How to Buy the AI Build-Out on the ASX — semiconductors, data centres, and the vector-database gap
- Noel Ong
- 2 days ago
- 12 min read
Part 1 argued that the world is spending enormous sums to produce AI, whether or not that spending yet buys value. This part follows the money into the physical supply chain the spending pays for, the chips and the buildings, and asks a plain question: which of it can an Australian investor actually own on the ASX, and which of it, including the much-hyped vector databases, is simply not listed here at all.
1.00 — THE PICKS-AND-SHOVELS IDEA, AND ITS LIMIT
Selling shovels in a token rush
There is an old rule about gold rushes: the steadier money was often made selling picks, shovels and denim to the miners rather than digging for the metal. The equivalent question for artificial intelligence is which businesses get paid no matter which AI application eventually wins.
Part 1 made the case that a company's AI spending should be judged on cost per outcome, the cost of a finished, useful result, rather than on how much it spends or how loudly it talks about AI. That test is deliberately hard on the buyers of AI. The suppliers are a different proposition. Whether or not a given bank or retailer ever proves its chatbot was worth it, the tokens (the small chunks of text an AI reads and writes, and the unit it is billed by) still had to be produced somewhere, on physical chips, inside physical buildings, drawing physical power. FIG. 01 lays out that supply chain as a stack, and marks the plain truth of it: an ASX investor can buy the bottom of the stack (the buildings and, speculatively, the chips) but not the fashionable middle (the models and the vector databases).
Now the limit, stated before any company is named. Australia is a small part of this supply chain. It has no large-scale chip fabrication, so most of its listed semiconductor names are early-stage technology-licensing hopefuls rather than profitable manufacturers. What follows maps what is here, what it does, and what each name's main risk is. It is not a list of recommendations, and the counterweights are as important as the descriptions.
2.00 — THE CHIP LAYER: SEMICONDUCTORS ON THE ASX
Small, speculative, and mostly selling ideas rather than chips
A semiconductor is the material, usually silicon, from which computer chips are made, and "the semiconductor industry" is the chain that turns it into working processors and memory. That chain has several rungs, and the handful of ASX names sit on different ones. FIG. 02 places them.
The names differ as much in maturity as in what they do. FIG. 03 maps them on two axes an investor cares about: how far each is from real, repeatable revenue, and how speculative it remains. Most cluster at the pre-revenue, high-risk corner. Only DUG Technology and, to a lesser extent, Revasum sit near the "actual revenue" end.
Before the profiles, PLATE 3 is a reminder of what this whole layer eventually produces: physical chips.
Design and intellectual property BRN · AXE
BrainChip Holdings (ASX: BRN) designs a low-power processor called Akida for "edge" AI, meaning AI that runs on a device such as a drone, sensor or piece of defence equipment rather than in a distant data centre. It licenses the design to others. For the year to December 2025, revenue was about US$1.89 million (up from about US$0.4 million a year earlier) against a net loss of about US$20 million, and a next-generation processor (the AKD2500) is expected in prototype during 2026.
Risk Revenue is minimal, the licensing model is unproven at commercial scale, and the company has burned cash for years. The share price reflects hope for the future, not current earnings.
PLATE 1 shows the Akida design BrainChip licenses to others, with its on-chip processing blocks and "neuron fabric".
Archer Materials (ASX: AXE) is a research-stage developer of a quantum computing chip and a separate biochip. Quantum computing is an experimental approach that could one day solve certain problems far faster than today's machines. Archer reported a net loss of about A$2.99 million for the half and generates almost no revenue.
Risk Highly speculative, pre-revenue, and any commercialisation is years away and uncertain.
Memory WBT · 4DS
Both names are chasing the same prize shown in PLATE 2: a memory that is as fast as today's working memory but keeps its data without power.
Weebit Nano (ASX: WBT) has developed a memory technology called ReRAM (resistive memory) that aims to be faster and more power-efficient than the Flash memory in today's devices, and licenses it to chipmakers. It reported a record half-year revenue of about A$5.6 million, referenced a licensing deal with Texas Instruments, held about A$82.8 million in cash, and guided to more than A$10 million of revenue in the 2026 financial year. Its market value is above A$1 billion.
Risk The valuation rests on future royalties that are still ramping, not on current profits, so it is priced for success that has not yet arrived.
4DS Memory (ASX: 4DS) is developing its own version of resistive memory aimed at high-density storage for mobile devices and data centres, which makes it the closest chip-level name to the "data storage for AI" theme. Revenue is negligible at about A$316,000, with roughly A$3.34 million in cash plus about A$5.45 million in deposits.
Risk Very early stage with almost no revenue; it needs a commercial manufacturing partner before it counts for much.
Materials BLG · SLX
BluGlass (ASX: BLG) makes laser components from gallium nitride, a semiconductor material used in lasers and power electronics, using its own manufacturing process. It reported revenue of about A$5.76 million (up about 14 per cent), a net loss of about A$4.10 million, and referenced a US defence contract and a project pipeline.
Risk Loss-making and facing the challenge of scaling a specialised process to volume.
SILEX Systems (ASX: SLX) belongs here only with a clear caveat. Its main business is laser uranium enrichment technology, not chips. Its semiconductor-relevant activity is a "Quantum Silicon" plant producing highly purified silicon-28, a material used to build quantum computing chips. That is a genuine but secondary part of the company.
Risk This is primarily an enrichment-technology company; anyone buying it for semiconductor exposure is really buying the enrichment story with a small chip option attached.
Equipment and high-performance computing RVS · DUG
Revasum (ASX: RVS) is the most literal "shovels" name on this list. It makes the grinding, polishing and planarising machines used to manufacture semiconductor wafers, including silicon carbide, and so sells tools to the industry rather than betting on any single chip winning.
Risk A small supplier into a cyclical capital-equipment market; demand rises and falls with chipmakers' own spending. Current financials should be checked directly.
DUG Technology (ASX: DUG) is different again: an established provider of high-performance computing (very large-scale number crunching), cloud services, geoscience work and software. Unlike the chip hopefuls it earns real revenue, and reported strong growth with its high-performance computing demand rising and cash flow turning positive.
Risk Small in a field where the giants of cloud computing set the price, so it must keep finding niches they do not serve.
3.00 — DATA STORAGE FOR AI: THE DATA CENTRE, NOT THE DATABASE
The vector database everyone talks about is not listed here
Ask where AI keeps its data and the fashionable answer is the vector database. The realistic answer for an ASX investor is that you cannot buy one here, so the real local exposure is the building the data sits in.
First, what a vector database is, in plain terms. Traditional databases find things by exact match, the way a phone book finds a name you spell correctly. A vector database stores information as strings of numbers that capture meaning, so it can find things that are similar in sense rather than identical in spelling. It is the technology behind most "chat with your documents" features, where an AI answers using a company's own material. It is central to how businesses are putting AI to work.
It is also not something you can own on the ASX. The specialist vector database companies are privately held, and the listed ways to touch the technology are large overseas firms that offer it as one feature among many. One of the more prominent independents, Marqo, has Australian roots but is private (a point worth confirming before relying on it). FIG. 04 sets out where the category actually trades.
So the ASX way to own "storage for AI" is the physical layer: the data centre. A data centre is a large, secure, heavily powered building full of computers that store data and run software for other companies. Its size is measured in megawatts (MW) of power, because power, not floor space, is what limits how much AI computing a site can run. FIG. 05 explains how a data centre turns power into profit, and where the cost-per-outcome test from Part 1 bites.
Australia has a genuine, investable cluster of listed operators. FIG. 06 compares the capacity figures the main names have disclosed, with an important caveat in the caption about how those figures are defined.
The listed data-centre names NXT · MAQ · DGT · GMG · IFT
NextDC (ASX: NXT) is described as the purest data centre business on the ASX: it builds and runs the buildings that house AI computing and storage. It referenced 667 megawatts of contracted capacity (as at April 2026, up from 416.6 megawatts in December 2025), a A$1.5 billion equity entitlement offer forming part of a A$2.2 billion capital plan announced in April 2026, and a partnership with OpenAI to build sovereign AI infrastructure in Sydney (data centres kept on Australian soil under Australian control).
Risk It spends enormous sums upfront; returns depend on filling that capacity with paying tenants at prices above its cost of capital.
PLATE 4 is an artist's impression of NextDC's S4 Sydney campus, one of the hyperscale sites behind its contracted capacity.
Macquarie Technology (ASX: MAQ) runs data centres alongside cloud and telecommunications businesses, and hosts a large share of Australian Federal Government computing, which is a demanding, sticky customer. A new 47 megawatt facility was referenced for opening in September 2026, with more in the pipeline. Its market value is around A$1.7 billion.
Risk Smaller than NextDC, so expansion carries more execution and funding risk relative to its size.
DigiCo Infrastructure REIT (ASX: DGT) is a data centre real estate investment trust (a listed property trust). In May 2026 it announced the sale of its Chicago data centre for about US$750 million to strengthen its balance sheet and fund growth, including an approved 88 megawatt expansion of its SYD1 Sydney campus. It has traded below its A$5.00 December 2024 listing price.
Risk Recently listed and already below its issue price; as a property trust it is sensitive to interest rates.
Goodman Group (ASX: GMG) is a large industrial property group that has pivoted hard towards data centres, referencing a development pipeline of about A$14.4 billion that is roughly 73 per cent weighted to them, and multi-gigawatt power capacity. It is the most diversified, and therefore lower-risk, way to touch the theme.
Risk Data centres are only a slice of a business still exposed to the broader property cycle, so the AI exposure is diluted.
PLATE 5 shows one of Goodman's data centre developments, the kind of asset its pipeline is now weighted towards.
Infratil (ASX: IFT) owns about half of CDC Data Centres, a major operator that referenced a 555 megawatt contract with a large cloud customer, taking its contracted capacity above one gigawatt. Infratil is a New Zealand-based, dual-listed infrastructure investor.
Risk A diversified infrastructure owner, so the AI exposure is real but indirect and shared with airports, energy and other assets.
The connective tissue MP1
Megaport (ASX: MP1) is not storage, but it is worth naming as the plumbing between clouds and data centres. It sells "network-as-a-service", on-demand connections that let businesses link up cloud and data-centre services, and has been positioning towards AI workloads.
Risk It provides connectivity rather than storage or compute, and competes in a fast-moving market; include it as an adjacency, not a core storage play.
4.00 — HOW THE TWO HALVES CONNECT
The cost-per-outcome test applies to the shovels too
It would be easy to read Part 1 and Part 2 as opposites: one urging caution on AI spending, the other pointing at ways to buy into it. They are the same discipline applied twice.
The test from Part 1 was cost per outcome, not inputs. Turn it on the picks-and-shovels names and it still bites. A data centre operator is worth it when it signs paying tenants at a return above its cost of capital, not when it announces a headline number of gigawatts. A chip hopeful is worth it when it converts its intellectual property into recurring royalties, not when it runs an impressive demonstration. A high-performance computing provider is worth it when it wins work the cloud giants cannot serve more cheaply. In each case the announcement is the bill; the signed contract or the royalty cheque is the value.
There is also a real tension between the two parts that a clear-eyed investor should hold. Part 1's cost discipline, if it takes hold, would slow the growth in tokens per task, which is part of what drives demand for all these shovels. The counter is that adoption has been growing faster than efficiency: Google's token volumes rose roughly seven-fold in a year even as the cost per token collapsed. That pattern, more use swamping cheaper units, is the bull case for the infrastructure. It is a reasonable bet, not a law, and Part 1's bearbox applies here too.
References & sources
Company activities and figures below are drawn from the secondary sources listed and are as at the companies' most recently reported results referenced there, broadly the first half of the 2026 financial year. Every company figure is market-sensitive and should be confirmed against each company's own ASX announcements and refreshed on publication day. All figures shown in the visuals are original Samso illustrations of the data named in each caption. Figures in this note were checked against the companies’ own ASX presentations and announcements where available (to July 2026); the NextDC OpenAI partnership, DigiCo’s US$750 million Chicago sale, Weebit’s Texas Instruments licence, Goodman’s pipeline, Macquarie’s 47MW facility and Infratil’s 555MW CDC contract were all confirmed against primary sources. This note names companies to illustrate where exposure exists; it is not a recommendation on any of them. Photographs and company diagrams labelled PLATE are reproduced from the companies' public ASX release presentations, with attribution in each caption, and are distinct from the original Samso illustrations labelled FIG.
Veye — "Top 5 ASX semiconductor stocks 2026": Weebit Nano (WBT), Archer Materials (AXE), BluGlass (BLG), 4DS Memory (4DS), BrainChip (BRN) descriptions and half-year figures.
Stocks Down Under — "Marvell... 3 ASX AI chip stocks WBT AXE BRN" and related coverage: Weebit Nano, Archer Materials and BrainChip technology and status; BrainChip market value and AKD-series roadmap.
Stockhead — "Revasum leads a pack of ASX semiconductor stocks...": Revasum (RVS) grinding, polishing and CMP wafer equipment; SILEX (SLX), BrainChip, Weebit Nano, 4DS and BluGlass activity summaries.
Silex Systems (SLX) company materials and Stocks Down Under coverage: primary business in laser uranium enrichment technology; "Quantum Silicon" (Q-Si) plant producing enriched silicon-28 for quantum computing.
DUG Technology (DUG) FY26 half-year results release and Stocks Down Under coverage: high-performance computing, cloud and geoscience services; revenue growth and cash flow.
Stocks Down Under — "The ASX AI infrastructure boom... data centre stocks": NextDC (NXT) 667 MW contracted, A$1.5 bn raise and OpenAI partnership; Macquarie Technology (MAQ) government hosting and 47 MW facility; DigiCo (DGT) Chicago sale and Sydney project; Goodman Group (GMG) pipeline and power capacity; Infratil (IFT) / CDC 555 MW contract and 1 GW-plus contracted capacity.
Kalkine and Megaport (MP1) investor materials: network-as-a-service cloud connectivity and AI-workload positioning.
AI Magazine — "Top 10 vector databases for AI"; MarketsandMarkets vector database market overview: specialist vector database vendors (Pinecone, Weaviate, Qdrant, Milvus/Zilliz, Marqo) and listed providers offering vector search as a feature (MongoDB, Microsoft, Amazon); no ASX-listed pure-play identified.
Reproduced company visuals (PLATES): BrainChip Holdings 2026 AGM presentation (6 May 2026); 4DS Memory Capital Raising Presentation (January 2025); NextDC Investor Presentation (20 April 2026); Goodman Group 1H FY26 Results Presentation (19 February 2026). Each reproduced with attribution in its caption.
The Samso Way – Seek the Research
Here at Samso, we pride ourselves on delivering content for investors that is independent and informed by over three decades of experience in the industry. Our content is well-researched and is only created if I see merit in discussing the company's story.
Our mission is simple: cut through the noise and spotlight what matters—genuine stories, grounded insights, and real opportunity.
Our content is well-researched and is only created if the team sees merit in discussing the company or concept. Investors can explore our three core platforms:
There may be numerous paths to success in investing, but the common thread among successful individuals is that they remain committed to making informed decisions. Equip yourself with the right knowledge and tools, and you will be well on your way to achieving your financial goals.
Most importantly, investors need to be absolutely diligent in understanding their own risk-reward tolerance and capabilities. Never bite off more than you can chew. As they say, Rome wasn't built in a day, and the Great Wall stood because it took centuries to complete.
The Samso Philosophy:
Stay curious. Stay sharp. And remember—digging deeper always uncovers the real value.
In Life, there is no such thing as a Free Lunch.
Never bite off more than you can chew is my parting comment.
Happy Investing, and the only four-letter word you need to know is DYOR.
To support our independent work, please head over to our Support Page and give us a helping hand in any of the ways listed. This is a new initiative for the Samso Platform, and it was always the concept of Samso when we started this journey in 2018.
Disclaimer
The information or opinions provided herein do not constitute investment advice, an offer, or solicitation to subscribe for, purchase, or sell the investment product(s) mentioned herein. It does not take into consideration, nor have any regard to your specific investment objectives, financial situation, risk profile, tax position and particular, or unique needs and constraints.
Samso is a trusted platform that equips dedicated investors with up-to-date industry knowledge and insights from top CEOs and thought leaders. By staying informed on business advancements and market trends, investors can enhance their financial decisions through a combination of expert guidance and their own research.
Samso Insights | www.samso.com.au | An Investor Lens on ASX-Listed Companies
Comments