For two days, Antwerp took center stage in the global chip industry with IMEC’s ITF World Conference

While international readers likely know Antwerp for its historical role as a global trade and financial hub, flourishing as Europe’s commercial capital in the 16th century, later rising to become the world’s diamond capital, and now home to one of the largest ports on the planet. Locally, Antwerp is simply called ’t Stad, “the city”, as if no other city really matters (and yes, I might just have revealed where I’m from). Having spent my entire professional career at the crossroads of tech and finance, you can imagine my delight at spending two full days at IMEC’s ITF World conference right in the heart of Antwerp, at the Queen Elisabeth Hall. This iconic venue, nestled between the Central Station and Antwerp Zoo on Koningin Astridplein, holds a special place in my memory. My connection to that spot goes back at least 40 years.

This past Tuesday and Wednesday, hundreds of engineers, executives, and investors converged there for the 2025 edition of the International Technology Forum (ITF), hosted by Belgium’s own R&D powerhouse, IMEC. Over two intense days, some of the brightest minds in semiconductors, AI, health tech, and computing infrastructure gathered to exchange ideas, challenge assumptions, and sketch the outlines of a post-Moore’s Law future.

What’s the main takeaway from my two days at ITF World? Exponential technologies are accelerating through the global economy and their influence is expanding at lightspeed. We’re entering a new era of purpose-built compute, agentic AI, and convergent innovation, where algorithms, materials, and hardware are no longer developed in isolation but co-designed from the ground up. In this blog, I’ll take you through some highlights of the event, featuring standout speakers, key insights, and takeaways that matter for investors.

Luc Van den Hove (CEO, IMEC) – "We need a new playbook"

In the opening keynote of ITF 2025, Luc Van den Hove, President and CEO of IMEC, delivered a clear and compelling message: the semiconductor industry is at an inflection point. Moore’s Law, the long-standing doctrine of doubling transistor density, is reaching physical, economic, and ecological limits. In response, Van den Hove called for a radically new approach to chip innovation, one built on co-optimization, cross-disciplinary design, and sustainability by default. Van den Hove began by acknowledging the immense pressure the industry faces from rising AI workloads. Training and running large AI models are now among the most power-hungry computational tasks on earth. According to IMEC projections, continuing on the current path could lead to AI consuming over 5% of global electricity within a decade. “More silicon and more compute won’t solve this,” he said. “Smarter systems will.” His proposed alternative is a new innovation playbook grounded in systems thinking. Rather than focusing on transistor miniaturization alone, Van den Hove championed heterogeneous integration combining logic, memory, interconnect, and power delivery in vertically stacked systems. This approach, enabled by advanced packaging and hybrid bonding, can dramatically reduce energy per operation and accelerate data movement efficiency.

IMEC, he said, is uniquely positioned to lead this transformation. As a pre-competitive R&D hub, IMEC works with nearly every leading semiconductor firm  from ASML and Intel to Samsung and TSMC to prototype materials, architectures, and tooling years ahead of commercial deployment. Van den Hove highlighted IMEC’s research into near-memory compute, optical I/O, backside power delivery, and sustainable materials as pillars of this future architecture.

A key part of his talk focused on the importance of hardware-software co-design. In the AI era, chip architectures must be shaped not only by physical constraints but also by the demands of emerging models and workloads. IMEC’s collaborations with AI companies, cloud hyperscalers, and system integrators aim to bridge that gap early in the innovation cycle. Van den Hove also touched on geopolitics and resilience, noting that advanced chips are now considered strategic assets. He argued that Europe must invest not only in fabrication capacity, but also in the foundational R&D that enables technological sovereignty. IMEC’s expanding footprint in domains like quantum computing, photonics, and bioelectronics reflects this mission.

“We need a new playbook,” he concluded. “One where sustainability, performance, and purpose drive innovation, not just scaling.”

Investor insight: Look to enablers of design-technology co-optimization, especially those working in 3D architectures, power-aware logic, and sustainability-aware tooling. In our view, investors could look to companies enabling design-technology co-optimization, like Synopsys, Cadence, ASML, and Besi, as the chip industry shifts from transistor scaling to function-level integration. Leaders in 3D packaging, power-aware logic, and sustainable semiconductor tooling such as Applied Materials, Lam Research, and Infineon are poised to drive the next wave of trillion-dollar semiconductor growth.

Johny Srouji (Apple) – "Designing for performance per watt"

Apple’s Vice President of Hardware Technologies, Johny Srouji gave one of the most practical keynotes. He chronicled Apple’s journey from purchasing standard-issue CPUs to developing the A-series and M-series chips. These chips are custom processors that allowed Apple to redefine what energy efficiency and performance looks like in consumer electronics. By tightly coupling silicon design with software needs, Apple has dramatically improved performance per watt in iPhones, iPads, and Macs. Srouji stressed that Apple’s edge came not only from high-end design talent but also from relentless attention to cross-layer optimization: from transistor layout to app-level user experience.

He walked the audience through real examples where modest improvements in latency or thermal performance cascaded into better battery life and user satisfaction. The vertical integration model, he said, gives Apple the levers to make such optimizations meaningful and replicable. “Disruptive products can’t be built on legacy chips,” Bronner said. “Silicon must serve the product, not the other way around.”

He also touched on the development process, which relies heavily on simulation, early prototyping, and iterative validation across software and hardware teams. This approach helps Apple compress product timelines while maintaining tight quality control.

Investor insight: As illustrated by Srouji’s talk, vertical integration is accelerating as OEMs and hyperscalers move deeper into custom chip design. Amazon’s indirect stake in AMD and its in-house silicon (like Graviton), alongside Broadcom and Marvell’s expansion in tailored ASICs, highlight a broader shift. This trend creates interesting opportunities across the value chain in EDA (Synopsys, Cadence), advanced packaging (Besi, Amkor), and IP licensing (Arm, Alphawave IP).

Terushi Shimizu (Chairman Sony Semiconductor) – “Smarter sensing for smarter AI”

Sony Semiconductor CEO Terushi Shimizu highlighted the role of intelligent image sensors in enabling real-time AI. Sony’s stacked CMOS sensors with built-in AI can process visual data directly on-chip, reducing latency and power use, which is ideal for edge AI in devices like autonomous cars and industrial robots. The company is also advancing event-based sensors for ultra-fast detection.

Investor insight: Sony is positioning itself at the forefront of low-power, edge-based AI sensing, with strong upside in automotive, robotics, and healthcare applications.

Smart Pill Demo – "Digesting data in real rime"

In one of the most interesting demonstrations of the event, Aniek Even, an IMEC researcher, ingested a sensor-laden smart pill on stage. As it traveled through her gastrointestinal tract, the pill transmitted real-time data on pH, core body temperature, and oxidative stress levels, all visualized live for the audience. Naturally, it set off a ripple of laughter, “ohs,” and “ahs” throughout the room. The pill contains a proprietary technology called Reno’s Balance, which measures the equilibrium between oxidants and antioxidants, a key biomarker in conditions like inflammatory bowel disease. Combined with AI algorithms, the system promises to offer clinicians a continuous stream of diagnostic insights, replacing the need for sporadic and invasive procedures like endoscopies. This technology opens the door to fully passive diagnostics: devices that require no patient input beyond ingestion but yield high-resolution, clinically actionable data.

Investor insight: Passive diagnostics and smart ingestion tech will redefine chronic disease management. Companies that integrate hardware, real-time biosensing, and AI interpretation stand to disrupt multiple layers of the medtech stack.

Clément Farabet (VP Research Google DeepMind) – "From models to agents"

One of the most compelling talks at ITF 2025 came from Clément Farabet of Google DeepMind, who shared the company’s latest vision for the evolution of artificial intelligence. His keynote coincided with Google I/O, where the company unveiled a wave of AI-powered products, which provided a timely boost for its share price after months of market pressure.

Farabet laid out DeepMind’s core thesis: AI is moving beyond passive foundation models toward agentic, context-aware systems.  Systems that don’t just respond, but perceive, reason, and act within their environment. This transformation, he argued, won’t just reshape software architectures, it will also require a fundamental overhaul of the hardware stack that powers AI.The keynote began with a retrospective. Over the past few years, DeepMind and its peers have pushed the frontiers of AI with models like AlphaFold, Gato, and Gemini. These systems showed that single models can operate across domains, from protein folding and gameplay to language understanding. But the next leap, the speaker explained, involves building AI that moves beyond narrow capabilities toward general intelligence. Farabet argued that we are still early in that journey, just beginning to transition from experimental prototypes to competent, reliable systems. But all the key ingredients are now in place. The speaker outlined four foundational pillars for artificial general intelligence (AGI):

1. Situated Intelligence: AI must understand and interact with the real world in real time, through sight, sound, and speech. A demo called Project Astra, powered by Google's Gemini model, showcased this vision. Gemini processes real-time audio and video inputs, remembers past interactions, and connects to the internet and other tools. This makes it capable of meaningful, context-aware assistance far beyond traditional voice assistants like Siri or Google Assistant.
2. Creative Intelligence: The next step is generative capability, creating entirely new content. The model Veo was presented as an example, generating high-resolution 4K video from text prompts. It even animates still images, inventing motion and expression from scratch. These abilities mark a shift from recognizing patterns to generating novel, high-quality media.
3. Helpful Intelligence: AGI should be genuinely useful. The speaker demonstrated NotebookLM, a tool that lets users upload content like PDFs or web articles and instantly turn them into personalized podcasts or study aids. This reflects a future where AI can assist with real-world cognitive tasks in a seamless and productive way.
4. Autonomous Intelligence: The final pillar is autonomy, the ability of AI to act independently, initiate tasks, and complete objectives without being micromanaged. DeepMind illustrated this with internal experiments where agents act as “virtual interns,” navigating company systems, submitting code, and responding to reviews. These agents gather context, make decisions, and execute actions, all while remaining aligned with human oversight. It marks a shift from reactive AI to proactive collaborators capable of genuine initiative.

Together, these developments signal the start of a new phase in AI, where models are not only capable of understanding and generating information but also of engaging with the world in human-like, helpful, and increasingly autonomous ways.

A significant portion of the talk focused on the infrastructure challenges this new class of agents introduces. Unlike traditional AI workloads, which are compute-intensive but largely batch-processed, agentic AI requires continuous, low-latency inference, persistent state, and high memory bandwidth. These agents must remember, respond, and adapt, often on edge devices, where energy and compute resources are limited. To support this, DeepMind is investing in Gemma, a new open-source model optimized for edge deployment. The goal is to democratize agentic AI and reduce reliance on massive cloud inference stacks. The speaker emphasized that on-device intelligence will be essential for privacy, responsiveness, and sustainability, especially in fields like healthcare, robotics, and education. Lastly, the keynote also addressed governance. As agents gain autonomy, DeepMind is developing behavioral safety layers, auditability mechanisms, and transparent reasoning modules that allow developers and users to understand and control how agents reach their conclusions.

Investor insight: Multi-modal AI agents will require new inference engines, edge-ready LLMs, and secure runtime environments, creating opportunities for companies enabling memory, context, and autonomy at scale. Key enablers include Nvidia and AMD (AI accelerators), Marvell, Broadcom and Arista (custom silicon and AI-centric networking), Qualcomm (edge inference), and Micron, Samsung and SK Hynix (HBM).

Scott DeBoer (EVP, Technology & Products, Micron) – "The power of memory in AI"

In a keynote filled with technical depth, Scott DeBoer, EVP of Technology and Products at Micron, made one thing clear: memory is now the rate-limiting factor in AI acceleration. While GPUs and compute architectures tend to steal the spotlight, DeBoer used his stage time at IMEC ITF 2025 to re-center the discussion on High-Bandwidth Memory (HBM), the unsung hero enabling the AI revolution.

“Without memory, AI doesn’t scale,” DeBoer stated. “And without power-efficient memory, it won’t be sustainable.”

He began with a breakdown of AI infrastructure trends. As model sizes and training datasets grow exponentially, GPUs are no longer compute-bound, they’re bandwidth-bound. Inference workloads, particularly those requiring multi-modal context and memory, now depend heavily on HBM stacks that sit physically adjacent to compute cores. This proximity enables terabytes per second of throughput with minimized latency and energy loss. Micron, a key supplier in this space, has developed its latest generation of HBM with up to 20 vertically stacked DRAM dies, delivering twice the bandwidth at 30% lower power than the previous generation. DeBoer emphasized that the complexity of building these structures rivals that of GPUs themselves, in fact, HBM stacks can contain more silicon area and thermal challenges than the accelerators they support.

He walked through the architectural features that make this possible: through-silicon vias (TSVs), advanced die thinning, hybrid bonding, and ultra-low-k dielectrics. He also detailed Micron’s investments in thermal management, not just heatsinks, but novel materials and interface designs that allow heat to dissipate evenly across stacks. Another highlight was Micron’s push toward co-packaged memory solutions that integrate HBM directly with compute dies through advanced packaging techniques like 2.5D interposers and 3D fan-out. DeBoer explained that memory must now be part of the system design conversation from day one, not an afterthought.

“Memory is no longer a peripheral. It’s a co-pilot in AI system performance,” he said.

He also hinted at Micron’s roadmap beyond HBM, including low-power DRAM for edge AI, automotive memory for centralized compute in vehicles, and emerging memory types that blur the line between storage and memory. Finally, DeBoer addressed supply chain constraints. As AI adoption surges, demand for HBM is exploding, but the materials, tools, and capacity needed to produce it are highly specialized. Micron is ramping CapEx and deepening partnerships with OSATs, foundries, and toolmakers to meet demand.

Investor insight: DRAM innovation, 3D packaging, and thermal-aware interfaces are becoming critical as HBM (High Bandwidth Memory) emerges as the next competitive frontier in AI infrastructure. Micron, SK Hynix, and Samsung lead in HBM development, while Besi, Amkor, and ASE Technology enable the advanced packaging it requires.

Christine King (Board Director, Skyworks) – "Adversity and the AI opportunity"

Christine King shared her journey from a single mom with a high school education to becoming the first female CEO of a semiconductor company. She overcame adversity by pursuing education, starting with welfare assistance, and eventually landing a job at IBM. King excelled in microprocessors, leading to significant roles and innovations. She founded an ASIC business, which grew to a billion-dollar enterprise, despite initial skepticism.

Investor insight: King’s message resonates beyond AI. It’s a mindset that applies equally to investing. It starts with having a clear vision, doing the homework, seeking out the right experts, inspiring those around you, and showing up to do the work, every single day. No matter the field, the fundamentals remain the same. And when setbacks inevitably come, remember: adversity is your ally. As King put it, looking back on her own life, “that’s when the real growth happened.”

Oskar Painter (Head of Quantum Computing at Amazon) – "Quantum: Nine Orders to Go"

In a keynote marked by clarity, Oskar Painter, Head of Quantum Computing at Amazon, delivered a sobering yet forward-looking assessment of where quantum computing stands today and how far it still has to go. Painter pulled no punches: practical, scalable quantum computers remain at least nine orders of magnitude away from the reliability and scale needed for useful applications. “Quantum computing is still in its vacuum tube era,” he said. “Promising, full of potential, but nowhere near deployment.”

Painter began by outlining the current state of the field. Most quantum computers today operate with tens to hundreds of noisy qubits, plagued by decoherence, gate errors, and environmental interference. To make quantum systems fault-tolerant, that is, capable of executing meaningful computations reliably, researchers must increase coherence times, reduce noise, and develop effective quantum error correction (QEC).

Painter emphasized that AWS is taking a full-stack approach to solving this problem. The company’s QED-C (Quantum Error Detection and Correction) platform focuses on all layers: from superconducting qubit design and cryogenic control systems to compiler stacks and high-level abstractions for developers.

While the challenges are immense, progress is happening. Painter described recent AWS breakthroughs in high-coherence transmons, fast gate times, and cryo-compatible multiplexed readout systems. He also highlighted the importance of materials innovation, pointing to low-defect substrates and novel deposition techniques that reduce surface losses and increase qubit fidelity. But hardware alone won’t solve quantum’s scalability issue. Painter dedicated a large portion of his talk to software: namely, how quantum-classical hybrid algorithms and early error mitigation techniques are allowing small quantum systems to perform useful approximations in chemistry, materials science, and optimization. These are the so-called “NISQ” (Noisy Intermediate-Scale Quantum) applications limited in scope but increasingly real.

He was quick to temper expectations: “We are still years away, perhaps a decade, from a system with fully logical qubits and commercial utility.” However, AWS is building toward that goal methodically, through investments in cryogenic integration, superconducting chip fabrication, and QEC hardware acceleration. Painter also touched on AWS’s vision for cloud-delivered quantum compute, which will integrate quantum processors into the same fabric as classical HPC infrastructure. This hybridization, he argued, is not a transitional phase, it will be the norm, even when scalable quantum machines arrive.

Investor insight: Quantum computing is not a near-term revenue story, but quantum-enabling technologies might be. Companies providing tools, materials, or control systems for cryogenic environments, RF design, superconducting chip fabrication, or QEC software might have more near-term commercialization paths. Established players like Amazon (AWS), Microsoft (Azure Quantum), Alphabet (Google Quantum AI), and IBM (IBM Quantum) are building full-stack or hybrid platforms, offering long-term upside with nearer-term optionality through cloud access, research leadership, and commercial engagement. Investors with more risk appetite may look to IonQ, Rigetti Computing, and D-Wave Quantum, pure-play quantum firms developing novel qubit architectures and early-stage commercial applications. That said, we continue to believe quantum nvestors should think in years, not quarters.

Carolina Aguilar (CEO, Inbrain Neuroelectronics) – "Decoding the Brain"

Speaking about a power-women, Carolina Aguilar has been build with the same stamina as Christine King. Carolina gave one of ITF’s most visionary talks, revealing how Inbrain is pioneering the interface between the human brain and artificial intelligence. Using biocompatible graphene electrodes, her company builds minimally invasive implants that deliver high-resolution, long-duration brain signal capture.

Unlike legacy implants, which often degrade over time and provoke immune responses, Inbrain’s graphene arrays are ultra-flexible and capable of capturing neural signals down to individual action potentials. These signals are processed using AI models that learn to interpret motor intent, potentially enabling assistive technologies for patients with neurodegenerative conditions. Aguilar presented early clinical results, including real-time decoding of patient intent for prosthetic control and neuromodulation applications. She emphasized the transformative potential of closed-loop systems, where a patient’s neural state is interpreted and acted upon instantly by a connected device. What I appreciated was how she subtly snubbed Elon Musk’s Neuralink by highlighting her company’s second-place ranking among the world’s top BCI firms, four spots ahead of Neuralink.

“We’ve moved from millimeters to neurons. From static maps to dynamic language,” she said, calling this the dawn of intelligent neuroprosthetics.

Investor insight: Neurotechnology is transitioning from research to application. For investors in listed equities, the positbilities are rather limited, but out of general interest it will be interesting to look for startups with clinical trial momentum, AI-driven interpretation layers, and materials platforms enabling next-gen BCIs.

Gael Close (VP Innovation, Melexis) – "Robotics with a Sense of Touch"

In a live demonstration that captivated the audience, Gael Close unveiled a robotic hand outfitted with Melexis-IMEC developed tactile sensors that could grasp a cup without breaking it and spilling any water. Unlike traditional robotic grippers that use force thresholds and rigidity, this system used dynamic feedback from 3D force sensors that measure shear, pressure, and subtle texture variations. This innovation opens new frontiers in human-robot interaction, especially for collaborative robots (cobots) in logistics, healthcare, and consumer environments. For instance, Close pointed out that a robot aiding physical therapy must adapt to soft tissue and movement resistance with surgical precision. That capability hinges on integrating compact, low-latency sensing hardware directly into actuators.

He also showcased how the data from these sensors can train machine learning models to improve robotic intuition over time, another layer of value creation. Close underscored the commercial urgency: “In 5 to 10 years, if your robot can’t sense and adapt, it won’t compete.” He noted that Melexis is already working with automotive and healthcare firms to bring these technologies into production.

Investor insight: Sensor-rich robotics is moving from labs to production lines, unlocking growth in MEMS, mechatronics, and tactile computing. Key beneficiaries include Teradyne (via Universal Robots and Mobile Industrial Robots), STMicroelectronics and Bosch (MEMS sensors), Keyence (sensing and automation), and Cognex (machine vision), all well positioned to power the next wave of intelligent automation.

Some final thoughts

As the lights dimmed in the Queen Elisabeth Hall and the buzz of two intense days began to settle, one thing was clear: the future is being prototyped right here, in labs, fabs, algorithms, and packaging lines and increasingly, right here in Antwerp. IMEC’s ITF World 2025 didn’t just showcase technological breakthroughs; it reminded us that innovation thrives where people, ideas, and infrastructure intersect. From quantum and neurotech to tactile robotics and energy-aware compute, exponential technologies are no longer theoretical, they are scaling.