Last week I mentioned that we should be seeing more M&A activity in 2026, and it seems IBM got a head start on that by announcing its intention to buy Confluent, which curates the highly successful Apache Kafka open source project. It’s a move that makes a lot of sense, given that IBM has a number of other data offerings — and continues to be the standard-bearer of open source products. Confluent specializes in data streaming, which makes developers’ lives easier (especially in hybrid environments) and provides a more event-driven data architecture, which is something IBM has been lacking. IBM is touting this as the foundation for a next-generation data platform, but frankly I am most interested to see how something like Confluent will intersect with and improve the existing IBM automation Software portfolio.

I also wanted to make note of two announcements that may typically fly under the radar but could have a big impact. For the past two years, many of us have lamented how the AI landscape is dominated by a few big players. But this week both Salesforce and Google made moves to open up their AI technology or data to other companies. These are the type of ecosystem-building moves that can lead to the next wave of startups and innovations. Salesforce has opened up its Agentforce 360 platform for partners to build new types of agents that leverage Salesforce’s applications and data services. Google has opened up a number of its services including Google Maps and Google Compute Engine as MCP tools, which will allow remote agents to either leverage or manipulate Google services remotely. While on paper these actions may seem a bit commonplace for many early-stage startups, they could also be a big accelerant for the big companies’ own solutions strategies.

Last week I mentioned that we should be seeing more M&A activity in 2026, and it seems IBM got a head start on that by announcing its intention to buy Confluent, which curates the highly successful Apache Kafka open source project. It’s a move that makes a lot of sense, given that IBM has a number of other data offerings — and continues to be the standard-bearer of open source products. Confluent specializes in data streaming, which makes developers’ lives easier (especially in hybrid environments) and provides a more event-driven data architecture, which is something IBM has been lacking. IBM is touting this as the foundation for a next-generation data platform, but frankly I am most interested to see how something like Confluent will intersect with and improve the existing IBM automation Software portfolio.

I also wanted to make note of two announcements that may typically fly under the radar but could have a big impact. For the past two years, many of us have lamented how the AI landscape is dominated by a few big players. But this week both Salesforce and Google made moves to open up their AI technology or data to other companies. These are the type of ecosystem-building moves that can lead to the next wave of startups and innovations. Salesforce has opened up its Agentforce 360 platform for partners to build new types of agents that leverage Salesforce’s applications and data services. Google has opened up a number of its services including Google Maps and Google Compute Engine as MCP tools, which will allow remote agents to either leverage or manipulate Google services remotely. While on paper these actions may seem a bit commonplace for many early-stage startups, they could also be a big accelerant for the big companies’ own solutions strategies.

Thoma Bravo finalized its acquisition of Verint Systems on November 26, 2025, immediately merging it with portfolio company Calabrio. Thoma Bravo operating partner Mike Lipps has assumed the role of chairman and interim CEO, replacing Dan Bodner. Private equity firms continue to consolidate the CX software sector, a trend reinforced by this deal and comparable take-private moves, including Hellman & Friedman’s purchase of Zendesk and Silver Lake’s acquisition of Qualtrics.

From a technical perspective, the combination targets the convergence of AI automation and human labor; Verint’s bot-heavy interaction layer handles self-service, while Calabrio’s suite optimizes the human agents handling escalations. This theoretically offers CIOs a unified platform for managing hybrid workforces. However, significant redundancy exists between the two companies’ workforce optimization (WFO) tools, creating roadmap ambiguity for current customers. The technical debt involved in rationalizing these distinct architectures may slow near-term innovation, risking a period of stagnation while the firm prioritizes backend integration over new feature development.

Adobe has released an update for the Adobe Premiere mobile application that integrates a dedicated workspace for creating and publishing YouTube Shorts. This integration provides a specific environment within the mobile app designed to align with the technical requirements of YouTube’s vertical video format. The update introduces capabilities such as exclusive effects, transitions, and title presets alongside a robust template system. Users can access pre-built templates that contain embedded text and effects, or generate original templates that other creators on the platform can reuse. The application also incorporates advanced audio processing tools, such as Enhance Speech for voiceover clarity and Generative Sound Effects, which use AI to generate sound elements from voice inputs. These features operate within a standard non-linear editing interface that supports unlimited multi-track timelines and manual controls for color and layering.

This development highlights the strategic necessity of short-form video as a primary channel for audience acquisition and corporate marketing. With YouTube Shorts averaging over 200 billion daily views, the format has evolved from casual social sharing into a critical medium for high-volume organic reach. Previously, achieving high production value for vertical video often required a fragmented workflow involving desktop software and file transfers. By embedding professional-grade features directly into the mobile pipeline, Adobe addresses the market shift where professional users require the speed of mobile-native editors without sacrificing technical control. This move could consolidate the workflow for enterprise and professional users who might otherwise migrate to lightweight competitors for their daily content needs.

I like IBM’s plan to acquire Confluent, despite the hefty price tag of $11 billion, to close the gap in its data platform. Confluent could give IBM a reliable way to handle real-time data, which has become necessary to move AI from projects to real operations. I’ve said many times that without consistent and trustworthy data, the rest of the AI stack doesn’t get very far. This deal could give IBM a chance to have a unified platform with its data, integration, and AI capabilities instead of working with separate tools that do not always fit together.

This acquisition also fits into IBM’s broader pattern of working to modernize its portfolio. Earlier deals like Txture, which strengthens IBM’s cloud modernization and infrastructure migration services, and its work with Cognitus, which expands IBM’s SAP transformation and industry-specific capabilities, show that IBM knows it needs to close the gaps between legacy estates, cloud platforms, and new AI-driven workloads. Those moves give IBM more credibility in migration, application modernization, and ERP transformation conversations, but they do not solve the real-time data challenge. Confluent could be seen as the missing piece that brings data processing into the center of IBM’s architecture discussion.

There is still a question of whether IBM can turn all of this into a unified platform. Integrating Confluent along with the capabilities gained from Txture and Cognitus into IBM’s broader portfolio will take time, and there is competitive pressure from Microsoft, Google Cloud, AWS, Oracle Cloud, Snowflake, Databricks, Cloudera, and Teradata as they continue to evolve their AI-focused data architectures.

Everything comes down to execution. The real test is whether IBM can make the data environment easier to work with and reduce the hurdles for data, governance, and AI workflows. Customers need that simplicity to move their AI plans from ideas to real operations. If IBM is able to deliver on that, organizations could get a platform that is easier to manage and more aligned with how they want to use AI. If not, this could become an acquisition that does not change IBM’s direction or its competitive position.

At NTT’s R&D Forum last month, the company demonstrated a photonics-electronics convergence device based on NTT’s photonics technology stack that facilitates communication between datacenters with a purpose-built optical switch that the company refers to as a PEC device. The multi-year PEC roadmap calls for building on the commercialized interconnect card that exists today to deliver inter-board, inter-package, and intra-package designs by 2032.

Co-packaged optics have become a hot topic of discussion this year, given their value for integrating optical components with network silicon and AI accelerators, which can improve signal loss over longer distances and dramatically reduce power consumption. NTT believes that it can achieve the same with PEC, and that could be a game-changer given the power-hungry needs of modern AI applications and workloads.

Oracle released its latest financial results to mixed reactions from the market, but I thought the results showed progress. When I look at how a business operates rather than how the stock traded, I see that Oracle has continued its movement toward cloud and AI expansion. In fiscal Q2 2026, Oracle’s total revenue was up 14% year over year to $16.1 billion, with non-GAAP EPS up 54% to $2.26 and GAAP EPS up 91% to $2.10, helped by a $2.7 billion pre-tax gain from the sale of its Ampere stake.

What matters is that Oracle is still growing while funding a much larger infrastructure build. Profitability has not disappeared, but it will be harder to sustain as spending increases, which puts more weight on execution going forward. Cloud remains the main driver. Total cloud revenue (IaaS plus SaaS) grew 34% YoY to $8.0 billion, and cloud infrastructure (IaaS) revenue was up 68% to $4.1 billion. Cloud applications (SaaS) revenue also grew, up 11% to $3.9 billion, led by Fusion Cloud ERP, up 18% to $1.1 billion, and NetSuite Cloud ERP, up 13% to $1.0 billion. That level of growth shows the demand for AI workloads and large-scale database environments, not just customers lifting existing applications to the cloud.

Backlog was the key data point this quarter. Remaining performance obligations reached $523 billion, up 438% YoY, driven by large multi-year cloud infrastructure agreements with customers such as Meta and NVIDIA. That backlog provides revenue visibility, but it also creates operational pressure. Oracle now has to add datacenter and AI capacity on schedule, bring it into production, and convert contracted commitments into steady usage without cost overruns.

The post-earnings stock reaction reflects short-term concerns. A modest revenue miss and a 3% decline in traditional software revenue show that legacy businesses remain under pressure, even as cloud growth begins to offset some of that weakness. Capital spending will increase as Oracle adds datacenter and AI capacity, which will constrain free cash flow in the near term. Given the size of the backlog, that tradeoff is expected.

Competition is always a big part of the picture. Oracle is operating alongside AWS, Microsoft, and Google, all of which have broader cloud ecosystems. Oracle does not need to displace those providers, but it does need to stay relevant in database-centric and AI-driven workloads where it has strengths. Its multi-cloud approach, including plans for 72 Oracle Multicloud Datacenters embedded within Amazon, Google, and Microsoft clouds, is a practical way to extend reach without forcing customers into a single-cloud model.

The quarter indicated strong demand for what Oracle is offering, but it also put the spotlight on how well the company executes and the margins it achieves. Oracle now has a clearer picture of what’s coming up, but the next step is to show that they can turn that demand into actual capacity, consistent usage, and better margins as time goes on.

When I talk about the next generation of ERP, I usually start with what is no longer working. Traditional ERP was built to be a system of record. It captured transactions, enforced controls, and produced reports, but most of that value usually showed up after decisions were made. Today that lag is a problem because businesses need ERP to help them respond while events are unfolding, not weeks later in a report. That is why I frame modern ERP as a system of action, one that connects data, context, and execution so work can move faster and with fewer handoffs.

This is where Industry 5.0 becomes relevant in very practical terms. Industry 5.0 is not about turning everything over to automation. It is about using AI to support people, reduce friction, and improve decision quality. In that model, ERP becomes the operational nerve center that brings together people, money, and process data, then uses intelligence to surface issues, suggest next steps, and guide execution. Instead of employees spending time reconciling systems or chasing approvals, the system helps prioritize what matters, while humans stay accountable for judgment and outcomes.

Workday has recently discussed its direction and how its platform is built around a shared data foundation for finance and HR, with AI embedded directly into everyday workflows instead of added on later. Tools like Ask Workday give users a simple way to get answers, while role-specific AI for finance and people teams supports routine decisions as work is happening. APIs and agent frameworks make it easier to extend those capabilities without breaking the core system. The emphasis on trust and governance matters because these systems now influence pay, hiring, compliance, and financial results. What separates this approach from earlier cloud ERP efforts is that it actually changes how work gets done, rather than just moving older systems to a different hosting environment.

I see this as progress not just from Workday, but across the ERP market. Vendors are building systems that operate with more real-time data, embedded analytics, and clearer coordination between people and automated workflows. This is already showing up in planning, workforce management, and financial operations, where ERP supports decisions as work happens rather than after the fact. In short, ERP is moving away from a passive back-office role and taking on a more active role in day-to-day operations.

The Bank of New York Mellon has integrated Google Cloud’s Gemini Enterprise into its proprietary AI platform, Eliza. This collaboration updates the platform’s infrastructure to harness Google’s agentic AI capabilities, specifically incorporating the Gemini 3 and Veo 3 models. The integration targets the deployment of multimodal AI agents across BNY’s workforce, intended to facilitate the development of internal tools for processing and synthesizing financial reports, datasets, and historical market trends.

The technical focus of this deployment is on automating data-intensive tasks and enhancing research workflows through what the companies describe as “agentic deep research.” By leveraging these newer model architectures, BNY aims to enable more complex reasoning and analysis within its existing enterprise environment, moving beyond standard text generation to actionable, agent-based financial operations.

Last week at Open Source Summit Japan 2025, the Linux Foundation formally announced the elevation of Space Grade Linux (SGL) from a Special Interest Group under the ELISA banner to a full Linux Foundation project. The new stack is called Papermoon, and its concept is similar to that of other domain-focused projects, such as Automotive Grade Linux (AGL) and the Civil Infrastructure Platform (CIP). Instead of building custom, one-off software stacks for each space mission, SGL provides a common software platform with space-ready board support, a Yocto build system, high reliability, and long-term maintenance. Papermoon will continue to build upon the safety-critical work done by ELISA over the past six years. You can read more about Papermoon in this New Stack article.

My take: There’s definitely a trend towards vertical OS clustering for embedded systems, and this is a good time to review the reasons for creating these “silos.” In theory, OSes like Linux and Windows already provide domain-independent abstraction layers, so why fragment the ecosystem into multiple distributions? Here’s why. Kernel and base OSes need tailoring for industries with specialized regulatory, lifecycle, hardware, and deployment realities. Domain-specific tailoring is costly and creates long-term technical debt without offering any competitive product differentiation, so it makes sense to implement these modifications once and for all. Space (SGL) is a great example of how this approach reduces development cost and time-to-market. We’re seeing similar progress in verticals such as automotive (AGL), industrial infrastructure (CIP), networking (OpenWrt), VyOS (telecom), edge (Alpine), virtualization (Proxim), and other domains.

Google has announced the rollout of a new feature called Android Emergency Live Video in the U.S., along with Mexico and Germany. The capability becomes available during an emergency call (e.g. via 911 in the U.S.) and enables the caller to share live video with the dispatcher. Google says the feature is encrypted and that the user can end the livestream at any time. There is also no need to enable this feature as it will be native to any Android using Android 8 or newer. I believe that this feature will undoubtedly save lives by giving emergency responders better and more real-time information.

SpaceX, AST SpaceMobile, and Amazon (via Amazon Leo, formerly Project Kuiper) are all trying to compete for the same LEO customers. I believe that SpaceX’s planned IPO shows that there’s still lots of appetite for satellite technology and high-bandwidth connections. However, I don’t believe that the new push to talk about datacenters in space is remotely a productive one; we should be focusing on the LEO constellations and their performance/capability. SpaceX is still the only network that’s commercially available to both home and mobile users and has significant customers, including airlines like Hawaiian and United. I believe that Amazon’s strategy of focusing on enterprise users and offering higher-speed solutions may work out well for a higher-ARPU business with better margins. AST SpaceMobile seems to be many of the telecom carriers’ hedge against SpaceX’s dominance that we are seeing today. But that said, AST SpaceMobile offers a very different solution, and we will need to see how it competes on a commercial level before really drawing any conclusions.

T-Mobile has begun to push its 5G RedCap solution, which I believe has the potential to be bigger than most people realize. 5G RedCap (which stands for “reduced capacity”) leverages T-Mobile’s Standalone 5G network to offer lower throughput than traditional 5G, but with lower power consumption and the same latency as 5G. This is important because it creates a lower-cost 5G solution that still delivers internet connectivity without sacrificing the user experience. The first place we are seeing this launched broadly is the Apple Watch Series 11 models, which all feature a 5G RedCap modem and connect to T-Mobile’s network to get a better, lower-latency 5G connection. While I’m still not a believer in cellular being particularly productive for smart watches, wearables in general are great candidates for RedCap modems, and I believe we’ll likely see them in AI smart glasses fairly soon — and maybe even AR glasses somewhere down the road.

Microsoft has integrated OpenAI’s GPT-5.2 into Microsoft 365 Copilot and Copilot Studio, bifurcating the architecture into “Thinking” and “Instant” configurations. The Thinking variant is engineered for complex problem-solving and strategic analysis, while the Instant version prioritizes low-latency tasks such as drafting. Within the enterprise environment, these models use “Work IQ” to query and reason across disparate data sources — including meeting transcripts and documents — marking a functional pivot from generative text to cross-entity reasoning. Deployment begins immediately for commercial license holders via a manual model selector, with Premium subscribers scheduled for early next year.

This implementation shifts operational complexity to the user, who must now actively optimize between inference speed and analytical depth based on the specific task. While the system moves closer to an agentic framework capable of deriving new insights, it relies heavily on the structural integrity of the underlying data graph. For developers, the automatic migration of Copilot Studio agents from GPT-5.1 to GPT-5.2 eliminates manual upgrades but introduces a critical need for regression testing, as the new model’s reasoning behaviors may impact how existing agents interpret prompts and execute logic flows.

Two new advances in quantum networking and quantum sensing could have significant impacts on quantum technology. First, deterministic all-photonic entanglement swapping has been demonstrated across remote quantum dots. That success should mean that a viable path to scalable quantum repeaters for a secure internet is very close. Second, there’s good news on the hardware front, in the form of an 8-qubit silicon spin array made in a 300mm CMOS process. It shows long coherence and high-fidelity operations. Both of these quantum successes directly map into my current focus areas in quantum networking, sensing, hardware, and scalable architectures.

All-Photonic Entanglement Swapping Using Remote Quantum Dots
A just-published paper titled “All-photonic entanglement swapping with remote quantum dots” describes a new method called entanglement swapping that allows two particles that have never interacted before to become entangled. This can be done when using photons, which also means it can be used for building quantum networks to support quantum versions of the internet. Up until now, we could only do entanglement swapping between remote systems by using light sources that only worked randomly. That was problematic because large, practical quantum networks require dependable light sources that work every time (i.e., deterministically). They must also be able to create photon pairs without degrading the quality of entanglement.

In this research, entanglement was demonstrated by swapping light pairs generated by two separate light sources (called quantum dots) made from gallium arsenide. Entangled photons were made by placing the quantum dots into special hybrid devices made from semiconductors and a material that responds to pressure. The entangled photons were made from two different quantum dots that were perfectly identical. This arrangement showed entanglement swapping with a high level of accuracy — so good that it was statistically far beyond anything possible with standard, non-quantum physics.

This work proves that quantum dots can be used as reliable light sources in large-scale quantum repeater networks. That moves us closer to building a quantum internet.

Building a Quantum Computer from Silicon
A major step was recently taken to build practical quantum computers using silicon spin qubits. What makes these qubits so promising is that they are made using the same manufacturing methods (CMOS) that produces chips for millions of laptops. Researchers at the University of New South Wales in Australia achieved a critical milestone by building and controlling an array of eight silicon spin qubits working coherently. This is a very big deal because previous efforts were typically limited to only a few qubits — without the ability to scale up to more qubits.

The eight-qubit chip was created using a standard 300mm CMOS factory process, demonstrating that the technology is compatible with large-scale, cost-effective manufacturing. These qubits showed excellent performance by maintaining coherence up to 1.31 milliseconds, which was long enough to perform complex calculations. A two-qubit gate operation was also successfully demonstrated.

For the first time, this research demonstrates that building large, reliable, and powerful quantum processors using standard silicon chip manufacturing technology is a realistic goal.

News broke this past Sunday about ServiceNow potentially acquiring cybersecurity startup Armis in a deal that could be worth $7 billion. If it moves forward, it would be ServiceNow’s largest acquisition to date, one that would continue to round out the company’s focus on moving beyond ITSM help-desk support to becoming a full-stack enterprise automation solutions provider.

Google is following Apple with the ability to use gestures to control its smart watch. With the Pixel Watch 4, Google is finally catching up to the Double Tap feature introduced on the Apple Watch in 2023. This is yet another accessibility feature that eventually becomes a mainstream technology, one that I believe will become extremely popular for things like AR where you don’t want to have to use your voice to control everything. We’re already seeing something similar with Meta’s neural band, but realistically I think most of us would prefer to just wear a watch and have that capability built into the watch rather than needing an additional new wearable. Something like the Mudra Link neural wristband makes sense because it attaches to your watch and has similar capabilities.

At AWS re:Invent 2025, what stood out in this area were the updates about how the PGA TOUR and AWS have moved to the use of agentic AI. The focus is on solving long-standing coverage and workflow gaps.

The multi-agent content engine built on Amazon Bedrock AgentCore can generate full articles for every player in the field, giving fans meaningful insights on players who usually get little broadcast attention. The PGA also expanded its AI-supported commentary, an evolution of what launched in TOURCAST at The Players Championship; this means that broadcasters can get near-real-time context and shot analysis without adding overhead.

The “Fan Experience” zone made this more concrete. Real-time birdie probability tools, immersive visuals, and a data-aware simulator showed how data can make the game easier to follow and more engaging without forcing fans to be experts.

What I also noticed is the impact of how AI agents are reducing manual content work and improving consistency, which matters as the PGA TOUR ramps up the volume and pace of digital coverage. The organization is using AWS’s AgentCore, Trainium, and Bedrock to create experiences that are more tailored to each person and based on data, all while managing the complexity of production.

Verizon recently announced pricing cuts across a large swath of its subscriber service plans. On the surface, it is a desperate attempt to claw back lost customers, and it could backfire from a profitability standpoint. The changes are underpinned by a three-year lock-in period — one that may force AT&T and T-Mobile to match Verizon as a defensive move.

Verizon’s ARPU (average revenue per user) is already under immense pressure, and a better strategy would have been to focus on more value-added services.