AI Special Topic Workshops · A One-Hour Session

Bubble Trouble:
Making Sense of the AI Economy

Everyone is asking whether AI is a bubble. This hour explains why that question has no tidy answer: economics, geopolitics, information warfare and military competition are all tangled together.

Figures in this deck are from 2024 to 2025 reporting; markets move fast, so treat every number as a snapshot, not a forecast.

ntworld.inkWorkshops01 / 18

The question everyone asks about the AI economy

"Is AI a bubble?" can't be answered by stock prices alone

The instinct

Treat it like any market story

Watch the share prices, compare the price-to-earnings ratios, wait for the correction. That works for ordinary manias; tulips, railway shares, crypto tokens.

The reality

Four stories tangled into one

Governments frame AI as national security, which keeps money flowing regardless of commercial returns. The same tools reshape the information environment that investors rely on. Each dimension feeds the others.

The frame for this hour: the AI bubble is what policy researchers call a wicked problem; a challenge with many stakeholders, conflicting incentives and no clean answer. The goal today isn't a verdict; it's a map.

ntworld.inkWorkshops02 / 18

First, grasp the scale

A buildout with no historical precedent

~$400B

Projected AI infrastructure spending in 2025 by the big technology companies; data centres, chips and power.

~80%

Share of American stock market gains accounted for by AI-related enterprises in late 2025.

10 mths

Tech companies are collectively funding the equivalent of a new Apollo program every ten months; by nominal dollars, more than any group of firms has ever spent on a single buildout.

And the concentration: the top five companies made up around 30% of the entire S&P 500, AI capital spending reached about 1.1% of US GDP in mid 2025, and the IMF warned of global economic impacts if valuations correct sharply. When something this big wobbles, everyone feels it.

ntworld.inkWorkshops03 / 18

Now look closer at the accounts

Record spending, record losses

$13B

OpenAI's projected revenue for 2025; extraordinary growth for a company that barely sold anything three years earlier.

$13.5B

OpenAI's net loss in the first half of 2025 alone; more than a year's projected revenue, gone in six months.

498

AI "unicorns"; start-ups valued above one billion dollars, most with little or no profit to show yet.

The bull case and the bear case use the same numbers: believers see an Amazon-style land grab where losses now buy dominance later; sceptics see valuations priced for a future that may never arrive. Both can point at this slide.

ntworld.inkWorkshops04 / 18

The pattern that worries the analysts

The money going in circles: hyperscalers buying from each other

·Microsoft invests in OpenAI, and OpenAI pays Microsoft for the cloud computing it runs on; a significant slice of "AI revenue" is the same dollars changing hands.

·Google, Amazon and Microsoft all buy NVIDIA chips while building competing AI products; the chip maker's boom is everyone else's capital expense.

·This circular flow makes it hard to see genuine external demand; how much of the AI economy is customers, and how much is the industry buying from itself?

Goldman Sachs Research, 2025

"The AI revenue buildup is currently still too concentrated, with hyperscalers cannibalising their own demand for AI. Unless we see more evidence of revenue diversification, AI revenue buildout and therefore valuation resilience may come under pressure."

ntworld.inkWorkshops05 / 18

Someone has drawn it, and it is worth a minute of looking

Every arrow is a real, announced deal

Bloomberg chart titled How Nvidia and OpenAI Fuel the AI Money Machine, showing circles for Nvidia, Microsoft, OpenAI, Oracle, AMD, CoreWeave, Intel, xAI and others, sized by market value, joined by coloured arrows for hardware, investment, services and venture capital

1

Circles are market value. Nvidia at $4.5 trillion, Microsoft at $3.9 trillion, OpenAI at $500 billion. The arrows are what they have agreed to pay or invest in each other.

2

Follow one loop. Nvidia agrees to invest up to $100 billion in OpenAI; OpenAI signs a $300 billion cloud deal with Oracle; Oracle spends tens of billions on Nvidia chips. The money comes back around.

3

None of this is illegal or even unusual on its own. The question a picture like this raises is the one from the last slide: how much of the demand is coming from outside the circle?

Chart: Bloomberg News, from "OpenAI's Nvidia, AMD Deals Boost $1 Trillion AI Boom With Circular Deals", 7 October 2025, bloomberg.com. The graphic itself carries no date, only the credit "Source: Bloomberg News reporting"; the figures are those reported at the time of publication and have moved since. Reproduced here for teaching and comment.

ntworld.inkWorkshops06 / 18

The idea that organises this whole hour

What makes a problem "wicked"?

A "wicked problem" in policy terms is a challenge that resists simple solutions. The AI bubble ticks every box:

·No clear problem definition; is the problem overinvestment, underinvestment, or investment in the wrong things? Depends who you ask.

·Many stakeholders with conflicting values; a venture capitalist, a Pentagon strategist, a social media platform and a Chinese planner all want different truths.

·Incomplete information; the key numbers (real demand, real capability, real returns) are exactly the ones nobody can verify.

·Interconnected causes and effects; pull one thread (say, export controls) and three other dimensions move. And the stakes for getting it wrong are high.

ntworld.inkWorkshops07 / 18

The map for the rest of the session

Four dimensions, deeply intertwined

Economic

The money

Investment flows, market valuations, circular financing, and the awkward question of when returns arrive.

Geopolitical

The great powers

US-China competition, chip export controls, supply chain choke points and strategic autonomy.

Information

The stories

Misinformation amplification, algorithmic curation, and narrative warfare about what AI can actually do.

Military

The arms race

Autonomous systems, intelligence analysis, and deterrence calculations that don't care about quarterly earnings.

Why four lenses: each dimension creates its own incentives to inflate or deflate AI claims. Consensus on "the truth about AI" is hard because the players are being rewarded for different answers.

ntworld.inkWorkshops08 / 18

The geopolitical dimension

A supply chain with three choke points

Taiwan

TSMC

Manufactures roughly 90% of the world's most advanced chips. There is no quick substitute anywhere on earth.

United States

NVIDIA

Dominates the AI accelerator market; the US uses export controls on these chips to slow China's AI development.

Netherlands

ASML

The sole supplier of the extreme ultraviolet lithography machines needed to make advanced chips at all.

·Critical minerals: China processes 60 to 90% of many minerals AI hardware depends on; Australia's rare earth deposits are both an opportunity and a strategic exposure.

·The DeepSeek complication: China's DeepSeek showed serious AI can be built despite export restrictions, and at far lower cost than western estimates; awkward for the controls narrative and for high-cost western business models at once.

ntworld.inkWorkshops09 / 18

The information ecosystem dimension

The same technology is rewiring what we believe

Researchers describe three interconnected actors in misinformation spread; AI turbocharges all of them, plus the tools meant to catch it:

Publishers

Making it is now cheap

AI slashes the cost of producing false content: fake articles, deepfake video, synthetic audio.

People

Telling is now hard

Individuals increasingly cannot distinguish AI-generated content from authentic material, especially under stress or cognitive load.

Platforms

Spreading is the business

Engagement-optimised algorithms promote sensational and divisive content; misinformation included.

Detection

The referee hallucinates

AI-based detection delivers mixed results, and AI's own hallucinations mean it can't be trusted to fact-check itself.

The slow damage: as synthetic content spreads, people lose the ability to trust any source; and that uncertainty corrodes not just politics but economic decisions, science and social cohesion. Including, circularly, decisions about AI investment itself.

ntworld.inkWorkshops10 / 18

The military dimension

An arms race that doesn't read earnings reports

·Both the US and China are deploying AI across air, land, sea and space operations; drones, underwater vehicles, ground robots, and intelligence systems that digest vast imagery and intercepted communications.

·RAND Corporation researchers identify five hard national security problems that artificial general intelligence could create, including new risks of US-China conflict.

·The guardrails are voluntary: Pentagon AI guidelines (2021, updated 2023) and a US-led declaration on military AI joined by over 50 countries; but neither the US nor China has accepted binding limits.

Beyond the Horizon ISSG

"The rapid pace of AI development, particularly in autonomous weapons, cyber warfare, and intelligence analysis, has triggered an arms race reminiscent of the Cold War nuclear buildup."

The bubble connection: if AI is framed as national survival, spending continues whatever the commercial returns; which props up the economics, which reinforces the framing.

ntworld.inkWorkshops11 / 18

The one question to carry out of this room

Every AI claim has a beneficiary

Incentives to inflate

"AI changes everything, invest now"

AI companies attracting investment and justifying valuations · defence contractors securing funding · governments justifying security spending and deterring adversaries · content creators farming engagement from hype.

Incentives to deflate

"It's all hype, stay away"

Short sellers profiting from falling valuations · competitors undermining rivals' perceived lead · safety researchers wanting development slowed · threatened industries defending existing business models.

The habit: almost every piece of AI news is shaped by someone's agenda. That doesn't make all claims equally suspect; it means the first question is always "who benefits from this narrative?"

ntworld.inkWorkshops12 / 18

Why this is one problem, not four

Each dimension amplifies the others

·Economic → geopolitical: stock valuations justify government AI spending; corporate lobbying shapes export controls and industrial policy.

·Geopolitical → military: US-China competition accelerates military AI; export controls decide who gets advanced capability.

·Military → information: military AI enables sophisticated influence operations; defence funding shapes which research (and which narratives) get amplified.

·Information → economic: hype narratives drive investment decisions; misinformation about capability inflates valuations. And around it goes.

The loop in one sentence: inflated valuations justify security spending, which reinforces the "AI is everything" narrative, which attracts more investment. Breaking in anywhere requires understanding all four dimensions at once; that's what makes it wicked.

ntworld.inkWorkshops13 / 18

The strangest evidence of all

The people selling the boom keep saying "bubble"

Sam Altman · OpenAI

"Are we in a phase where investors as a whole are overexcited about AI? My opinion is yes."

Sundar Pichai · Google

"There are elements of irrationality in the AI market right now."

Jamie Dimon · JP Morgan

"AI is real... but some money invested now will be wasted."

Australian Financial Review · September 2025

"If we really are in another share-market bubble, it's surely the most anticipated example in history."

Read it through the incentives lens: when the CEOs with the most to gain from hype are managing expectations downward, they may be hedging reputations ahead of a correction; or sincerely warning. Either way, "even the insiders say bubble" is itself a narrative worth interrogating. Ray Dalio of Bridgewater has raised similar warnings about AI investment moving too fast.

ntworld.inkWorkshops14 / 18

The parallel everyone reaches for

The dot-com bubble: the technology was real anyway

Then

1999 to 2000

Revolutionary technology, massive investment, circular financing between firms; then a 78% collapse in the NASDAQ. The internet still changed the world; most of the 1999 companies didn't live to see it.

The infrastructure echo

Fibre then, data centres now

As author Bethany McLean notes, the 1990s fibre-optic overbuild left infrastructure that sat unneeded for decades. Today's question: will the current data centre boom rhyme with that; useful eventually, ruinous for whoever paid?

The takeaway from history: "bubble" and "real revolution" are not opposites. The dot-com crash and the internet's triumph both happened. AI could follow the same script; a financial reckoning and a genuine transformation, in either order.

ntworld.inkWorkshops15 / 18

Why any of this matters north of the Berrimah Line

Bubble dynamics reach the Territory in five ways

·Buying AI: vendor claims are shaped by bubble incentives. The Australian Government's 2024 Copilot trial found real productivity gains, but more modest than the marketing suggested; evaluate before you procure.

·Workforce: the same Copilot evaluation flagged that AI efficiencies could fall hardest on administrative roles; women hold about 75% of the potentially affected roles in the APS.

·Information integrity: remote NT communities with limited connectivity can be especially exposed to misinformation; media literacy is bubble insurance.

·Indigenous data sovereignty: the Copilot trial also noted outputs biased toward western norms; AI adoption must respect Indigenous Cultural and Intellectual Property frameworks.

·Strategic resources: the NT's critical minerals connect directly to the AI supply chain contest; Australia's rare earth agreements with the US create opportunity and geopolitical exposure at once.

ntworld.inkWorkshops16 / 18

The takeaway: five things to remember

Reading the AI economy without panic

·It's not just a market story; economics, geopolitics, information and military competition are one tangled system.

·Ask who benefits; every claim about AI capability, up or down, has someone rewarded for your believing it.

·Watch the circular money; revenue between hyperscalers is not the same as demand from customers.

·Bubble and revolution can both be true; the dot-com crash didn't stop the internet, and an AI correction wouldn't stop AI.

·Act locally; for NT organisations this means sceptical procurement, watching workforce impacts, and building media literacy before you need it.

ntworld.inkWorkshops17 / 18

Sources used

Where this deck's claims come from

Goldman Sachs Research (2025): the circular-financing warning quoted on the money slides, and the analysis of AI spending approaching productivity limits

Market and company figures (2025 reporting): ~$400B projected AI infrastructure spend, ~80% of US market gains, ~30% top-five S&P concentration, ~1.1% of US GDP (Q2 2025), OpenAI's $13B projected revenue and $13.5B H1 loss, 498 AI unicorns, and the IMF valuation warning

RAND Corporation: the five hard national security problems AGI could create · Beyond the Horizon ISSG: the arms-race assessment quoted on the military slide

Australian Government Microsoft 365 Copilot trial evaluation (2024): the productivity, workforce (75% of affected APS roles held by women) and cultural-bias findings on the NT slide

Australian Financial Review (September 2025): "the most anticipated example in history" · executive remarks by Sam Altman, Sundar Pichai, Jamie Dimon and Ray Dalio as reported in 2025 coverage

Bethany McLean on the 1990s fibre-optic overbuild, and standard accounts of the 1999 to 2000 dot-com collapse (NASDAQ down 78%)

Prepared August 2026 for the NT World Ink AI Special Topic Workshops, from the NT World Ink research file on the AI bubble. Market figures age quickly; treat them as a late-2025 snapshot.

ntworld.inkWorkshops18 / 18

Bubble Trouble:Making Sense of the AI Economy