What a Difference 20 Years Makes: Lessons from Computex and the AI Infrastructure Boom
- 3 days ago
- 9 min read
By Oliver Campbell │ ICM HPQC Fund │ June 2026
“Written by humans, please don’t blame the robots for our typos”
This year marked 20 years since my first Computex event in Taipei, and so much had changed that the event itself was barely recognisable versus 2006. Reflecting on the past 20 years is relevant because analysing — or guessing — where compute and the tech industry will be 20 years from now is a fool's errand. A far more valuable pursuit is to understand companies bottom-up: their business models, and how those models tie into reducing cost and driving efficiency across the industry. This tends to be the best way to identify the survivors — because in this industry, as history consistently shows, the survivors tend to be the winners. Here is what has changed over 20 years:
The technology itself
International exposure and the attendee demographic
Platforms and networks, not components
Economics have shifted
Cost over TAM (total addressable market) — token prices and power as the defining chokepoints
Bubble-chatter
Key Takeaways
The AI investment fever is real — but the more durable opportunity lies in cost reduction, not headline capacity expansion (capex). We focus on technologies that bring down the unit cost of compute and therefore tokens.
Supply chain economics have structurally improved: balance sheets are stronger, cooperation has replaced cut-throat ASP competition, and foundry and supply relationships now matter more than just price.
Token cost and price inflation are a genuine near-term risk to AI adoption. Power management, photonics, interconnect fabrics, and purpose-built integrated circuitry (IC) design are the picks-and-shovels investments we find most compelling.
Several on-the-ground indicators at Computex — from sold-out keynotes to executives moving markets with offhand comments — warrant the usual caution one associates with cycle exuberance.
The Technology Itself
Bear in mind that in 2006, the iPhone had not yet been announced and Netflix was still mailing out DVDs. The largest U.S. companies by market cap were Exxon Mobil, Microsoft, Citigroup, and General Electric. Microsoft was the biggest tech company in the world, with annual revenue of $45bn and a market cap of around $235bn — versus $3.1 trillion at the time of writing. And yet, despite that extraordinary run, it is only the fourth largest company in the world, sitting behind Nvidia, Apple, and Alphabet. The biggest tech company in the world in 2006 went on to be a 13-bagger (increased by 13 times) over 20 years and still got lapped.
The contrast in Computex content between then and now was perhaps the starkest measure of how far the industry had come. Computex 2006 was centred on raw processor competition and new component launches — dubbed the "Dawn of Dual-core", as Intel and AMD went to war over clock speeds, chipsets and overclocking bragging rights. ‘Two cores’ were cutting-edge. The keynotes were delivered by chip architects, and AI, machine learning, and ARM Holdings’ chips for PCs were completely absent from the conversation.
The theme of Computex 2026 was "AI Together" and the keynote lineup read more like Davos than a hardware trade show. Jensen Huang of Nvidia, Intel CEO Lip-Bu Tan, Qualcomm’s Cristiano Amon and Marvell’s Matt Murphy all took the stage. Where 2006 had chip architects debating core counts, 2026 had tech CEOs mapping out civilisational ambition. Where the excitement in 2006 was Intel’s first dual-core processor, in 2026 Intel unveiled Nova Lake — a consumer desktop chip with up to 52 cores, built on its new 18A process.
The chip is no longer just a processor. It has become the nervous system of a new kind of intelligence. The conversation has moved from megahertz to models, from clock speeds to clusters, from nanometres to networks, and from desktops to data centres consuming the power of small cities. AI was not a feature on show in Taipei last week — it was the operating assumption behind everything.
The show floor told the infrastructure story as clearly as any keynote. Marvell’s CEO delivered a keynote titled "The Future of AI Scaling Depends on Connectivity" — a sentence that would have been incomprehensible at Computex 2006, where networking was an afterthought.
“AI infrastructure must continue scaling as a unified compute resource. As compute performance continues to advance, the next phase of AI innovation will increasingly depend on how efficiently data can move across accelerators, servers, racks, campuses and geographically distributed data centres at higher bandwidth, lower latency and lower power1”
International exposure and the attendee demographic
20 years ago Computex felt like a largely domestic affair; Taiwanese purchasing managers meeting Taiwanese suppliers and perhaps some international customers too. Last week in Taipei, there were booths from Invest in Israel, the Czech Republic and France. Even the Government of Ontario had sponsored a round to highlight Canadian startups.
The composition of the crowd had changed just as dramatically. Computex once measured its success by the value of purchase orders written on the floor. By 2026 it was an eclectic blend of global senior executives, software and hardware engineers, investors, and press. That breadth made the keynotes genuinely difficult to calibrate. When Jensen Huang argued that AI (artificial intelligence) would create more jobs, he was speaking specifically about software engineers — yet a significant portion of his audience faced the opposite implication from agentic AI. The line between celebrating an industry, promoting a company, and making supply chain partners feel valued was a fine one, and the wider world was listening more closely than ever.
The cultural shift on the show floor was equally visible. Where once booth entertainment skewed toward spectacle for its own sake, by 2026 it ran to hands-on product demos and real world applications — perhaps a reflection of an audience that had grown considerably more technical, senior and mature. The “booth babes” of 2006 have been largely consigned to the past.
Pictured below was Compal’s booth. Compal was previously an Original Design Manufacturer (ODM) for notebook PCs. Their theme in 2026 was "The Engine of Intelligence" and they organised their showcase around four main pillars: AI Infrastructure, Connectivity, Smart Healthcare, and Quantum plus Biotech AI via partnerships in academia. The era of the superstar processor or memory technology may be far from over, but the overriding messages at Computex 2026 were about partnership, networks, and shifting up the value chain.
Economics have shifted
The Taiwan tech manufacturing base that provided the supply chain underpinning the AI boom cut its teeth in very aggressive circumstances. Outside perhaps the foundries, manufacturers spent decades absorbing annual ~10% average selling price (ASP) declines on already thin margins, hoping to make up the difference on high volumes and asset turns.
In many cases their working capital was strained, meaning they were effectively financing their branded customers. At the foundry level today, it is no longer a question of price — it is a question of relationships and capacity allocation. In 2006, DRAM and TFT-LCD had daily price quotes in newspapers as the commodities they once were, today memory is a high value and critical component and the returns and cashflows have accrued accordingly (see SK Hynix below).
Cooperation has become the name of the game. Jensen Huang of Nvidia used part of his keynote to thank the supply chain partners whose work made the AI build possible. Stripping out the impact of surging demand on margins and returns was difficult, but cashflows and balance sheets across the sector are in materially better shape than 20 years ago — and, arguably, more sustainably structured too.
The real constraints: cost and power
Total addressable market (TAM) forecasts are certainly compelling, but the reality is that rising token prices risk stymying demand, and power supply remains the binding constraint. The AI boom is proving extraordinarily expensive to operate — and the next competitive frontier, arguably more consequential than raw model capability, is driving down the cost of inference. Token prices continue to rise, and if they reach levels that trigger demand destruction, that would represent a material near-term headwind to AI adoption. The countervailing pressure is that for most enterprises, adopting AI is no longer just a growth option; it is a competitive necessity.
A critical input into token prices is the energy cost. Nvidia explained to us that a one gigawatt data centre did not actually represent one gigawatt of usable capacity — it was capped at around 60% by the supply itself, to prevent spikes and guard against irregular supply. Their DSX Max LPS (dynamic power management software designed to maximise AI "token performance per megawatt" within a fixed power budget) is designed to maximise available power and squeeze more tokens out of each unit of power. One key message from Nvidia is the revenue-generating potential of tokens, alongside an acknowledgement that if prices kept rising, demand could suffer.
Reducing token costs, and compute more generally, has long been a central theme at ICM HPQC. Unit cost economics — not total addressable market projections — remain the more reliable lens through which to assess investment durability in this cycle. The ancillary technologies supporting headline capex expansion look increasingly attractive on this basis. As a result, at HPQC we continue to focus on tech that improves the efficiency and financial sustainability of compute rather than simply increase its scale.
With legitimate questions over whether corporate and government balance sheets could sustain current spending trajectories, solutions that reduce cost per token without requiring more gigawatts of capacity stand to become disproportionately valuable.
Bubble-chatter
There is a lot of talk, especially in public markets, about what is happening to corporate valuations in the compute space. It is understandable that nearer-term valuations can take a back seat to longer-term growth assumptions, but it is folly to ignore them entirely and a handful of observations from the week gave us pause:
A company developing advanced power solutions for data centres was unable to secure a booth even with 12 months’ notice and had to share floor space with a distribution partner.
Keynotes at Computex this year were either standing room only or had been fully booked months in advance. In several cases, joining an online broadcast was the only way to access.
There were frequently queues at the booths with “bouncers” monitoring footfall, especially when announcements were made. This was not common 20 years ago.
So many attendees were trading their personal stock accounts intra-day. It was very common to see, not just in the booths, but also in presentations, people with their portfolio screens open. Typically in Taiwan, red means up (auspicious) and green means down, so you would see red screens and happy faces.
Corporate executives often talked up sectors, which was understandable, but they also talked up listed companies. Over the weeks surrounding Computex, we saw some extreme share price moves across the sector— possibly driven by comments made at the event.
Conclusion
Computex has evolved remarkably over 20 years and the scale and scope of debate has shifted decisively from a focus on individual components to what role the industry will play in society. The excitement this year mirrored the broader enthusiasm around AI and the demand for compute infrastructure — and it raised real questions, both about public market valuations and the industry’s capacity to sustain capital expenditure at those levels. Within that context, however, we found a great deal to be positive about. Power management, photonics, interconnect fabrics, and purpose-built IC designs are all capable of driving token costs lower and improving returns on capex. We believe the picks-and-shovels layer of the AI stack offers investment returns at least as compelling as the headline infrastructure story — and, over a full cycle, arguably more durable too. Every lazy article about technology ends by saying how quaint these technologies and observations will look in the future, but it is always true. Roll on 2046.
Oliver Campbell
Investment Principal of the ICM HPQC Fund
MAS Licensed Representative, ICM Global Funds Pte Ltd
June 2026
Important Note:
The information in this article should not be considered an offer or solicitation to deal in the ICM HPQC Fund (Registration number T22VC0112B SF003) (the “Sub-fund”). The information is provided on a general basis for informational purposes only and is not to be relied upon as investment, legal, tax, or other advice. It does not take into account the investment objectives, financial situation, or particular needs of any specific investor. Investors should seek relevant professional advice before making any investment decision. The information presented has been obtained from sources believed to be reliable, but no representation or warranty is given or may be implied that it is accurate or complete. The Investment Manager reserves the right to amend the information contained herein at any time, without notice. Investments in the Sub-fund are subject to investment risks, including the possible loss of the principal amount invested. All forms of investments carry risks, including the risk of losing all of the invested amount. Investors should read the prospectus before deciding whether to acquire the units in ICM HPQC Fund. The value of investments and the income derived therefrom may fall or rise. Past performance is not indicative of future performance. This document is intended solely for institutional investors and accredited investors as defined under the Securities and Futures Act 2001 of Singapore. The whole or any part of this work may not be reproduced, copied or transmitted or any of its contents disclosed to third parties without ICM Global Fund’s express written consent. This advertisement or publication has not been reviewed by the Monetary Authority of Singapore.
ICM HPQC Fund a registered Sub-Fund of the ICMGF VCC (the VCC), a variable capital company incorporated in the Republic of Singapore. The assets and liabilities of ICM HPQC Fund are segregated from other Sub-Funds of the VCC, in accordance with Section 29 of the VCC Act.
Get to know us better:
