The word does not appear.
Not once. Not 'digital assets.' Not 'stablecoin.' Not 'distributed ledger.' In a document built to project the Australian economy to the mid-2060s — four decades of demography, productivity, fiscal drag and structural change — the technology that settles billions of dollars a day across public chains inside that same jurisdiction gets zero lines, zero footnotes, zero appendix entries.
Artificial intelligence gets a section. It gets named as one of the structural transformations reshaping the country. It gets modelling assumptions attached to it.
That asymmetry is the news. Coinbase filed a public response within a day, and the compressed version of it reads like this: the report ignores the digital financial infrastructure that AI agents will eventually need.
Demand-side claim dressed as a supply-side observation. I've spent the week pulling it apart — reading the document the way I read contract source, not for what it argues but for the structure of the omission. A diff tells you more than a manifesto. And this diff is unusually clean.
What the document actually is
The Intergenerational Report is not a policy platform. It is a projection model with narrative scaffolding bolted to it. Treasury produces it on a five-year cycle — 2002, 2007, 2010, 2015, 2021, and the edition now circulating — and its purpose is narrow and specific: take the population, participation and productivity assumptions, run them forty years forward, and show what happens to the Commonwealth's balance sheet.
Its five named shifts have been stable across recent editions. The care economy. Climate and the energy transition. Digital and data, which has now been reframed around AI. Demographics. And the geopolitical reordering of trade and supply chains.
The mechanical spine of the whole thing is productivity. The 2021 edition ran on an assumed labour productivity growth rate of roughly 1.5 per cent a year — deliberately above the approximately 1.2 per cent average of the three preceding decades. That gap between assumption and history is where the fiscal arithmetic either holds or fails. Every structural story in the document has to be translated into that number, or into a spending line, or it does not get in.
That is the entire explanation for the silence. Not ideology. Model scope.
AI qualifies because it can be plugged into total factor productivity. You can argue about the coefficient, but you can argue about the coefficient, which means it belongs in the document. Crypto qualifies for nothing, because in Treasury's unit of analysis it has no fiscal line item. It does not consume budget. It does not produce a measurable national income stream that survives an audit trail. It does not change the aged care or NDIS or interest-payment projections by a dollar that anybody can defend in a footnote.
Absence from a fiscal projection is not a verdict on technology. It is a statement about what the projection can measure. That distinction is where every hot take on this story has already gone wrong.
The Australian ledger, outside the document
Here's the tension worth sitting with. Australia is not a low-adoption market. It has historically ranked near the top of global crypto adoption indices on a per-capita basis — retail participation rates that put it alongside the United States and well ahead of most of continental Europe. Several million Australians have held digital assets at some point in the last cycle. That is not a fringe statistic.
What the country also has is an existing regulatory framework that already reaches crypto. ASIC supervises digital asset financial products. Treasury has run consultations on digital asset platform licensing, with the stated direction being to bring exchange and custody businesses inside the existing Australian Financial Services Licence perimeter rather than create a bespoke regime. The Australian Taxation Office has run multi-year data-matching programs against exchange records, and the capital gains footprint it has reconstructed from those programs is not small — it runs into the billions in taxable gains across individual returns.
So the correct read is not 'Australia has no crypto policy.' Australia has crypto enforcement, crypto taxation, and a licensing reform in progress. What it does not have is crypto strategy. Enforcement is a ledger of what you take. Strategy is a ledger of what you build. The 40-year plan is a strategy document, and crypto is not in the strategy layer.
There is one more structural detail that matters more than it looks. Australia's superannuation system is one of the largest pools of long-horizon capital on earth, on the order of four trillion Australian dollars, and it is the natural buyer for any asset class the national strategy blesses. When an asset class is absent from the long-horizon document, the default posture of every prudential decision-maker in that system is not hostility. It is inertia. You cannot allocate against a strategy that does not name you.
What an AI agent actually needs — and where the argument breaks
Strip the Coinbase statement down and you get five requirements. An autonomous agent transacting in the economy needs identity, authorization, a payment rail, settlement proof, and some form of reputation or recourse. That's the whole stack.
Of those five, only two plausibly require a public chain under any conditions at all, and only under specific ones.
Take the payment rail first, because that's the load-bearing claim. The reason machine payments are a genuine problem is not that existing rails are slow. It's that existing rails have fixed per-transaction costs that destroy sub-cent economics. Card interchange plus the fixed acquirer fee means anything below roughly a dollar is structurally unviable on card rails — you are paying thirty cents to move half a cent. ACH and SWIFT batch, which means they settle on a clock, not on a request. An agent that needs to buy one API call, one inference, one kilobyte of data, in the middle of a task, cannot wait for a batch window and cannot pay a fixed fee larger than the purchase.
That is the actual specification. Not 'agents need blockchain.' Agents need deterministic settlement below the fixed-cost floor of the legacy rails.
The technical answer that has emerged over the last eighteen months is HTTP 402 — the status code that sat dormant in the spec for three decades — operationalized as a machine-payable response. Coinbase shipped a version of this as x402, settling on Base in USDC. Google's AP2 work on agent payments extends the same idea into authorization and mandate structure. And there's an identity layer forming around registries that let an agent prove what it is and who stands behind it, which is what the ERC-8004 discussions in Ethereum governance forums are circling.
Now the part nobody in the narrative wants to hear. Most agent payments do not need a chain. If a single firm's agents are transacting with that same firm's services, the ledger is internal and the settlement is a database write. Chains win in exactly one scenario: when the trust boundary between buyer and seller does not align with any shared counterparty account. Cross-jurisdiction, cross-organizational, anonymous-to-each-other, sub-cent, high-frequency. That is a real market. It is also much smaller than the total addressable market being implied when people say 'AI agents will need digital financial infrastructure.'
The argument is coherent. The size of the argument is being inflated by roughly an order of magnitude, and nobody in the policy conversation has priced that in.
The four-tenths-of-a-cent problem
Follow the economics down one more level, because this is where the narrative collides with the hardware.
An agent buying inference pays fractions of a cent per call. The payment mechanism must cost less than the thing purchased, or the agent's economics invert and the whole use case evaporates. On Ethereum mainnet, a simple transfer in a quiet market costs a few cents and in a congested market costs dollars. Volatility is just fear wearing a disguise — except for an autonomous agent, which cannot feel fear and therefore cannot absorb variance. It needs a price it can budget against. Gas volatility is not a sentiment problem for a machine. It is an accounting failure.
The bridge that solves this is account abstraction with a paymaster: the agent spends a stablecoin, the paymaster absorbs gas on the back end. That converts an unpredictable cost into a predictable one. It also relocates the gas risk onto whoever runs the paymaster — usually the same operator running the sequencer.
Which brings me to the structural conclusion I did not expect to reach when I started this. The AI-agent payments thesis is, mechanically, an optimistic-rollup thesis, not a zero-knowledge thesis, and almost nobody has said so out loud.
Run the unit economics. ZK proving costs per transaction remain brutally high. I've been through the numbers repeatedly since the 2020 Curve audit work, and the story hasn't changed: unless gas returns to bull-market levels of congestion, ZK operators are bleeding on every proof they generate for a transaction worth a fraction of a cent. Cut the fee to what an agent can pay and the proving cost eats it. The math only closes at fee levels that no micropayment use case can fund.
Optimistic rollups don't have that problem, because they don't prove per transaction. They post data, assume validity, and rely on a challenge window. That's cheap at the margin. It's why the agent-payment experiments are landing where they're landing, and it's why the ZK camp's answer to this — better provers, hardware acceleration, recursive proofs — is a 2027 answer to a 2025 problem.
Who gets the float
Now ask the question that separates analysts from commentators: if this thesis is right, whose revenue line does it become?
USDC reserve income flows to Circle, and a contractual share of it flows to Coinbase. Coinbase's own stablecoin-related revenue ran close to a billion dollars annually through the last reported cycle, sitting inside its subscription-and-services segment rather than its trading revenue. Base sequencer fees accrue to Coinbase as well. Neither is enormous today. Both scale directly with the thing the public statement is advocating for.
I want to be precise about this, because it's the part that gets misread as cynicism. The statement is not dishonest. It is also not disinterested. Coinbase is a listed company whose equity story increasingly depends on non-trading revenue — stablecoin economics, custody, chain fees — precisely because spot trading revenue is cyclical and the current tape is chop. A sideways market is a market where transaction revenue flatlines and infrastructure revenue is the only growth line left on the slide.
So the policy intervention is revenue defense with a technically valid argument attached. Both things are true at once.
And there's a second-order effect that matters for anyone trading this narrative: if machine payments do scale, the order flow is enormous, high-frequency and structurally predictable. Which means the interesting question is not whether it happens on-chain. It's who captures the value of moving it. Intent-based routing, if it becomes the standard for agent payments, doesn't eliminate MEV — it relocates it into off-chain solver auctions where a handful of firms see the flow before it settles. That's a concentration risk dressed as a usability upgrade.
The Australian divergence nobody is modelling
Back to the document, and the thing I think is genuinely under-reported.
The interesting fact is not that Australia's strategy layer ignores crypto. It's that Australia's household layer doesn't. High retail participation, an enforcement apparatus already generating tax revenue from that participation, and a national strategy that treats the whole category as not existing — that is a divergence, and divergences of that shape resolve eventually, usually in the direction of the thing that already generates tax receipts.
There's a precedent I watched in real time. When the US spot Bitcoin ETFs launched in 2024, I worked with a Cape Town hedge fund to decompose IBIT's on-chain inflow signature by trading session. The prevailing narrative was retail-driven accumulation. The tape said something else — a measurable concentration of institutional accumulation during Asian hours, against the US retail flow. The divergence between narrative and flow lasted months before the narrative caught up.
This is the same shape. The narrative says Australia has no crypto economy. The tax data says otherwise. The strategy document has simply not been updated to the tax data's reality yet, and forty-year models update slowly by design.
The contrarian read: the omission may be the good outcome
Here's the angle you won't find in the outrage posts.
Everyone is framing this as Australia ignoring crypto. The sharper framing is that AI earned its place in the document because AI can be modelled as a productivity input, and crypto didn't because it can't be. That's a measurement failure on the industry's side, not a hostility failure on Treasury's. If you want into the next forty-year plan, the deliverable is not a louder statement about agents needing payment rails. The deliverable is a defensible number showing digital settlement infrastructure raising measured productivity in an Australian industry, filed with a methodology an actuary can attack and survive. That doesn't exist yet. When it does, the document will name it, and it will name it without anyone lobbying.
And then the uncomfortable part. A forty-year plan that names you is a plan that taxes you, licenses you and designates you. Australia has already chosen its licensing path — bring platforms inside the existing AFSL perimeter rather than build a bespoke regime. Being absent from the fiscal outlook means no sector-specific levy, no carve-outs, no designation as a strategic industry with the reporting obligations that follow. The 2021 mint rush taught me that the button is a lever, not a purchase — the act of being included is never the act of being rewarded. Inclusion has costs, and the industry has spent a decade asking for them without pricing them.
The flip side is real too. Silence is not a ban, but it is also not a subsidy, and the subsidies — tax treatment, prudential permission, clear legislative standing — are what institutional capital actually needs before it moves at size. Yields were too good to be true, so we didn't allocate; the same logic runs the other way. A four-trillion-dollar superannuation pool will not allocate to an asset class the national strategy doesn't measure, no matter how many Australians already own it in their brokerage accounts.
What I'm watching from here
Three signals, in order of information content.
The first is whether the next Intergenerational Report names it. Not whether a minister does — ministers name things weekly. Whether the modelling annex references digital settlement at all. If it does, that's a strategy-layer shift and the super funds move within a cycle.
The second is the ASIC and Treasury licensing legislation as it lands. Silence in a forty-year plan is soft. Licensing rules are hard, and they determine whether an Australian-facing exchange can custody, clear and offer derivatives inside the perimeter or has to leave.
The third is machine-payment volume that is real rather than announced. Not integrations — settled value, on a public chain, denominated in stablecoins, at a transaction count per second that no human generates. That number is the only honest test of whether the AI-agent infrastructure argument is a thesis or a talking point.
Everything else this week has been narrative. Narrative doesn't move capital. Measured flow does.