A new generation of HPE and Dell servers, the busiest month yet for open AI models, and a deadline approaching for everyone covered by NIS2. What any of this means for your infrastructure.
On 7 October HPE introduced the first four models of its thirteenth ProLiant generation, built on 6th-generation AMD EPYC processors: the DL525 and DL585a for demanding enterprise workloads, and the XD245 and XD285 in OCP form factors with higher rack density and more efficient cooling. The generation is designed with inference and agentic AI workloads in mind, offers more memory bandwidth and I/O, and brings automated security and management through Compute Ops Management.
What this means for you: if you are planning a server refresh for 2027, Gen13 is the generation to budget for – Gen11 is reaching the end of its life cycle, and Gen12 will stay in price lists for at least another year. For ordinary virtualisation workloads, Gen12 remains a good choice at a better price; for AI inference and data-intensive systems it is worth waiting for the Intel-based Gen13 models that HPE has announced as the next step.
Source: HPE Newsroom, 7 Oct 2026
On 21 October Dell expanded the Dell AI Factory with NVIDIA. The most interesting piece is the PowerEdge R8822 – an air-cooled, two-socket server with NVIDIA Vera CPUs in a standard 19-inch rack, built for agentic AI workloads, with support for up to four GPUs and available from October. The larger systems – the M9882 with 256 Vera CPUs in a single rack and the XE9810 inference server – follow in the first quarter of 2027. The Dell Pro Precision workstation with the GB10 chip and 64 GB of memory has been available since 23 October. Dell also reported that the AI Factory now has more than 6,500 customers.
What this means for you: Dell is saying that AI agents need CPUs and memory first, not just GPUs – good news for most organisations, because the R8822 fits into an existing server room without liquid cooling. For companies that want to try AI on a small scale, the Pro Precision with GB10 is the first realistic way to run a larger model on a desk for the price of a workstation.
Source: Dell Technologies, 21 Oct 2026
In three weeks, more models with freely downloadable weights were released than in the whole summer. On 15 October Meta released Muse Spark Open (816 billion parameters, 42 billion active, a one-million-token context window) – the strongest open model from the United States, but under Meta’s own licence and needing roughly 1 TB of memory. On the same day NVIDIA released Nemotron 3.5 Bolt (74 billion parameters in total, 9 billion active, with image and video input), IBM released Granite Switch 4.1 in 8B and 30B versions under Apache 2.0, and Aleph Alpha released Kolibri-1 for German and English, which runs on a single H200 or B200 card. Mistral published the weights of its Large 4 model at the end of the month.
What this means for you: for deployment on your own infrastructure, the important model is not the biggest one but the one that fits into the memory you have. Nemotron 3.5 Bolt and Granite 4.1 are sizes that run on a single server with two GPUs and can handle an assistant over company documents or form processing. When choosing, look at the licence: Apache 2.0 and MIT come without restrictions, while the licences of Meta and some Chinese labs have conditions that should be read before deployment in a public institution.
Source: deepset, open-weight releases July–October 2026
On 27 October HPE announced the integration of HPE Networking EdgeConnect (SASE) with HPE Mist and HPE Aruba Central – wired, wireless and security networking under a single agentic framework intended to automate operations and troubleshooting. Mist support arrives in 2026, Aruba Central support in the first half of 2027.
What this means for you: if your network runs on Aruba, nothing changes for you right now – but at your next expansion or SD-WAN tender it is worth expecting that HPE is pushing a single console for the whole network. For mixed environments (Aruba and Huawei) it remains important that monitoring is handled above the vendor level, not separately in each one.
Source: HPE, 27 Oct 2026
The amendment to the Slovak Cybersecurity Act that transposes the NIS2 directive has been in force since 1 January 2025. Organisations were required to have security measures in place by March 2026, and by March 2027 they face their first audit or self-assessment through the Unified Cybersecurity Information System of the National Security Authority (NBÚ). Operators of essential services that are not critical may replace the audit with a self-assessment; critical operators must use a certified auditor. The Act also requires a cybersecurity manager with direct access to the statutory body, and written agreements on security measures with suppliers whose work touches the operation of the organisation’s systems.
What this means for you: if you are in the NBÚ register, November is the last sensible moment to start the self-assessment so that findings can be remediated by March. And if suppliers service your servers or networks, the contracts with them must include security measures – including the contracts with us.
Source: itbezpecnost.sk – NIS2 and the Cybersecurity Act 2026
Write to us through the contact form – we will tell you which parts of this review concern you and which can wait.
Book a consultation ›