matt Kimball)
- Meta Oakley Vanguard Smart Glasses — Harnessing Meta AI for an Active Lifestyle (Anshel Sag)
- NVIDIA at CES 2026 — Vera Rubin and the Changing Shape of AI Infrastructure (Matt Kimball)
Analyst Insights Published
- MI&S Weekly Analyst Insights — Week Ending January 9, 2026 (MI&S Analyst Team)
Podcasts Published
The Enterprise Applications Podcast (Melody Brue, Robert Kramer)
Game Time Tech (Melody Brue, Robert Kramer)
Press Citations
AMD / Ryzen AI / Patrick Moorhead / Interactive Crypto
AMD’s Ryzen AI Breakthrough: Why Experts Predict a $200 Stock Price by 2026
Anthropic / Claude Code / Jason Andersen / InfoWorld
Anthropic expands Claude Code beyond developer tasks with Cowork
Apple & Google / Gemini / Patrick Moorhead / Business World
Why Apple Choosing Google Gemini For Siri Could Be Masterstroke
IBM / Sovereign AI / Jason Andersen / Fierce Network
IBM drills down to the core of the sovereign AI problem
IBM / Sovereign AI / Jason Andersen / Tech Target
IBM prepares hybrid cloud twist for sovereign AI
Iran internet shutdown, Cybersecurity / Robert Kramer / CSO
Iran’s partial internet shutdown may be a windfall for cybersecurity intel
Meta / VR / Anshel Sag / Android Central
Meta isn’t giving up on VR, but it just evaporated any goodwill it had left
Microsoft / AI Datacenters, Design, Impact on Communities / Matt Kimball / Network World
Microsoft tells communities it will ‘pay its way’ as AI data center resource usage sparks backlash
PC Gaming / Anshel Sag / IGN
2026 Is Going to Suck for PC Gaming
RAM, RTX 5070 / Anshel Sag / IGN
With the RTX 5070 Ti Reportedly Dead, the RAM Crisis Has Officially Hit Graphics Cards
Xreal & Virtue / AR Glasses, Legal Battle over Patent Infringement / Anshel Sag / Android Central
Xreal takes legal battle with Viture to the US over patent infringement claims
Xreal & Virtue / AR Glasses, Legal Battle over Patent Infringement / Anshel Sag / The Verge
Xreal sues Viture over AR glasses patent
Intel / GPU Engineering Hire / Anshel Sag / CRN
Intel Hires Qualcomm Executive to Lead GPU Engineering for Data Centers
NVIDIA / DGX Spark / Anshel Sag / Business Insider
CEO Jensen Huang stepped in as criticism mounted over a key Nvidia product launch, internal emails show
Press Release Quotes / Vendor Blogs
Lightbits Labs / AI, Infrastructure / Matt Kimball / Business Wire
Lightbits Labs Delivers Record Growth as Organizations Standardize on NVMe over TCP for AI-Ready Infrastructure
New Gear or Software We Are Using and Testing (New)
- Anker Nano Charger (Anshel Sag)
- Dell 52” 6K Monitor (Anshel Sag)
- Gemini 3 (Jason Andersen)
- NotebookLM (Jason Andersen)
- NZXT Player Three Prime Gaming PC (Anshel Sag)
- Lenovo Legion Go 2 (Anshel Sag)
- Samsung Galaxy XR (Anshel Sag)
- OnePlus 15 (Anshel Sag)
- Oppo Find X9 Pro (Anshel Sag)
- Apple M5 MacBook Pro (Anshel Sag)
- ROG Xbox Ally X Gaming Handheld (Anshel Sag)
- Google Pixel 10 Pro Fold (Anshel Sag)
- Google Pixel 10 Pro XL (Anshel Sag)
- Meta Ray Ban Display (Anshel Sag)
- Meta Oakley Vanguard (Anshel Sag)
- Miku Pro Baby Monitor (Anshel Sag)
- HP Z2 Mini G1a (Anshel Sag)
- Bleequp Ranger AI Smart Glasses (Anshel Sag)
- Naya Create Modular Keyboard (Anshel Sag)
- Alienware 18 Area 51 Gaming Laptop (Anshel Sag)
- Samsung Galaxy Fold7 (Anshel Sag)
- XREAL One Pro (Anshel Sag)
- XREAL Eye AR Glasses (Anshel Sag)
- Anonymous Tablet (Anshel Sag)
- Alienware Area 51 Desktop & 27-inch 4K OLED monitor (Anshel Sag)
- Poco F7 Ultra Smartphone (Anshel Sag)
- 2.0 Antec Flux Pro Case (Anshel Sag)
- 2.1 GeForce RTX 5070 and 5070 Ti (Anshel Sag)
- Steelseries Arctis Nova Pro Wireless Headset (Anshel Sag)
- HP EliteBook X G1a (Anshel Sag)
- Kindle Colorsoft (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.
- World Economic Forum, January 19-22, Davos, Switzerland (Patrick Moorhead)
- Microsoft AI Tour, January 21, New York City (Jason Alexander, Robert Kramer – virtual)
- MIT RealityHack, January 22-25, Cambridge, MA (Anshel Sag)
- Acumatica Summit 2026, January 25-28, Seattle (Robert Kramer)
- World Economic Forum, January 19-22, Davos, Switzerland (Patrick Moorhead)
- Microsoft AI Tour, January 21, New York City (Jason Alexander, Robert Kramer – virtual)
- MIT RealityHack, January 22-25, Cambridge, MA (Anshel Sag)
- Acumatica Summit 2026, January 25-28, Seattle (Robert Kramer)
- Western Digital Innovation Day, February 2-4, New York (Matt Kimball)
- Net Suite’s SuiteConnect, February 11, New York (Robert Kramer)
- ZohoDay 2026, February 16-19, Austin (Melody Brue)
- SAP Analyst Innovation Council 2026, February 23-25, Alpharetta, GA (Robert Kramer)
- VAST Data VAST Forward Conference, Salt Lake City, February 24-27 (Matt Kimball)
- RingCentral Analyst Summit, February 25-27, Scottsdale, Arizona (Melody Brue)
- MWC 2026, March 1-4, Barcelona (Anshel Sag)
- NVIDIA GTC, March 16-19, San Jose (Anshel Sag)
- Microsoft FabCon, March 16-20, Atlanta (Robert Kramer)
- Nutanix .NEXT ‘26, April 2-9, Chicago (Matt Kimball)
- Infor Analyst Summit, April 13-15, Atlanta (Robert Kramer)
- Adobe Summit 2026, April 20-22, Las Vegas (Melody Brue, Robert Kramer)
- Google Cloud Next, April 22-24, Las Vegas (Patrick Moorhead)
- SAP Sapphire, May 11-13, Orlando (Robert Kramer)
- Blue Yonder, May 17-20, San Diego (Robert Kramer)
- Zendesk Relate, May 18-20, Denver (Melody Brue)
- Broadcom Mainframe Analyst Summit, Boston, June 8-10 (Matt Kimball)
July events coming soon.
August events coming soon.
September events coming soon.
- SAP Connect, October 5-7, Las Vegas (Robert Kramer)
- Oracle AI World + SuiteWorld 2026, October 26-29, Las Vegas (Robert Kramer)
November events coming soon.
December events coming soon.
Subscribe
Want to talk to the team? Get in touch here!







