Last week I mentioned a potential stealth project by OpenAI called “Strawberry.” It was pure speculation based on a Reuters rumor that OpenAI had created a very powerful model and a possible precursor to artificial general intelligence (AGI).

First, let me explain that LMSYS Chatbot Arena is a platform developed by Last Mile Systems that allows users to interact with, evaluate, and compare various chatbot models. It has a leaderboard of models being evaluated. It’s also a research tool that provides a way for developers to get data on model performance and preferences. OpenAI has previously tested models before releasing them.

Adding fuel to the rumor flame, it was reported that an anonymous model scored higher on the leaderboard in reasoning than GPT-4o. Strawberry? When I checked the leaderboard just before filing this piece, it was gone. Everyone keeps hoping that one of these new models will have a real breakout in reasoning; it will send up a flare that a powerful new form of AI has arrived.

Since we are on the subject of LMSYS, I may as well cover a new model, Mistral Large 2, which has 123 billion parameters and has done quite well on the leaderboard. Compared to the previous Mistral model, it is strong in code with over 80 languages, math, and reasoning. It doesn’t have a huge context window—128,000 tokens. That said, its instruction-following beats the Llama 3.1 405B model, and it is currently leading the Arena hard leaderboards. I believe there is much more to come from the Mistral AI team.

Last week I mentioned a potential stealth project by OpenAI called “Strawberry.” It was pure speculation based on a Reuters rumor that OpenAI had created a very powerful model and a possible precursor to artificial general intelligence (AGI).

First, let me explain that LMSYS Chatbot Arena is a platform developed by Last Mile Systems that allows users to interact with, evaluate, and compare various chatbot models. It has a leaderboard of models being evaluated. It’s also a research tool that provides a way for developers to get data on model performance and preferences. OpenAI has previously tested models before releasing them.

Adding fuel to the rumor flame, it was reported that an anonymous model scored higher on the leaderboard in reasoning than GPT-4o. Strawberry? When I checked the leaderboard just before filing this piece, it was gone. Everyone keeps hoping that one of these new models will have a real breakout in reasoning; it will send up a flare that a powerful new form of AI has arrived.

Since we are on the subject of LMSYS, I may as well cover a new model, Mistral Large 2, which has 123 billion parameters and has done quite well on the leaderboard. Compared to the previous Mistral model, it is strong in code with over 80 languages, math, and reasoning. It doesn’t have a huge context window—128,000 tokens. That said, its instruction-following beats the Llama 3.1 405B model, and it is currently leading the Arena hard leaderboards. I believe there is much more to come from the Mistral AI team.

A simple oversight and a possible Python catastrophe averted — Security issues often start in the development process, and this can manifest in many ways. Sometimes, a developer will create a security loophole to save time with the intention of removing that loophole before they build and deploy the software to production. But intentions don’t always equal reality, and that’s when bad things can happen. Companies like JFrog have tools that customers use to mitigate security risks throughout the development lifecycle. JFrog also takes a novel approach to testing its own products and services by running them against public code repositories. Fortunately, that testing process saved a lot of developers from an unintentional but potentially catastrophic security breach of multiple popular frameworks including Python. I had a chance to meet with the JFrog team this week to discuss its blog post that told the story. Stay tuned for a deeper dive from me on this topic soon.

Cloud revenues are way up. Is it AI? — The big three cloud providers all held their quarterly earnings calls over the past week, and all three reported very strong cloud performance. With all of the AI hype, it’s an easy connection to make that AI is a big driver of this success. But reality may be different. Microsoft, which grew its cloud revenue at 29% year over year, was the only company that spoke to AI’s impact on that growth. CEO Satya Nadella attributed less than a third of the growth to AI. And while AWS did not give any insight on how much of its cloud growth (19% year over year) could be attributed to AI, it did point out that a large share came from migrations of older on-premise apps. I think that the steady march of containerization and microservice-driven app development, combined with a dwindling number of IT resources, has hit a crucial tipping point. And, with or without AI, cloud businesses will continue to post solid growth.

How do enterprise IT organizations, global system integrators, IT consultants, and others map a comprehensive cloud deployment/migration strategy that accounts for data sovereignty, resilience, availability, and lowest-cost multi-cloud environments? This is a challenge I hear repeatedly as I talk with IT executives and others. Some company could create a nice market by developing such a tool that could be used on an ad hoc basis by enterprise IT organizations. I have heard of many companies claiming to have such a tool, but have yet to see a comprehensive solution commercially available.

Given the dominance of NVIDIA in the AI acceleration space, how do OEMs differentiate in addressing the enterprise datacenter market? More precisely, how does Dell truly differentiate from HPE and Lenovo? Likewise, what becomes HPE’s calling card that it can apply to the server market when all of the attention seems to be focused on silicon? As much as OEMs already differentiate in this market, the likes of Supermicro continue to grow in share (and recognition), and it is important for server vendors to continue to drive differentiation through platform innovation. A good example of this is the security HPE built into its Gen10 platform—a set of capabilities that caused the company to gain a strong market position. As Intel and AMD roll out new generations of silicon that will lead to platform refreshes, it’s important for this platform differentiation to continue.

Latest numbers indicate that AMD continues to close the gap with Intel in the datacenter space as EPYC processors reach just a touch above 24% market share. This approaches, if not matches, AMD’s all-time high for server market share, and is certainly the highest it has reached in almost two decades. While AMD has experienced a strong run in the cloud, its somewhat recent growth (and acceleration of growth) in the enterprise is perhaps more impressive, as it demonstrates a shift in the transactional server market landscape. While Intel has a very solid product in Sierra Forest/Granite Rapids, and an equally solid roadmap, it will be challenging for it to reestablish footing in this commercial server business.

There’s been a good amount of ridicule directed at NVIDIA (and to a lesser extent, TSMC) around the recent delays encountered in delivering the Blackwell GPU to the market. But at one time or another every (and I mean every) silicon provider has run into challenges that have caused delays in shipping for good reason. It’s because chip design and manufacturing is really hard. Is it rocket science? No. But do you know what the folks at NASA say? “It isn’t silicon design!”

Good on NVIDIA for identifying the Blackwell challenges and pressing pause to ensure that only high-quality silicon ships to the marketplace. And let’s hope it doesn’t happen again.

Dynatrace announced its financial results for the first quarter of fiscal 2025, with a 19% increase in ARR to $1.541 billion and a 20% rise in total quarterly revenue to $399 million. Key developments included the addition of new platform extensions Site Reliability Guardian App, Davis Anomaly Detection App, and Vulnerabilities App, expansion of security features with Kubernetes Security Posture Management, and partnerships such as being the first AWS partner to integrate with its Application Migration Service. This aligns with industry trends towards higher demand for observability and security solutions. Dynatrace’s growth is a sign of enterprises maintaining their current strategies for data management, performance, and security.

IBM released its 2024 Cost of a Data Breach report — The global average cost of a data breach rose 10% to $4.88 million. Shadow data, or data stored outside the main secured system, contributes to these costs. IBM’s report shows that security AI and automation can save an average of $2.22 million. It stresses the importance of AI-driven security, breach response training, and securing generative AI to reduce risks and expenses.

My thoughts: This isn’t surprising. Enterprises are constantly adding SaaS applications to their technology stacks, which increases the volume of data. This expansion creates more opportunities for data breaches and increases security risks.

Box announced its acquisition of Alphamoon, an AI startup specializing in intelligent document processing. This is a strategic move to strengthen the Box AI content management platform. By automating metadata extraction from complex documents using OCR, large language models, and a no-code interface, Alphamoon’s technology empowers users to streamline workflows and make data-driven decisions. This acquisition complements Box’s earlier purchase of Crooze, a no-code workflow automation platform, and solidifies its vision for an Intelligent Content Cloud. Combining these technologies enables Box to move beyond storing and managing content to intelligently process and automate workflows around it.

Network detection and response (NDR) can be an effective tool for monitoring and detecting suspicious activity. HPE Aruba Networking recently announced its offering in this area that applies AI to behavioral analytics. For this to be effective, infrastructure providers must have significant data lakes to train and refine models. HPE is among a handful of providers that can effectively lean into its own data lakes to provide customers with optimal NDR outcomes.

Körber Supply Chain Software has acquired Transportation Management Systems (TMS) provider MercuryGate. This expands Körber’s capabilities in transportation management, giving it a more integrated solution across the supply chain. The acquisition is part of Körber’s broader strategy to provide advanced features for logistical visibility to help companies make good decisions especially when there are transportation disruptions. This acquisition also places Körber in a stronger competitive position within the supply chain software market against SCM competitors Blue Yonder and Manhattan Associates.

PayPal has launched Fastlane, a streamlined guest checkout solution, for all U.S. merchants. By securely storing customer data, Fastlane allows returning shoppers to complete transactions with a single click, while new users can opt in to save their details for future use. During its initial testing phase, Fastlane demonstrated significant success, leading to a 34% increase in conversions compared to traditional guest checkout processes. PayPal anticipates that Fastlane’s widespread availability will help merchants reduce cart abandonment rates, particularly during the upcoming holiday shopping season.

Streamlined checkouts enhance the user experience by eliminating repetitive data entry and simplifying the payment process, particularly on mobile devices. Customers feel more comfortable sharing their information with reputable providers who they are familiar with and who have a reputation for secure data handling.

Fastlane is now generally available on PayPal Complete Payments and PayPal Braintree. Merchants can also access Fastlane through platforms such as Adobe Commerce, BigCommerce, Salesforce Commerce Cloud, and others for U.S. merchants—meaning it could have a substantial footprint.

Dayforce reported stronger-than-expected Q2 2024 results, showing significant growth in customer base and adoption of its AI-powered solutions. This positive performance further establishes the company as a formidable competitor in the expanding HCM market. Dayforce’s appeal lies in its cloud-based platform, which offers a modern and agile alternative to traditional on-premises systems, making it particularly attractive for enterprises. The company’s strategic emphasis on human-focused AI integration appears to be resonating with customers. On the earnings call, the company reported that revenue for its on-demand pay solution, Dayforce Wallet, is expected to more than double this year, and it is the fastest-growing product at the company.

When I first covered Dayforce Wallet in my research, I saw much potential in the offering. Dayforce was already emerging as a leader in the space with earned wage access, and the roadmap of features indicated that the company was very focused on providing additional financial health tools for workers who get paid through the Dayforce platform. Dayforce has processed $4 billion in on-demand, early direct deposit, and paycard payments to employees using Dayforce Wallet.

Dayforce Wallet is also integrated with Dayforce’s existing platform so the technology can produce accurate, real-time calculations and remain in compliance with relevant regulations. Dayforce has the opportunity to unlock significant revenue streams by expanding its offerings to include services such as loans, insurance, mortgages, and investments through the app and marketplace.

Mbed, Arm’s software development platform for Cortex-M embedded devices, has reached its end of life. The announcement did not surprise Mbed developers because Arm decreased support years ago and no longer maintains the open-source Mbed OS project. The online tools will remain accessible until July 2026.

The Mbed story began 20 years ago when Arm introduced its first 32-bit microcontroller, the Cortex-M3. These tiny computer systems required specialized development toolchains and techniques, and the learning curve was steep, so a couple of smart engineers built a simple online interactive development environment (IDE) and devised a clever way to inject a system image into a prototype device using a simple USB connection. Partner chip companies were eager to provide compatible boards, so over the next few years, Mbed introduced a new generation of programmers to the world of embedded devices.

Since then, despite many upgrades to the OS, libraries, and toolchain, Mbed has been used mainly for experimentation, education (it inspired the BBC micro:bit kits), and prototyping. Although Arduino adopted Mbed libraries and tooling inside many of its platforms, Mbed never captured much market share in commercial applications because professional developers use mainstream IDEs, and Mbed OS was not competitive. Mbed’s demise is the latest evidence of IoT’s evolution from custom, one-off, DIY projects to platform-based, scalable product development. Arduino and most other Mbed customers are switching from Mbed OS to Zephyr. Zephyr is emerging as the clear winner in the microcontroller OS wars, and some enthusiastic fans refer to it as “the Linux of microcontrollers.”

As Arm’s Mbed platform closes up shop, Raspberry Pi is open for business with Pico 2, a new version of the company’s small (21mm x 51mm), inexpensive, microcontroller-powered single-board computer. Pico 2 is a significant upgrade from the original Cortex-M0+-based Pico, featuring two processor options—a dual-core Cortex M33 or a dual-core RISC-V Hazard3. I think this is the first module of its type to have a checkbox option for Arm vs. RISC-V processors.

The new board has twice the memory, twice the flash, and several security enhancements, including secure boot. Developers write software for all Picos in MicroPython, CircuitPython, C, or C++. The Pico W variant of the original Pico has built-in Wi-Fi and Bluetooth (Infineon 43439), and I expect a Pico 2 W by the end of this year. Prices are astonishingly low—about the cost of a fast food value meal. The original Pico is $4, the Pico W is $6, and the Pico 2 is $5. I suppose the Pico 2 W will be $7.

Raspberry Pi Linux boards are great for learning and experimentation, and many industrial applications also use the platform, but the original Pico has no security and a very small processor. The Pico 2, with its additional processing power, security, and memory, might find its way into some low-volume commercial products.

Google announced the Google TV Streamer, a TV-attached box (not a dongle) similar to Roku, Apple TV, and Amazon FireTV. Customers can pre-order for $99; that’s competitive with higher-end units like Apple TV but much more costly than Chromecast dongles and other cost-optimized streamers. The Google TV Streamer defines a new product category because it runs Android, uses Gemini to curate TV experiences, and functions as a Matter controller. The company hasn’t yet made press units available, so expect my review in late September.

Will the Matter controller support third-party extensions? That’s important because that feature would fulfill Matter’s promise of controller unification. The alternative is ugly; each third-party interactive app would need its own controller and ecosystem, which would be chaotic. Stay tuned!

During the Black Hat conference, a group of researchers discovered and disclosed a 5G baseband vulnerability found in most 5G modems. They created a tool to let people find and fix their own vulnerabilities as well, but it also demonstrated how supporting older standards makes 5G more vulnerable because many hacks of cellular networks force you onto a lower, less secure G.

Vodafone has found that 39% of businesses surveyed in the UK are ready for 5G standalone services and has begun rolling out a separate 5G SA service for enterprise and SMB customers. 5G SA will likely benefit businesses the most, while also offering operators additional revenue to help pay off the added cost of deploying a new 5G core and even the expanded 5G networks.

The Pentagon’s FutureG office continues to explore and deploy new 5G use cases for the U.S. Department of Defense worldwide. It recently disclosed a project in Africa where it partnered with Anduril to build surveillance towers for security purposes in AFRICOM to help protect American bases.

Zoom has expanded its offerings beyond video meetings with the launch of Zoom Docs, an AI-powered collaborative document creation and editing solution integrated with its AI assistant, Zoom AI Companion. Zoom Docs aims to improve teamwork and streamline workflows within the Zoom Workplace platform, enabling users to collaborate more effectively and access critical information within a unified workspace.

With this new offering, Zoom is directly challenging established players such as Google and Microsoft in the collaborative document space. By integrating AI capabilities and leveraging its existing video conferencing strengths, Zoom is positioning itself as a comprehensive productivity platform for the hybrid workplace. This move highlights Zoom’s ambition to compete beyond its core video conferencing market and become a more holistic solution.

A U.S. federal judge has ruled that Google has a search monopoly, but Apple says there is no price it would pay Microsoft to use Bing as the default search. I believe that competition for search is good for consumers, and while I enjoy using Google, it has become bloated and less useful over the years.

Google is killing the Chromecast in favor of a MediaTek-powered TV set-top box which has many of the same capabilities but adds Thread connectivity and Matter support for enhanced smart home controls. (See more from Bill Curtis under the “IoT and Edge” heading on this page.) This indicates to me that Google is looking to expand the power of its ecosystem into the smart home market—and to tighten the integration across Google’s devices.

Sony has made the PSVR2 PC adapter kit available for $60. People are discovering that it’s a virtual link box that takes advantage of an already existing standard to connect the PSVR2 to a PC.

QuEra Computing has launched the QuEra Quantum Alliance Partner Program. The objective is to accelerate the development of neutral-atom quantum computers. Initial members of the alliance have expertise in various fields. The membership includes BIP, BlueQubit, Classiq, E4 Computer Engineering, ITQAN Al Khaleej, Kipu Quantum, Links Foundation, Pawsey Supercomputing Centre, Phasecraft, QAI Ventures, qBraid, QCWare, QMWare, QPerfect, Quantum Machines, QunaSys, Strangeworks, Venturus, and Wolfram Research.

Yuval Boger, chief commercial officer at QuEra Computing, said that the alliance represents QuEra’s commitment to fostering collaboration and driving innovation in the quantum computing landscape. Members of the group will be able to pool resources and share expertise for accelerated development, the promotion of new technologies, improved market penetration, and a more competitive and innovative tech industry.

Last week I attended Black Hat 2024, meeting with Fastly, HP, HPE, IBM, Infoblox, NTT, Wiz, and others. I was particularly impressed with what Wiz is accomplishing in cloud security with a complete portfolio and an easy-to-navigate user interface that consolidates multiple functionalities. The company is one to watch—especially now, given its confidence in walking away from a huge payday to be acquired by Google in lieu of a future IPO.

Dell and Alienware are honoring Intel’s extended warranty on Raptor Lake CPUs, offering a five-year warranty on chips. This is a great move because the failure rates for these chips are actually still quite low and will represent only a small uptick in service costs, but making this move will build a lot of brand loyalty and trust for Dell and Alienware. Lenovo and Acer have yet to disclose their policies on these Intel chips.

Humane, the company that makes the AI Pin, has been shown to have significant product returns on its already meager sales. The AI Pin continues to struggle because of its high cost, low battery life, limited functionality, and outdated chipset. Ultimately, Humane would have been better suited for wearables like what Meta has built with Ray-Ban.

Snapdragon, Manchester United’s current shirt sponsor, is reportedly exploring acquiring the naming rights to the football club’s iconic stadium, Old Trafford. Beyond brand visibility, the naming rights could enable Snapdragon to transform Old Trafford into a showcase for its cutting-edge technology. Qualcomm imagines a stadium with state-of-the-art connectivity, immersive fan experiences powered by Snapdragon processors, and innovative digital solutions that redefine how fans interact with the game. If the move succeeds, it would create a unique and memorable experience for fans and serve as a powerful demonstration of Snapdragon’s capabilities to a global audience.

In another significant move associated with Snapdragon’s Manchester United sponsorship, Qualcomm and Microsoft announced a partnership to add the Copilot+ PC logo to the back of the club’s jerseys. Qualcomm’s Snapdragon X Series processors exclusively power the next-gen Copilot+ PCs. The team displayed the new kit this past weekend at the Community Shield game that pitted Manchester United against crosstown rivals (and defending Premier League champions) Manchester City. The collaboration with Microsoft to highlight the Copilot+ PC logo on the jersey is a testament to the strong partnership between Qualcomm and Microsoft in PCs. Another interesting point of this deal is that the Copilot+ PC logo will appear only on the players’ jerseys, not the replica jerseys available for sale. This could indicate that Qualcomm intends to rotate partners or brands on this sports jersey “real estate.”

As we watch this year’s Olympics, have you noticed the advancements in sporting gear? The 2024 Paris Olympics has been a showcase for the intersection of sports and technology. Here are some key examples:

• Moisture-wicking tops reduce muscle vibration and fatigue for track and field athletes
• Swimwear made from polyurethane and hydrophobic textiles repels water and minimizes drag.
• Lightweight Nike shoes with carbon-fiber plates provide durability and grip on track surfaces. Patents cover the shoe’s sole structure, cushioning arrangement, and cleat patterns.
• Biometric sensors, motion-capture instruments, and data analytics platforms enable athletes to fine-tune their techniques and analyze physical performance.
• Virtual reality (VR) goggles help the Australian swim team visualize and optimize the speed of relay changeovers.
• AI monitors social media for cyber abuse and helps law enforcement track unusual patterns to improve event security.

It’s great to see the evolution of technology in sports. I expect many of these innovations to eventually be commercialized for consumer markets.

T-Mobile recently announced an initiative in partnership with Cradlepoint to lower the barriers to the adoption of 5G-connected PCs, as well as LTE and 5G fixed wireless access services. The carrier will work with IT distributors Ingram Micro and TD SYNNEX to offer a channel subsidy program designed to lower the cost of mobile-broadband-enabled PCs and FWA customer premise equipment. It is a great start, but less than stellar provisioning of these devices in the past must be addressed for the program to be ultimately successful.