A New AI World Order and the Coming Decline of the Hyperscale

This CIR DeepTech begins with speculations on how we at CIR see the structure of high-tech industry changing internationally particularly where South Korea might stand in this story in terms of innovation and commercialization. This is prompted by the Korean presence at the recent CES and is designed to guide readers on where Korea might fit into their future product strategy. Then we review the just-announced market report publication schedule from CIR (DeepTech’s publisher) for much of 2026. As usual our planned reports each cover deep technologies that will leverage the data center infrastructure to produce new business revenues.

But this time it’s different. As has been true for 40 years or so, CIR’s publications will cover photonics, fiber optics, and AI, but with our acquisition of IQT Research from 3DR Holdings, we will be adding quantum data centers to the mix, with perhaps a little bit of robotics to spice things up in the edge network.

Then we jump into what we confidently predict will be the big trend of AI in 2026 and 2027 – non-hyperscale data centers and the coming counterrevolution against LLMs. Yes, you read that correctly. Of course, our revelations on this matter will be tucked in neatly below the paywall. So, you will have to take out a paid subscription to read it. But if you do you, I can promise you not just data center prophesies but a free copy of our highly regarded CIR Guide to New Technologies for the AI Data Center, which would cost you about $1,000 if you bought this Guide directly from our Web site.

Korea Joins the Pack: Finally a New World Order for AI

My late father often kvetched about how, all too often in his opinion, new technologies were invented in Europe (including the UK) and commercialized in the US. His point, I think, was that some geographies are, it seems clever, at being clever and worth watching for that reason alone. These places may or may not, and for various reasons, be good at turning their ideas into money – the UK. no; Japan (to some extent) and the US yes. I once told a British guy who had come up with an idea for early silicon photonics transceiver that he would soon move to Silicon Valley, because that’s where the money was made. As I remember it, he was quite rude to me, but sure enough, two years later he was waking up to his San Jose mornings.

As I got older and progressed down my path to market analyst-hood, I looked to the seemingly endless reports and other musings on future technology coming out of EU and Japan as subtle and well-thought out pieces on where (in my case) telecom was headed. It didn’t matter to me that no one – including the Japanese – thought that (say) uber-ambitious plans for ISDN were actually headed anywhere commercially. My guidance to product planners everywhere was then, and is now, to read such documents with enthusiasm as useful input for the marketing mix.

So, does Hyundai Motor’s ZER01NE indicate that South Korea is another place we should now look for product ideas. What was on display at CES was more than just the usual Korean consumer products and mobile electronics. In the exhibit below I try to show how the ideas embodied in what was on show might inspire the next generation of mobility, energy, and artificial intelligence and other technologies covered in this Substack.

Incidentally, the Hyundai connection does not imply that all the ideas come from Hyundai. ZER01NE is a Hyundai Accelerator and only one of the entities at CES was an in-house Hyundai start up, plus the fact that the Hyundai innovation hub has office in the US, Germany, Israel, China and Singapore, yet the Korean connection is undeniable. Specifically, Hyundai has said that ZER01NE is a bridge between Korea’s innovation ecosystem and the global venture network. Through the ZER01NE Company Builder, Hyundai invests in internal entrepreneurs who bring ideas beyond the automotive domain. To date, 40 in-house startups have been successfully spun off into independent ventures. In parallel, the ZER01NE Accelerator identifies and supports promising external startups. It has been claimed that in recent years there has been a broad shift in Korea’s industrial approach—from linear, in-house R&D to distributed, network-based collaboration with startups, research institutes, and individual engineers.

CIR Debuts Its Market Report Publishing Program for 2026

Communications Industry Researchers (CIR) has launched its 2026 Report Program for companies that can’t afford to guess where growth will emerge. Covering more than 25 critical markets, our reports deliver decision-grade intelligence for executives who must quickly turn technology into revenue. Our 2026 program targets where capital, customers, and competition are focused: quantum technology, data centers, optical networking, and AI. Each report highlights how to win business and funding. Additional reports will appear in the second half of 2026. The exhibit below shows CIR plans for the coming year.

CIR reports are not marketing fluff. They are for companies who must justify investment, accelerate sales cycles, and prove traction. We show how technologies move from demos and trials, to contracts, and large-scale deployments. We identify barriers that stall adoption and strategies that companies are using to break through challenges. Every CIR report delivers roadmaps and insight you can’t extract from public sources. Readers see how the successful players are structuring product strategies, manufacturing, R&D priorities, and go-to-market plans. This is intelligence for supporting pricing decisions, partnering, acquisition strategy, and fund raising.

We make it easy to validate our reports before buying. Customers can review tables of contents and report samples. If a report does not match your needs, CIR analysts can deliver custom studies that answer revenue, investment, and positioning questions. 2025 saw a major expansion at CIR. Enhancements to www.cir-inc.com are still in progress. If the “Reports Tab” does not provide what you need, contact us directly:

For Purchases and Report Samples: Missy Wade — missy@cir-inc.com.
For Detailed Report Guidance: Lawrence Gasman — lawrence@cir-inc.com.

So anyway, for our paid subscribers only (to join us click on https://cirdeeptech.substack.com/subscribe), you will find more support for the view for which CIR DeepTech is becoming famous – namely that what needs to change technologically are the AI platforms, much more so than the thermal, interconnect and processing technology. Our point has essentially been that there is nothing magical about LLMs and that we might be able to do as well with small-language models (SLMs) or some kind of rethought LLM at the very least without the burden of untested technology. As the discussion below indicates perhaps the future won’t be one replete with giant data centers and almost unbelievable capital expenditures.

Today’s LLM training and running ultra-fast inference have suggested that in the future ever-larger models in the future will require ever-larger, more powerful and centralized infrastructure. However, there are more challenges to this assumption, arguing that today’s AI infrastructure is shaped primarily by historical choices and business models. In any case, many real world AI implementations already use smaller models and one statistic that we have seen quoted is that SLLs can handle 70–80% of enterprise tasks, with only the most complex reasoning requiring large-scale systems in the future. Below the paywall we discuss the possibility of distributed AI models and their implications.