This week Microsoft hosted its Ignite event in Chicago and, as expected, it was very heavy on AI agents in terms of announcements. This makes a lot of sense, and I intend to do a deeper research piece on all of the announcements very soon. What stands out so far is the breadth of the announcements. Microsoft is a massive technology company providing value to many different stakeholders. So, it was not a surprise to see agentic aspects throughout its developer tools such as Copilot Studio, as well as embedded into standard products such as Office 365. What I wonder is how quickly we will see the agent category start to fracture into sub-categories. This will likely be a healthy thing, since my conversations with sales and marketing leaders tells me that education around agents and AI in general is seriously lacking. One click down into how this stuff works—and the implied business value of different types of agents—might do wonders for everyone.

This week I got the chance to meet the founding team at Zavvis Technologies to discuss their approach to agentic applications. The discussion provided a cool perspective on startups in the age of AI. The access to LLMs via hyperscalers and horizontal tooling for data science and application development enables new approaches to innovating. What Zavvis is doing is digging into how agents can improve the CFO function in a company. To be specific, it’s not really about automating financial operations and processes, but more a set of agents to help the CFO understand options and see opportunities in a mass of data. To me, it’s a very interesting approach for a couple of reasons. First is simply the fact that a company can now use technology to take AI deep into a very specific functional area. Second, the agents should help blend together structured and unstructured data, which may in fact enhance the CFO’s role. That said, this approach is a bit tricky, given that the CFO role in particular is known for risk aversion—which suggests a different sort of go-to-market strategy for Zavvis. I am excited to hear what the company learns as it gets moving.

In October, IBM announced the availability of its Granite 3.0 models; since then it has been engaging many different ecosystems talking about the open source value proposition for LLMs. While IBM is not the only player here, it does have an evolved take on how transparent vendors can be about LLMs. But one item that got lost in the shuffle was IBM introducing one Granite variant specifically to implement guardrails on speech and bias. IBM has a notion of this LLM sitting in front of other LLMs. For an analogy, think of when someone in a TV studio “bleeps” out a word they are not allowed to broadcast. I was talking to another client about this today and I started to think about the utility of purpose-built LLMs being used as a front-end. It’s an interesting notion, and whether it’s going to win out against other rules-based guardrails and security measures is unknown. But it also suggests that we are about to enter into some really serious architectural discussions for AI in 2025.

This week Microsoft hosted its Ignite event in Chicago and, as expected, it was very heavy on AI agents in terms of announcements. This makes a lot of sense, and I intend to do a deeper research piece on all of the announcements very soon. What stands out so far is the breadth of the announcements. Microsoft is a massive technology company providing value to many different stakeholders. So, it was not a surprise to see agentic aspects throughout its developer tools such as Copilot Studio, as well as embedded into standard products such as Office 365. What I wonder is how quickly we will see the agent category start to fracture into sub-categories. This will likely be a healthy thing, since my conversations with sales and marketing leaders tells me that education around agents and AI in general is seriously lacking. One click down into how this stuff works—and the implied business value of different types of agents—might do wonders for everyone.

This week I got the chance to meet the founding team at Zavvis Technologies to discuss their approach to agentic applications. The discussion provided a cool perspective on startups in the age of AI. The access to LLMs via hyperscalers and horizontal tooling for data science and application development enables new approaches to innovating. What Zavvis is doing is digging into how agents can improve the CFO function in a company. To be specific, it’s not really about automating financial operations and processes, but more a set of agents to help the CFO understand options and see opportunities in a mass of data. To me, it’s a very interesting approach for a couple of reasons. First is simply the fact that a company can now use technology to take AI deep into a very specific functional area. Second, the agents should help blend together structured and unstructured data, which may in fact enhance the CFO’s role. That said, this approach is a bit tricky, given that the CFO role in particular is known for risk aversion—which suggests a different sort of go-to-market strategy for Zavvis. I am excited to hear what the company learns as it gets moving.

In October, IBM announced the availability of its Granite 3.0 models; since then it has been engaging many different ecosystems talking about the open source value proposition for LLMs. While IBM is not the only player here, it does have an evolved take on how transparent vendors can be about LLMs. But one item that got lost in the shuffle was IBM introducing one Granite variant specifically to implement guardrails on speech and bias. IBM has a notion of this LLM sitting in front of other LLMs. For an analogy, think of when someone in a TV studio “bleeps” out a word they are not allowed to broadcast. I was talking to another client about this today and I started to think about the utility of purpose-built LLMs being used as a front-end. It’s an interesting notion, and whether it’s going to win out against other rules-based guardrails and security measures is unknown. But it also suggests that we are about to enter into some really serious architectural discussions for AI in 2025.

Microsoft just released Magentic-One, a multi-agent AI system that can handle open-ended tasks common to daily life. The system’s multiple agents each have specialized functions. These agents are controlled by an orchestrator agent that acts as a monitor and supervisor. The open source system is new from the standpoint that it is active instead of passive and can provide recommendations and execute tasks. According to Microsoft, Magentic-One excels at software development, data analysis, and navigating the internet.

My biggest takeaway from the Microsoft Ignite conference is how much the company has invested in its infrastructure. It has added an HSM for crypto management and a DPU for networking and storage acceleration to complement its Cobalt CPU, Maia AI accelerator, and existing security platform. Effectively, Microsoft has joined AWS and Google Cloud in developing custom silicon to deliver a full compute experience.

In addition to this, the company worked with AMD to develop a custom chip—the EPYC 9V64H—to support virtualized HPC workloads. This chip will be outfitted with HBM3 memory and double the infinity fabric. While this is an incredibly powerful compute platform, what is perhaps more interesting is to see the dominant position AMD has taken in the cloud space. Custom chip work for the CSPs was once the domain of Intel and Intel only.

Finally, Microsoft has expanded its partnership with Oracle by activating an additional 12 regions for Oracle Database@Azure and making the environment managed and governed through Azure Resource Manager and Purview, respectively. Effectively, Oracle Database@Azure is now fully integrated as a native service.

Part 1: SC24 came and went, and boy was it a ride! There are so many storylines to trace from the big supercomputing conference, so let me just share a few.

My biggest takeaway is a little bit esoteric. When I attended SC15 in Austin in 2015, it felt like I was walking through a science fiction magazine because the technology was so disconnected from what was happening in the enterprise datacenter at that moment. In particular, it was very focused on the big national and academic labs. At that event, I saw and talked about topics that would eventually become common in the enterprise—but not for years.

By contrast, walking through the show floor this week in Atlanta, I could immediately tie all of the innovations I saw to uses by the average enterprise trying to operationalize AI. Put more succinctly, over the past decade the pace of innovation has drastically accelerated from the innovators to the deployers.

Here’s another big takeaway: holy liquid cooling, Batman! I have been watching the liquid cooling market for some time. Some of the earlier players such as Vertiv, LiquidStack, CoolIT, GRC, and Motivair are now part of a much larger market peppered with logos—both the familiar and the new. I saw a total of 50 companies listed on the exhibitor list, including 22 for liquid cooling. And this doesn’t count the likes of Delta, Schneider Electric, and some of the power and infrastructure companies that have either already joined the market or are looking to enter it. The (smart) acquisition of JetCool by Flex is a good example of this.

Overall, I’m quite impressed with the amount of liquid cooling I’ve seen from cooling vendors and OEMs alike. Lenovo has been pushing Neptune for a long time, and we saw HPE start aggressively telling its liquid cooling story this summer at its Discover conference. Now Dell is really starting to jump in the game (for instance, the XE9712 racks they are shipping to CoreWeave are liquid-cooled).

With this said, I think we are still very early in this cooling game, and what we are seeing in today’s market is kind of like the days of discovering fire and inventing the wheel. As warm-water cooling is starting to find a place in the market, look for two-phase direct-to-chip (D2C) cooling to play a bigger role, as it is so much more able to address the heat density that we see on chips. Longer term, I think immersive technologies will be niche in application and will eventually bridge to cooling technologies we aren’t even covering today.

The last big takeaway is about the silicon innovation going into this market across the entirety of the data journey. I have been in the tech industry for over 30 years and I’ve never seen so much innovation in the silicon space. Most of us see NVIDIA’s biggest threat coming from AMD, Intel, and Arm—or maybe even a Qualcomm or Marvell. However, don’t overlook the many, many innovation engines in the chip industry like NeuReality, Tenstorrent, Cerebras, Untether, or others.

Part 2: The line from supercomputing to enterprise computing has become short and straight.

We’ve seen HPC-like requirements creeping into the enterprise for some time. First it was the larger enterprise organizations with workloads like crash simulation and high frequency trading. Big data, EDA, and data analytics really pushed this requirement for accelerated compute and more bespoke storage and networking to populate the enterprise datacenter. But AI has totally disrupted the game and, yes, it has brought supercomputing into the enterprise in a major way. And to the edge—and wherever else there’s data. This is why we see such big market sizing and CAGRs associated with AI. It’s not just about the chips, servers, storage, and networking; it’s about the cost of deploying, tuning, and managing these environments. And because of its nascency, there is so little knowledge to share—certainly no institutional knowledge or “muscle memory.” Because of this, I see the consulting companies playing a big role in the AI journey.

Like I said when I was talking about the SC15 conference earlier, the line from supercomputing to mainstream enterprise used to be long and crooked. That line is now very short and very straight. No longer is technology being developed and then kind of iterated on for eventual broader consumption. The ability for technology to be broadly adopted (and used) in a commercial way is now a primary concern for any startup playing in this space.

Canva recently appointed Kelly Steckelberg, former Zoom CFO, to the same position within its organization. Steckelberg brings a wealth of experience, having successfully steered Zoom through its IPO and a period of rapid growth. Canva is currently valued at approximately $32 billion, with more than $2 billion in annual recurring revenue. It has seen significant success in expanding into the enterprise market, with 95% of Fortune 500 companies as users. Although Canva states there are no immediate plans for an IPO, Steckelberg’s appointment and the company’s strong financial performance suggest a public offering could be on the horizon. Steckelberg says she sees tremendous opportunity at Canva, and I believe the company is very fortunate to have her. At a recent Zoom event, she expressed confidence in her successor, Michelle Chang. Chang joins Zoom from Microsoft, where she served as corporate vice president and CFO of the Commercial Sales and Partner Organization. Chang will be front and center next week as Zoom presents its earnings. Analysts project Zoom to deliver year-over-year earnings growth driven by higher revenues when it releases its Q3 2025 financial results for the quarter ending October 2024. Chang will be a critical part of Zoom’s next growth phase as the company moves from a video conferencing company to an AI-driven collaboration and productivity platform.

Microsoft Fabric announced significant updates to its unified data platform at Ignite 2024. Fabric Databases, initially including SQL Database, now include transactional capabilities with built-in security, vector search, RAG support, and Azure AI integration, enabling the development of AI-optimized applications. OneLake, the platform’s multi-cloud data lake, now has enhanced multi-cloud and on-premises data integration with Azure SQL DB Mirroring. Several workload-specific updates were also announced, including sustainability data solutions, AI functions for text analysis in notebooks, and a GraphQL API for simplified data access. AI capabilities expanded with conversational AI tools, Azure Event Hubs KQL database support, and integration with Azure AI Foundry.

These updates strengthen Microsoft’s position against competitors, enhancing Fabric’s appeal as a unified platform for data management and AI development. By addressing enterprise requirements, Fabric reinforces Microsoft’s ability to compete with other major players in data management and AI such as Google Cloud and AWS.

IBM has modernized Db2 with a new AI-powered database assistant. As data demands grow, database systems must evolve to keep pace, and last week IBM released Db2 12.1, incorporating a slew of AI features. The new release addresses key challenges faced by database administrators and introduces the Database Assistant, developed using IBM watsonx. This assistant delivers instant answers, real-time monitoring, and intelligent troubleshooting. Explore these innovations further in my latest Forbes article, which features Miran Badzak, IBM’s program director for databases.

LogicMonitor is advancing in the hybrid observability industry with an $800 million investment to integrate AI into datacenter operations. Led by CEO Christina Kosmowski, the company focuses on helping businesses reduce costs and scale AI while improving efficiency and meeting sustainability goals. This funding looks to strengthen LogicMonitor’s role in supporting modern datacenter management and AI-driven operations.

Cloudera announces the acquisition of Octapai’s data lineage and catalog platform, expanding its data catalog and metadata management capabilities. With this move, Cloudera will be able to provide customers with visibility across data solutions, allowing them to use trusted data for AI, predictive analytics, and decision-making tools. Key benefits include improved data discoverability, quality, governance, and migration support.

Change is in the air as we approach the new year: Amazon Web Services has brought in Julia White as CMO. White was recently the Chief Marketing and Solutions Officer at SAP. Before that, she was at Microsoft for two decades, including as corporate vice president for Azure product marketing. Her expertise spans cloud services, AI, and product messaging, making her well-suited to AWS’s strategic needs. With this move, AWS looks to strengthen its position in the competitive cloud market against Microsoft Azure and Google Cloud. With AWS quarterly profits surpassing $10 billion for the first time, White’s leadership is expected to enhance AWS’s focus on cloud computing and AI innovation. This leadership change follows recent executive departures at AWS, including former CEO Adam Selipsky and VP of AI Matt Wood. Wood has since joined PwC as the firm’s first Commercial Technology and Innovation Officer.

Microsoft launched Flight Simulator 2024 with significantly enhanced features and new flying jobs that enable you to pilot virtually any kind of flying craft, from hot air balloons to helicopters and dirigibles. Unfortunately, Microsoft didn’t adequately anticipate the demand for the game, and servers crashed; the company apologized to gamers for not being prepared for the launch to be such a hit. I believe part of the success came from the game being included with Game Pass. Flight Simulator is the second big title this month to launch on Game Pass after Call of Duty launched earlier in November.

Qualcomm Investor Day — Nakul Duggal (Qualcomm’s group GM for automotive, industrial, and cloud) presented Qualcomm’s IIoT deployment model for AI at the edge. The company foresees a $50 billion market opportunity for edge intelligence by 2029 and has defined a path to achieve that goal. The path includes new edge computing chips (the Qualcomm IQ series) designed to support a comprehensive edge deployment architecture. The architecture aligns with industry-wide trends that make IIoT much more scalable.

Here’s how it works. Qualcomm customers develop AI-powered applications in the cloud for deployment on both cloud and local platforms. AI-accelerated on-premises “AI edge boxes” run cloud-native computing software environments on appropriately scaled compute platforms. The development model is “build in the cloud, deploy on the edge” using the same software infrastructure. However, IT-managed, cloud-native platforms do not extend all the way down to the chaotic world of OT (operational technology) devices. These small embedded platforms are often highly customized and optimized for specific tasks. OT devices require unique software stacks, device management services, and connectivity schemes. The result is a line of demarcation that separates small OT platforms from large IT systems running distributed cloud environments. Mr. Duggal explained the Qualcomm IQ Series processors and subsystems in terms of this model, enabling a new generation of on-premises compute platforms with the power and scale to address a wide range of vertical industries.

MY TAKE: AI at the edge is the new IIoT north star. I’ve advocated this three-tier architecture (cloud, distributed cloud, device) for years, and it’s great to see Qualcomm and other big suppliers follow the same pattern.

Microsoft Ignite — Not surprisingly, Ignite focused on AI in the cloud and at the edge. From an IIoT standpoint, Azure’s Adaptive Cloud approach was the star of the show, enabling AI to work across sites, clouds, distributed computing, and devices. The Azure IIoT model is consistent with my “fractal” view of IIoT intelligence, with cloud-native environments scaling from global clouds to local on-premises servers. Microsoft’s AI enablement products, data fabric, event grid, event hubs, storage, Power BI interfaces, and other services run on the whole range of platforms, while Azure IoT Operations (enabled by Azure Arc) implements device data interfaces and manages the OT data at the edge. The IIoT devices are, essentially, peripherals communicating via standards-based protocols.

MY TAKE: Azure IoT Operations is emerging as the interface between the chaotic world of OT devices and the structured world of Microsoft’s AI-enhanced IT systems. Other hyperscalers and platform suppliers are moving in this same direction, allowing enterprise applications (i.e., ERP) to immediately scale up OT-enabled AI-powered solutions with minimal dependencies on IIoT device systems. Even though Ignite didn’t have many IoT-specific sessions, the industrywide trend to separate chaotic device development from high-growth, AI-driven business transformation came across loud and clear.

Shure, a company known for high-quality audio equipment, has partnered with Microsoft by integrating its products with the Microsoft Device Ecosystem Platform (MDEP). This collaboration allows Shure to develop new audio solutions for Android devices for Microsoft Teams Rooms. What this means for Shure is enhanced security measures that meet Microsoft’s high standards, improved compatibility with Microsoft Teams Rooms, and the opportunity to tap into new markets. For example, government agencies that rely on Microsoft Teams for secure communication could now outfit their conference rooms with Shure microphones and audio processors that integrate with their existing systems. This partnership paves the way for Shure to deliver enhanced audio experiences to a broader range of users who depend on Microsoft Teams for collaboration.

Box held its annual BoxWorks event in person for the first time in several years and announced a suite of new AI-powered tools focused on helping businesses unlock the value of their content. These include AI Studio for building custom AI agents, Box Apps for automating workflows, and enhanced security features. Box also supports nonprofits using AI for social good through its Impact Fund. I provide more detail about BoxWorks and the company’s strategy for addressing organizations’ growing need to manage and extract value from increasing content volumes in this Forbes contribution.

Amazon announced the new Echo Show 21, its biggest and most capable smart home display with a built-in smart home hub. This new smart home hub includes support for both Matter and Thread wireless standards, making it an anchor for your smart home and putting Amazon at the center of that experience. While I’m not sure I have the need or space for this, there is also a 15-inch version which may be easier to fit into smaller kitchens. Amazon has designed the Echo Show 21 as a wall-mounted display; I may have to test it myself to understand the utility of having such a large display in the kitchen.

At Microsoft Ignite, the company announced a few updates to Copilot and Window 365, including a thin client called the Link. The Link appears to be Microsoft’s way of satisfying the market need for low-cost products while driving Windows 365 virtual PC usage. While many of Microsoft’s partners like HP, Dell, and Lenovo already offer thin clients, this appears to be Microsoft’s own approach focused around Windows 365 only. Microsoft also announced the new Copilot Actions, which gives prompt templates to help automate repetitive tasks. While this isn’t quite a way to enable scripting, I do think this should help to improve Copilot usage.

Microsoft and Atom Computing successfully entangled 24 logical qubits using neutral atoms under control of Atom Computing lasers. This is the record for the highest number of entangled logical qubits. Logical qubits are constructed from multiple physical qubits, and they allow complex quantum algorithms to be run. The system demonstrated high gate fidelities: 99.963% for single-qubit gates and 99.56% for two-qubit gates, making this the highest neutral-atom two-qubit gate fidelity in a commercial system.

A characteristic of neutral atom quantum computers is the tendency of atoms to disappear during operations. The team developed a method to detect and replace lost atoms without disrupting computations. As a benchmark, the researchers ran the Bernstein-Vazirani algorithm, which identifies a hidden binary string. The 20 logical qubits (created from 80 physical qubits) found the secret code in a single attempt, outperforming classical counterparts that must run the search many times to find all the bits.

This is good news for quantum computing. Great progress is being made with logical qubits, in fact doubling the number that the Microsoft-Atom Computing team has accomplished in just a few short weeks.

Qualcomm held its investor day—the first in three years—and updated analysts and investors on the progress it has made over the last three years in its product diversity strategy. The company also stated that it expects its smartphone business to be just 50% of revenue by 2030. This is a significant goal because the company currently generates 75% of its revenue from its handset business. In addition to that lofty goal, the company also revealed that it is planning an even more affordable PC processor soon, which will enable $600 CoPilot+ PCs. Additionally, it said that the next generation of its X Elite processors will be powered by the third generation of its Oryon CPU cores, which have done exceptionally well in benchmarks against Apple.

The whole industry eagerly awaited NVIDIA’s earnings, and the company beat on revenue and profit and guided slightly above expectations . . . yet still got punished in after-hours trading. The reality is that people’s expectations of NVIDIA are simply unrealistic and heavily tainted by retail hype, even though the company is now generating $35 billion in revenue per quarter and almost $20 billion in profit per quarter. NVIDIA is basically printing money compared to almost everyone else in the industry and dwarfs many of its nearest competitors. I believe that we are still very much in the early phases of AI, and while some of the AI model builders may be hitting a bit of a wall with training, NVIDIA says that demand for training chips remains high — and it is convinced that it is also well-positioned with chips for inference.

IBM has partnered with the Ultimate Fighting Championship to become UFC’s Official Global AI Partner. The UFC Insights Engine, built with IBM watsonx, utilizes data and AI to analyze live bouts, fighter tendencies, projected match outcomes, and methods of victory. This provides fans with detailed information to deepen their interest in the sport. It is another great example of sports technology enriching a sport and driving fan engagement.

Solidigm has launched a new 122TB SSD designed to reduce energy consumption in data centers, which are facing increasing demand and costs due to the rise of AI. This new drive offers significantly higher storage density and efficiency, which should lead to lower energy bills and a smaller physical footprint. This is crucial for sustainability and allows companies to invest more in AI development. Read more about Solidigm’s 122TB drive in my recent Forbes article.