ChatGPT and Claude will be the face of AI, however behind the screens lies a fancy ecosystem of chips, information, vitality and people.
Throughout the Asia Pacific, management over these layers is reshaping energy balances, binding allies and rivals into intricate types of dependence.
That is a synthetic intelligence (AI) chip.*
Chips like these energy the gadgets and platforms we use on daily basis: from ChatGPT and Google suggestions, to parking help and smartphone cameras.
Probably the most superior ones comprise tons of of billions of transistors: microscopic switches performing trillions of calculations a second, making them a number of the most complicated objects ever made.
With elements just a few atoms thick, that means “a speck of mud” can damage tons of of hundreds of {dollars}’ value of product, solely a handful of corporations can produce chips like this.
Their sudden significance has remodeled area of interest expertise manufacturing companies into issues of international coverage and nationwide safety.
However chips are only one piece of a bigger machine, with out which, they’re successfully ineffective.
AI is usually portrayed by Silicon Valley and hovering valuations, however these are simply the floor layers of what insiders and business leaders like NVIDIA CEO Jensen Huang have popularly described as “AI’s layered cake”: a world ecosystem of chips, information, vitality, fashions and functions (or some variation of this).
Every layer relies on the others: fashions want information and compute energy; computing requires chips; chips function in information centres; and all of it requires vitality.
However right here’s the twist: no nation controls each layer.
In reality, important capabilities are so globally dispersed but concentrated in just a few locations, that geopolitical allies and rivals alike are sometimes pressured to rely on each other: a dynamic some say has prevented anyone nation from dominating AI, whereas additionally elevating the price of battle.
Loading…
“The person interface — ChatGPT, Claude — is absolutely simply the tip of the iceberg,” Robyn Klingler-Vidra, an AI political financial system knowledgeable who authored a brand new report on key Asian gamers, says.
“There are such a lot of enormously helpful layers of the tech stack, unfold throughout ‘area of interest superpowers’ within the Asia Pacific, and firms taking part in with them, that AI underscores one thing folks could not intuitively see: why governments are so involved about deep commerce and demanding minerals.“
As a part of a wider year-long ABC mission, we spoke to folks throughout the AI panorama: from on a regular basis customers to previous and current representatives of main international corporations and rising gamers — SK Hynix, TSMC, OpenAI, Google, 01.AI and Microsoft — with many others agreeing to talk on background.
Pieced collectively, their insights inform a narrative way more nuanced than chatbots and inventory costs: a tangled system of energy and dependence formed by assets, diplomacy, and noteworthy accidents — and with penalties for Australia.
Beneath the AI hood: Laying it on skinny
Loading…
To know the AI race, we first want to grasp how the bodily system works and what nations are competing over.
There isn’t a universally agreed model of the so-called AI “stack” — some name it a sequence, cake, or ecosystem — however for simplicity, consider it as chips, information, vitality and software program, with the primary three being largely bodily.
On the coronary heart of the bodily stack are chips, whose significance is troublesome to overstate: once you immediate Google or ask AI a query, it’s chips doing the computing behind the scenes.
“Put merely, with out exponential advances in AI chips, we might not have right this moment’s AI capabilities,” Chris Miller, creator of the 2022 guide Chip Battle, says.
Whereas US firm NVIDIA designs most of the world’s main AI chips, it doesn’t manufacture them.
That work falls largely to corporations such because the Taiwan Semiconductor Manufacturing Firm (TSMC), which makes the vast majority of the world’s most superior AI processors. However even TSMC can’t function alone, and depends closely on specialised expertise and supplies from different nations to fabricate them.
“Superior chip manufacturing just isn’t one factor: Taiwan could make the world’s most superior AI chips, but it surely will depend on lithography machines from the Netherlands, ultra-pure supplies and chemical compounds from Japan, reminiscence elements from South Korea,” Gregory C Allen, AI coverage knowledgeable and founding father of Resolution Tree, says.
These corporations have spent a long time, out of the limelight, turning into terribly specialised in several, however typically complementary, elements of the chain: relationships that require monumental belief.
“We don’t compete with our clients,” a TSMC spokesperson says. “That belief permits us to work intently with the world’s main innovators, aligning our manufacturing roadmaps with their product designs years earlier than manufacturing begins.“
That division of labour has allowed a swathe of corporations to turn into extremely specialised, whereas additionally concentrating capabilities in a handful of locations.
As AI’s capabilities grew extra highly effective, these provide chains turned more and more political, with governments proscribing entry to chips and the expertise wanted to make them: because the US famously did in 2022 when it started proscribing superior chip exports to China, simply weeks forward of ChatGPT’s public launch.
“In a parallel universe the place the US by no means imposed these export controls, China would nearly actually already be the chief,” Allen says.
These restrictions have since developed, whereas new ones have unfold up and down the provision chain — from Japan, to the Netherlands, to China — profoundly shaping AI’s growth.
In different phrases, the geography of right this moment’s AI race is not merely the product of technological competitors. It has, partly, been intentionally engineered: a key level we’ll return to shortly.
However chips are just one a part of the AI stack, that means even when you safe entry to them, they really do little or no on their very own.
The facility of superior processors is simply totally realised when hundreds are mixed with high-speed reminiscence and networking, assembled into servers and racks, and related collectively inside huge information centres, the place the calculations behind our search queries, Netflix suggestions, and map instructions really happen.
Knowledge centres have existed for many years, however the generative-AI increase has quickly elevated their scale, visibility and strategic significance.
“I think about information centre builders have been formidable, however demand has most likely outstripped even what they anticipated,” says Klingler-Vidra.
“It is believable now that information centres have been being deliberate with out anybody totally imagining simply how data-hungry AI would turn into.“
These amenities are successfully large computer systems: warehouses full of processors, networking gear and cooling techniques, working collectively to coach and run more and more highly effective AI fashions.
Just like chips, nonetheless, operating information techniques at scale creates one other bottleneck and dependency: vitality. Each process takes a sip of energy, so information centres require monumental quantities of electrical energy, together with grid connections, land, cooling and, typically, water.
Wherever information centres are constructed, their urge for food for energy is reshaping vitality techniques round them, a strain nations together with Singapore and the US have been already confronting earlier than the AI increase.
“China is much forward within the scale of its electrical system and accessible energy capability, and computation requires monumental quantities of electrical energy,” says Ya-Qin Zhang, chair of AI research at Tsinghua College and former chairman of Microsoft China, noting that China generates greater than twice as a lot electrical energy because the US and over a 3rd of the world’s provide.
In Australia, comparable considerations are solely beginning to get louder. As data-centre growth surges, specialists warn the sector may require as a lot electrical energy as presently required to energy tens of millions of households inside a decade.
Nonetheless, solely as soon as these techniques are in place — chips, information centres and vitality — will we attain the layers most individuals recognise as AI: fashions, the skilled techniques that generate predictions and responses, like OpenAI’s GPT, and functions, the software program that folks use, like ChatGPT.
Firms akin to OpenAI and Anthropic function largely at this seen layer, whereas relying on the broader worldwide system.
“Companions and suppliers are central to AI progress as a result of compute provide just isn’t solely scarce, however have to be secured and managed years prematurely,” a spokesperson for OpenAI informed the ABC, noting that the corporate beforehand relied extra closely on particular person compute suppliers.
“We now work throughout a diversified ecosystem, which improves resilience and creates compute certainty, [which] instantly impacts whether or not AI can scale.”
Zooming out, highly effective AI relies upon as a lot on specialised chemical compounds and vitality as on laptop scientists. The place these dependencies turn into troublesome to switch, expertise intersects with financial leverage and geopolitical technique — a actuality mirrored in paperwork just like the 2025 US AI Motion Plan.
A redrawn Asia Pacific: Divided by politics, united by necessity
On uncooked numbers alone, the AI stability of energy seems pretty simple.
The US dominates in information centres, non-public funding and main fashions, whereas China leads in electrical energy technology, robotics, analysis output and manufacturing capability.
At face worth, that implies a well-known two-power contest; however the focus of important capabilities throughout the stack tells a distinct story.
We have seen that whereas the US may dominate by way of software program or chip design, it is Taiwan that dominates advanced-chip manufacturing.
Taiwan’s function, nonetheless, relies upon closely on specialised gear, supplies and chemical compounds from the Netherlands, Japan and South Korea.
So whereas the US may lead in design, a scarcity of Japanese chemical compounds or a important Dutch lithography element can grind the system to a halt.
This is not idea: it is occurred earlier than, and continues to occur.
In 2019, a stoush between Japan and South Korea choked provides of chipmaking supplies; in 2022, Russia’s invasion of Ukraine disrupted half the world’s neon provide, sending costs hovering.
Nearer to residence, Singapore and Australia could search to turn into regional information hubs, however these ambitions rely on software program, chips and large quantities of vitality.
In 2019, Singapore needed to pause new information centre growth for a number of years as demand strained restricted energy, land and water assets.
Improvement resumed in 2022 beneath tighter effectivity guidelines, the identical yr Singapore started importing electrical energy from Laos by way of Thailand and Malaysia.
AI energy subsequently does not merely belong to whoever has essentially the most cash or information centres: it may well sit inside an obscure chemical or commerce settlement that no one notices till it disappears.
“What occurs to the world if Taiwan magically disappears? Financial apocalypse. What occurs to the semiconductor world if the Netherlands magically disappears? Much less sudden, however nonetheless a disaster. What occurs when you lose the US? Identical factor. What occurs when you lose Korea? Identical factor,” Allen explains.
“There are a number of important nodes within the provide chain, very like there are a number of important organs in your physique: you want all of them.”
These dependencies also can lower throughout alliances. For instance, South Korea, a US ally, provides important semiconductor expertise whereas remaining deeply tied to China, whilst Washington restricts China’s entry to a number of the most superior applied sciences.
This geography displays a mixture of deliberate coverage, business choices, political necessity and historic accident.
For instance, observers argue that the 2022 export controls supposed to restrict Chinese language AI growth have additionally inspired higher effectivity and funding in home options, an impact NVIDIA CEO Jensen Huang has described because the controls “backfiring”.
“It is a double-edged sword,” Kai-Fu Lee, founding father of Chinese language AI agency 01.AI and former head of Google China, says. “When you do not get [top-end NVIDIA chips], you must determine methods to make issues quicker, you are pressured to turn into extra environment friendly.“
Dynamics like these — current throughout the provision chain, from uncommon earth refining and specialised expertise to subsea cables and proprietary information — complicate the thought of AI as a easy “race”.
“China refines greater than 90 per cent of the world’s uncommon earths: that represents an enormous supply of energy in a system predicated on their availability,” Klingler-Vidra says.
“No nation controls each layer, so being extraordinarily troublesome to switch in a single layer could matter greater than attempting to dominate all the factor.”
For smaller economies, that creates alternatives.
“As a substitute of attempting to do every part and ending up mediocre at every part, discover the stuff you could be world-class at and use them to get every part else you want,” Allen says.
“What Australia wants is strategic relevance, and ideally, strategic indispensability, which it may well obtain not by constructing every part itself, however by controlling one thing the broader system wants.“
However strategic significance cuts each methods: turning into indispensable to the stack, additionally means inheriting the dangers and prices.
Alternatives, transparency, and the worth of getting it mistaken
()
Almost 4 years since ChatGPT’s launch, public discourse round AI stays dominated by frontier corporations — from breakthrough capabilities to warnings about existential hazard — with far much less consideration paid to the techniques beneath them.
Observers level out that whereas a number of the opacity is intentional, a lot of it’s a consequence of how the business developed: corporations could have authorized causes to keep away from showing dominant, whereas governments could worry financial retaliation from buying and selling companions for talking bluntly about AI.
“The self-censorship is an actual problem,” Allen says.
“Politically, economically, it is comprehensible; but when the general public do not hear the plain fact, how are they going to know who to vote for or which insurance policies to assist?”
In July, Australian Prime Minister Anthony Albanese declared AI a nationwide precedence. However some have been fast to level out why turning that ambition into coverage is far tougher than merely attracting funding or constructing extra infrastructure.
No nation controls every part it wants. For Australia, the query is subsequently not the right way to turn into technologically self-sufficient, however the place to construct energy, and which dependencies it’s prepared to simply accept.
One present AI technique is to seize home worth by the roles, abilities and infrastructure constructed to assist international AI fashions, like information centres.
However some specialists warn investments of this kind danger turning Australia into an “infrastructure enclave” that consumes home land and electrical energy, with out guaranteeing the much-needed mental property or financial leverage.
They keep that Australia already has benefits in vitality, important minerals, and industries like mining and agriculture: areas the place it may apply AI to capitalise on present strengths.
“Australia has a variety of information centres, however many are foreign-owned or funded by outdoors funding, which could make us robust in that layer, however will not essentially outcome within the worth or leverage that different areas may present,” Vikas Kumar, who authored a latest AI chain competitiveness framework, says.
“The chance at hand is to mix AI with sectors the place Australia already has deep, in-demand information, to create exportable services.
“As an illustration, if I will Korea, constructing that relationship can be my primary motivation: securing key elements, whereas providing functions or vitality they want in return.“
This strategy makes AI coverage as a lot about diplomacy as expertise and personal funding: commerce coverage, alliances, immigration and vitality choices can shift technological leverage nearly as a lot as breakthroughs in a laboratory.
A number of Chinese language and American AI corporations informed the ABC on background they have been already working in the direction of new partnerships with main Australian industries, together with mining; OpenAI additionally pointed to its December 2025 initiative with CommBank, Coles and Wesfarmers.
By the identical token, observers have been cautious to level out that interdependence throughout the stack doesn’t robotically equate to safety, as Europe learnt from its reliance on Russian vitality.
“It seems it wasn’t low-cost, we simply weren’t counting accurately,” Allen says, pointing to comparable dependencies like Chinese language uncommon earths, which can be costlier than assumed as soon as “strategic vulnerability” is factored in.
()
Affect over important elements of the AI system subsequently issues not solely to these constructing highly effective AI fashions and functions, but additionally to these with the leverage to constrain them.
For now, as main powers guess that management in AI will instantly translate into wider financial and geopolitical energy, the speed and scale of funding throughout the chain — from vitality to chip design to frontier software program — continues to be extraordinary, reflecting the sheer measurement of that guess.
“Within the US, we’re spending extra on AI each 10 months than all the Apollo Moon program price from begin to end over 13 years: it’s an astonishing outlay of capital right into a technological query mark that would doubtlessly backfire,” Allen says.
“The Industrial Revolution reshaped the worldwide stability of energy and nearly each facet of human life: you may actually see its affect from area. My prediction is that, with the extent of change occurring throughout all the AI stack, that is the type of second we’re dwelling by now, therefore why it feels so ominous.”
For Australia, like a lot of the remainder of the world, there isn’t any single technological race to win. There’s solely an online of selections to barter: what to construct, what to commerce, whom to belief, the place to turn into indispensable, and the place dependence turns into harmful.
Learn the story in Chinese language (中文版) or Indonesian.
Credit
- Reporting & Producing: Steven Viney
- Viewers Engagement & Extra Reporting: Doug Dingwall
- Graphics/Design & Extra Reporting: Jarrod Fankhauser
- Extra Media: TSMC, SKHYNIX, NVIDIA, Getty, Reuters
- Translations: Natasya Salim & Michael Li
*Editor’s Word: Some visuals and data have been simplified or tailored for readability. The place data was unavailable or requests not met, we relied on knowledgeable evaluation and cheap approximations. The rotating chip visible is an amalgamation of a number of main AI chip schematics, moderately than a illustration of any single mannequin.