Monday (telco diary) | AI sovereignty for Industry 4.0

0
1
Monday (telco diary) | AI sovereignty for Industry 4.0


From the newsletter (sign-up if you want it sooner): Europe’s AI sovereignty push is gathering pace with the launch of Soofi S, while telecoms and governments reposition networks, chips, submarine cables, private 5G, and data centre connectivity as strategic assets in the emerging AI economy.

Here’s a round-up of news and other bits today, starting with some non-telco news about AI frontier models in Europe – but very much a part of the sovereignty discussion and new economic workloads that telcos are engaged in. Plus Deutsche Telekom is involved. So, Soofi S has just been released, in case you missed it – as the first model from the Sovereign Open Source Foundation Models (SOOFI) project, coordinated by the German AI Association (KI Bundesverband), with key research partners including the Fraunhofer Institute and the German Research Center for Artificial Intelligence (DFKI), plus six German universities. It is a 30-billion-parameter mixture-of-experts AI foundation model designed for efficient inference and long-context workloads.

Soofi S follows Mistral AI, from France, as an alternative to models from the likes of OpenAI, Anthropic, and Google, plus their Chinese peers. Mistral is negotiating to raise approximately €3 billion ($3.5 billion) at a valuation of about €20 billion. But Soofi is different, too – particularly to its non-European counterparts. Its S-model is a publicly funded open-source product. It is trained on Deutsche Telekom’s Industrial AI Cloud, and pitched for industrial AI apps. Even compared to Mistral AI – with US-style commercial models, cloud deals, and benchmarks – Soofi is an Industry 4.0 project. Mistral is Europe’s attempt to build a global contender; Soofi is its attempt to build an industrial champ. The importance of both has shifted after the US killed foreign access to Anthropic’s best models, since overturned.

China is considering similar restrictions on its top models. AI access and ownership is a hot topic for governments, as we know. There is related news today, also – including closer coverage of the industrial telecoms space. Gartner has a good review, for one, about the top AI chip makers across 40-odd categories – Nvidia, AMD, and Broadcom are holding up the US end, among the ones-to-beat; TSMC is from Taiwan; Infineon is German. In straighter telco matters, new ITU recommendations and FCC regulations show how submarine cables are being recast as critical infrastructure, with the ITU looking at international resilience bottlenecks, and the FCC tightening security (and loosening some building regs) around US systems, particularly vulnerable terminal equipment in landing stations.

Besides, there appears to be some lobbying by AT&T of the Mexican regulator, set to auction private 5G licenses in the country. You can imagine the fuss it is making. Meanwhile, the Australian regulator has new planning rules for the 1.9 GHz band, which (as Dean Bubley pointed out on LinkedIn; with thanks) includes expanded support for non-3GPP private DECT NR broadband, with new “indoor-only” rules for indoor comms at 1900-1920 MHz, and a reserved outdoor tranche at 1900-1910 MHz spectrum for railway comms. Which means, slowly, the industrial space is bubbling along; note, as well, we have a follow-up to last week’s ramble about private 5G, based on new SNS figures about the market’s five-year CAGR (34%) and valuation ($6.6 billion) – specifically about private 5G’s role in physical AI.

Also, for note: the DCI and DC-related connectivity bonanza is all-go. Finnish operator Elisa has signed its first large-scale data center connectivity deals, it says, and HKT is to launch Hong Kong’s first 3.2 Tbps AI interconnect superhighway using hollow-core fibre – to cut latency by about 30%. Lots to get your teeth into, then.