The Future of Tech - Key Themes for 2026 and Beyond

Speaker A: Foreign.

Speaker B: Welcome to our, uh, conversation exploring the technology shifts set to reshape value creation, capital allocation and deal making from 2025 to 2035. This podcast is aimed at investors, M and A deal makers and corporate strategists. We will highlight the most important themes driving change in the world of technology and the opportunities that come with them. Um, my name's Alexander Love and I'm a senior editor for Global Data. AI continues to dominate boardroom discussions. The technology has continued to evolve substantially even since the introduction of ChatGPT in November 2022, which was a substantial leap in itself. New layers and new functionality since are redefining how AI will be of value. Joining me today is Isabel Aldehir, who is Principal Analyst, Strategic Intelligence Division at Global Data. Isabelle has been tracking these shifts and what they mean for businesses navigating the next decade. We will be focusing today largely on AI, but also quantum computing. Key themes that every tech investor and dealmaker should have on their radar. This, uh, podcast is brought to you by Sterling Technology, the provider of premium virtual data room solutions for secure sharing of content and collaboration for M and A deal making and capital raising.

Speaker A: Think.

Speaker B: Hi Isabelle, welcome to the podcast.

Speaker A: Thank you. Pleasure to be here.

Speaker B: Firstly, I wanted to talk, sort of give an overview of the investment opportunity in AI. And let's start with the big picture about why this decade is different and where value is likely to concentrate. So from your perspective, what has made generative AI a genuine step change for investors in the tech space rather than, uh, just the next step in a long line of advances?

Speaker A: I would say we are experiencing now the very well timed intersection of several technologies that have been under development for decades. Uh, now we're reaching a point of complementary maturity. Generative AI didn't appear out of nowhere. Those who are well acquainted with the field will know that the earliest chatbots date back to 1960s. You know, these are very simple chatbots, not based on neural networks at the time and certainly not transformers. But they were essentially the very early groundwork for where we are now. The Transformer model developed by Google, uh, the backbone of most generative AI models today. Uh, it was developed almost 10 years ago in 2017. Uh, but it wasn't until OpenAI made use of it when developing ChatGPT that its full potential was really grasped. Same with the chip technology that's providing all the compute. The chips industry is again decades in the making. Since 1950s Nvidia, which must be by now one of the best known companies in the world. Um, it's Been around since the 1990s. Most of its existence it was a relatively unheard of company and it wasn't even an AI chips company. But now its name is everywhere. Its GPUs, its graphics processing units have become fundamental hardware for running generative AI models. And it's not just Nvidia anymore. Uh, amd, Broadcom, Google, uh, they're all pushing out their AI training and inference, uh, chips. And of course memory is also super important in the AI world. And that's where Micron, sk, Hynix and uh, Samsung, uh, also come into play. Yes, AI is a next step in a long line of advances, but it is a pretty big next step and it's because of the happy marriage of numerous different technologies coming, uh, to this, a very uh, fruitful point following decades of what you might call independent development in the sense that much of that development was not even exclusively intended for AI.

Speaker B: So looking forward between 2025 and 2035, what can investors in the tech space expect? Um, do you think AI will remain at the center of conversations over the coming dec.

Speaker A: The short answer is yes. Uh, certainly AI is still going through a lot of evolution. It's not a technology that has reached a point of stagnation or plateau, um, and it has a lot of growth potential, especially with the Agentic AI component which was largely unheard of in 2024. But last year, in 2025 talk uh, of Agentic AI became ubiquitous. And now in 2026 we're already seeing a lot of real, uh, agentic use cases. For example agentic E wallets where AI agents now have the cap payments independently. And a lot of fintech and traditional payment providers are rolling this out as we speak. Uh, and what I mean by agentic AI, I'm referring to AI systems that can make decisions and execute them. And this is different to generative AI which generates content but doesn't autonomously perform a subsequent action to meet an objective. Of course, what is already at the center of many conversations, uh, that this topic of return investment, both from the perspective of AI adopters, the enterprises in particular, and also from the perspective of the AI developers and ROI is hindered by how expensive it is to run the infrastructure supporting AI. And as models become more and more intelligent and capable of reasoning, which is happening every time a new model is released, the more processing is required to carry out those reasoning steps and so the greater the consumption of tokens, the greater the compute power required, the more data centers are needed, uh, the greater the energy usage and so on. And so forth. And so the business model is going to be very important because, you know, how are we going to keep producing and running increasingly intelligent models whilst affording the simultaneous increases in expenditure? And we're seeing quite a big divergence between rivals. OpenAI and Anthropic. Uh, like most, if not all pure play AI companies, they are operating at massive losses every year. However, Anthropic is focusing its attention on enterprise customers and based on estimates, something like 80% of Anthropic's revenue comes from enterprise customers rather than consumers, which is a very different picture to OpenAI's revenue split. Uh, but then OpenAI is currently trialing ads amongst its US customers, uh, short as a means of assessing how well, uh, advertising can support the business financially. Something that Anthropic has stated, uh, in a rather creative fashion that it will not copy. So there are diverging business models and diverging opinions on advertising too. It remains to be seen how profitability will be achieved. Hence all the conversations these days around the AI bubble. Uh, and also just talking about the pure play AI companies. There are a lot of startups entering the market now. Many of these are founded by former employees from the larger firms. For example, Advanced Machine Intelligence Labs was founded this by Jan Lecun, previously chief scientist at Meta. You know, the amounts of money being raised straight from the get go is really quite astounding. Thinking Machines Lab has already raised I think $2 billion and is seeking to raise its valuation from 12 to $50 billion. And the uh, Advanced Machine Intelligence Labs is looking to raise about half a billion straight out of the gate. And what is particularly impressive is that much of this funding is being obtained before any major products have been released. You know, investors are going off of potential. You know, yes, there's plenty of activity in the AI space and I would expect topics around, um, technological development, competition, talent, poaching, roi, you know, as well as other areas I didn't mention, such as the energy dilemma and ethical quandaries to remain at the center of conversations over the next years. There's a lot going on at the moment.

Speaker B: Yes, of course. And you mentioned ROI a couple of times. I wanted to come back to that. Why do you think it's difficult for companies adopting AI to provide figures and what other metrics are important to measure success of AI implementation?

Speaker A: I mean, I think we are being premature expecting ROI numbers so soon. It takes time to adopt AI. You know, you don't just flick a switch and light up all your workloads with AI, there are a Lot of steps to go from ideation to pilot to uh, full scale production and then beyond that to AI native. Considering that ChatGPT was only released at the end of 2022, it's difficult to report an ROI. It's possible we're expecting too much right now. In many cases, AI is being applied to very specific functions in a company, almost as an add on you really being integrated across an entire company. There are very few AI native companies out there and um, that refers to when a company is fully integrated at its core with AI. When Microsoft PowerPoint started to become a staple on workplace computers, replacing the more manual preparation of presentation materials, I don't think businesses were isolating the roi. But I understand the dilemma. A lot more money is going into AI systems compared to other productivity tools. So how do we know all this effort and financial investment is worth it without quantifying it somehow? Uh, in the meantime, until ROI does become calculable, there are other metrics that can be captured on more immediate timescales, such as customer satisfaction, uh, employee AI adoption rates, failure prevention rates and so on. Any, uh, specific metrics would of course depend upon the industry in question. But the point I want to make is that ROI is not the only metric to judge successful implementation. And it's interesting to see what's happened in the market recently, uh, uh, or in February, uh, Anthropic released two AI features, Cowork and Claude Code Security, uh, both of which of course caused a lot of, uh, market disruption amongst data providers and information services stocks and also cybersecurity stocks. And separately, the release of some other AI tools by other companies caused, uh, disruption in wealth management and price comparison stocks. And the market reactions were literally immediate. Uh, there was no waiting in seeing whether these AI plugins actually replicate and reproduce the same value as incumbent firms do. And particularly in the case of Anthropic's, uh, CLAUDE Code Security, which scans code for security vulnerabilities. It barely scratches the surface of what cybersecurity companies deliver. And yet major stocks fell nearly 10% the same day and even more by the following week. Investors are clearly anticipating the disruptive impact of AI and the success that it will bring. And there's a trend that we are evidently seeing where investors are expecting immediate impact from AI tools, which is intensifying the pressure for ROI and equivalent metrics.

Speaker B: I wanted to look at the future of AI and what is coming next. You mentioned agentic AI earlier and how it is maturing. What is it that makes agenc AI agents so attractive to enterprises.

Speaker A: I'll start off by explaining the problem companies are having with generative AI, and then I'll explain the attraction of AI agents. So with generative AI, as I'm sure all the listeners know, you prompt the model and it will produce some kind of summary or output based on the training data it has. And yes, this can save an employee a lot of time, but it's not really transforming a business's operations just yet. It's primarily a productivity enabling tool. And this goes back to the topic of roi, the difficulty of calculating it. It can be difficult to isolate precise financial savings from productivity tools without a lot of very diligent tracking. Now, uh, onto AI agents, uh, a company can assign objectives and targets to AI agents and they can set various constraints and then allow the AI agents to operate to achieve those targets. And this is where the business disruption occurs because it's no longer just about using AI to save time or produce content more quickly. AI agents, um, will receive information from various sources, including for example, IoT sensors, uh, internal company databases, any kind of specialized knowledge sets. Uh, they may also be prompting large language models independently. And once the AI agent obtains the information it needs, it can execute a decision to meet its objectives, whether that's regulating an equipment's output on a factory floor or making particular purchases when conditions are most optimal. AI agents are now, uh, replicating the work that humans traditionally did, uh, potentially at higher accuracy and at lower cost. And then with multi agent systems where AI agents communicate with each other and collaborate, this removes the silos that humans usually operate in and which make it so difficult for interdepartmental across division collaboration, especially in large organizations. And so the ultimate aim from the enterprise perspective is not just to substitute human roles with AI agents. There's a limit to how much you can do that, but ideally to generate also new revenue streams. And that's where the real value would be unlocked.

Speaker B: Of course, now regarding large language models, they dominate discussions around generative AI. What do you think about critics who say that these LLMs are reaching a point of diminishing returns? And do you think we will see other kinds of models in the future?

Speaker A: People started talking about diminishing returns quite early, uh, back in 2024, maybe even the end of 2023, when the emphasis was on larger and larger models with more data and trillion plus parameters. And yes, there were indications that results were not much better than previous iterations. And that's where this critique of diminishing returns started. But I think with the advent of reasoning models, there was another sizable jump forward because LLM's answers became much more robust and explainable. And when LLMs are combined with that agentic component I mentioned just now, there's a lot of value to be captured, but there is still wall that LMS will eventually reach. And this is something that Demis Esabas and Iann Lecun and several other leading figures in the AI world are talking about a lot now that LLMs cannot actually be considered intelligent, nor will they ever be able to emulate, um, especially human intelligence in the way that we learn and think. Generative AI today built on large language models is not intelligent, although it does an excellent job at uh, imitating intelligence. So you can say that LLMs are predictive tokenization machines. They are highly adept at, uh, learning statistical relationships between words and data, but they lack comprehension, you know, that they don't actually understand the user's input, um, or the output generated. And Yann Lecun, who I mentioned earlier, he has said he is no longer interested in large language models and he left Meta at the end of 2025 to launch his own startup, Advanced, uh, Machine Intelligence Labs, which has just raised a billion dollars. And it is exclusively focused on something called world models. And world models are fundamentally very different to large language models. Uh, so world models, they possess a physical understanding of the situation or environment they are representing and they essentially make predictions using the knowledge of physical rules and boundaries about the future state of that situation. Uh, in this sense they have comprehension because then they're not just detecting patterns, they're understanding cause and effect. And this is much more like, uh, human intelligence. And it's a particularly relevant technology for robotics and autonomous vehicles where there are so many possible future physical states. And there are companies like Uh1x and uh Wave that are exploring world models exactly for these applications. Another recent startup in 2024 is World Labs. And of course there's also the one by Yann Lecun I just mentioned. So this space is quite startup dominated. Google does have world models, uh, produced by Hasabis and the DeepMind subsidiary. They're called uh, Geni and Meta has produced world models, but that was largely driven again by Yann Lecun who has now left because he didn't think he could explore them sufficiently with so much emphasis still on large language models. So it remains to be seen if Meta will continue in house with them now that Lecun has left. But even if large tech companies don't pursue world models in house, I wouldn't Take that as a condemnation or a verdict against world models. These large companies have poured billions into LLMs and you can't just change track suddenly, whereas the startups have a lot more flexibility to do the groundwork. And I would imagine that um, the large tech companies will be monitoring closely to make timely acquisitions or ACRI hiring as is now quite common. But there is a growing consensus that to achieve artificial general intelligence and subsequently superintelligence, it won't be with language models, it will be with world models.

Speaker B: And uh, the winners of course won't be the ones who deploy fastest, they'll be the ones who manage the risks and constraints better than everyone else. Just off the back of that, regarding execution risks, security and resilience, that's what I wanted to look at next. Because as you know, AI isn't the only technology that is going to disrupt in the next decade. Developments in quantum computing are making headlines not just in technical journals, but also in mainstream media. And while there are many opportunities for quantum computing to solve problems that our existing computers cannot, many people are concerned about the threats of quantum computing as well. What kind of threats are ah, we talking about here and just how serious are they?

Speaker A: They, well quantum computing is something of a double edged sword. So yes it can do computations that are impossible to do on uh, today's classical computers alone due to quantum principles like uh, entanglement and superposition. And there's so many opportunities that quantum computing will bring, particularly uh, in uh, things like optimization, multivariable decision making, molecular simulations and so on. Quantum computing is indeed hitting the mainstream media. There are uh, numerous companies in uh, finance, pharmaceuticals, uh, automotive industries, uh, partnering with quantum computing companies to carry out pilots and experiment with different use cases. Uh, some industry names that come to mind are JP Morgan, hsbc, Johnson Johnson, Asher, uh, Zenecker, he and I. There's nowhere near an exhaustive list. There are a lot of names looking at quantum these days and uh, some of the companies these names are partnering with are big tech companies like IBM, Google, Microsoft, but also a lot of um, uh, pure play quantum startups like IonQ, PSI, Quantum D Wave, Xanadu and AT Global Data. We think that from 2030 we will start to see quantum computing commercialization begin in earnest as those pilots mature. And from 2035 we are anticipating a point of maturity to be reached where the quantum opportunities are very real and accessible. But also the same can be said about the threats. And that's why it's a double edged sword. A major risk that quantum computers will pose in the future is the ability to run Shor's algorithm, which would effectively decrypt much of today's standard encryption protocols, allowing data to be hacked. Not just recent data, but all historic data protected by these encryption protocols is at risk. And um, interestingly, Shor's algorithm was actually developed in the 1990s, but it requires a quantum computer to be executed. And alarmingly, there are hackers stealing data now, now knowing that they can't access it for the time being. But, um, are waiting for quantum computing to provide the key essentially. And this is called harvest now, decrypt later. And so there is a race against time to develop different methods of quantum safe security. And um, many businesses might not be aware of this threat. Some people might still think quantum is nothing but hype, uh, and I would suggest a less cynical view. The US is mandating the transition to quantum safe cryptography and the EU is following a similar route. So it is paramount for all, um, for all strategy and cybersecurity teams to know that this is coming and to ensure that your systems and data remain safe in the years to come.

Speaker B: And on that note, if you were to give some advice to enterprises now to prepare for these quantum computing threats, what would that advice be?

Speaker A: There are a couple of recommendations that come to mind. Uh, firstly, exploring the right partnerships. Uh, for example, Orange, uh, the telecommunications company, is rolling out quantum safe networking in the form of post quantum cryptography for its customers so that data transmitted across the network is safeguarded. Uh, and Orange is not alone, uh, within the telecom space there are others like Deutsche Telekom, SK Telecom, Vodafone, all working to deliver quantum safe networks. Uh, it's not just telecoms, um, but you need quantum safe security from your cloud provider too, for example. But in general you want to make sure that the partners you're working with have scope for continued innovation in this area. Because particularly in the case of post quantum cryptography, the algorithms being developed now might not be secure in the future as quantum, uh, attacks are expected or anticipated to become more and more sophisticated. Similar, um, to today's classical cybersecurity threats, um, encounter prevention. There will always be something of a cat and mouse game going on. Uh, and internally companies need to have a chief information security officer who understands the threat of quantum computing and knows what steps need to be taken. Um, quantum is a difficult subject matter and given the high stake risks, companies need to hire explicitly for quantum experts.

Speaker B: Thank you Isabelle for your insights and thank you to our audience for listening to this podcast brought to you by Sterling Technology, the provider of premium virtual data room solutions trusted by technology dealmakers to support their most important transactions.

Speaker A: Sam.