If AI reshapes work, it reshapes the marginal buyer. And the marginal buyer is what moves asset prices.

By the team at Banyantree Investment Group

───

The superannuation contribution file arrives overnight.

It is an unlovely thing, the remittance advice that tells the fund which accounts to credit, a list of amounts and identifiers that looks more like plumbing than finance. On a good day it is perfectly dull: the totals reconcile, the cash lands, the system ticks over. In the morning, someone checks it with the same mild attention you give the weather. Not because it is interesting, but because it is there, every day, whether markets are calm or hysterical.

For years it changes the way rainfall changes, within a familiar band. Then, gradually, it does not.

The list is still long, still regular, still boring, but it is lighter in places you do not see in headlines. Headcount is down in quiet corners. Hours have become more elastic. A certain kind of corporate language grows sunnier: efficiency, leverage, productivity. The market calls it progress. The file calls it flows.

If you spend enough time around markets, you begin to notice that we talk about them as if they are arguments. Fundamentals versus narratives. Bulls versus bears. Value versus growth. But what keeps markets continuous is less philosophical than that. Someone, somewhere, has to keep showing up with cash, not once, but repeatedly, in numbers large enough to matter. Not because they are inspired. Because they are enrolled.

Australia has a particularly visible version of this enrolment. Superannuation is tied to wages and collected by default. It is not that every dollar ends up buying Australian shares. It does not. It is that an enormous, habitual flow is converted into investment exposure across portfolios, local and global, whether the underlying investor feels confident, tired, distracted, or simply busy living their life.

You can call it a system. You can call it a policy. In practice it behaves like a pipe.

And when a large pipe changes, the room changes with it.

Why this matters now is that three quiet supports have started to shift at the same time. The Super Guarantee has reached its legislated ceiling of 12 per cent, so contribution growth now depends far more on wages and employment than on policy ratchets. White-collar employment growth has softened, and professional-services wage growth has cooled at the margin. Meanwhile the AI platforms that sit at the top of the index are pulling more weight, literally, inside benchmark-aware portfolios. None of this is a crash signal. It is a mechanism becoming visible moment.

This is where AI enters the story. Not as a sector you pick, or a slogan you fear, but as a force operating simultaneously on both ends of the chain. From below, it threatens the wage income that feeds the file. From above, it concentrates the index that the file ultimately purchases. These are not separate arguments that happen to share a subject. They are two expressions of the same underlying transfer: economic value migrating from the broad labour income of the knowledge-work class to the narrow capital income of the platforms doing the displacing.

The workers whose wages fund the pipe are also, through their default superannuation allocations, buying growing stakes in the companies replacing their tasks. The loop is not conspiratorial. The mechanism is clear, even if its magnitude and pace remain genuinely uncertain. And the mechanism is already running.

"At the margin" is a phrase used in markets to describe something simple: prices are set by the next buyer, not the average one. The marginal buyer is not necessarily the largest pool of capital in the world. It is the buyer who shows up, reliably, when bids and offers meet. Change that buyer, or change how reliably they appear, and you can change market behaviour without changing the story people tell themselves.

There are two kinds of marginal buyer worth distinguishing. The trading marginal buyer sets prices on the day through liquidity, positioning, and market structure: visible, discussed, modelled. The allocation marginal buyer is slower, easier to miss, and far more powerful over months and years, because it supplies persistent net risk-bearing capacity. Superannuation flows are the allocation marginal buyer for Australian-listed assets. For global exposure, where the largest funds now hold forty to sixty per cent of growth assets internationally, the mechanism is less direct. The marginal buyer becomes global, and the chain runs through different pipes. What follows is most legible for the domestic market, where the paycheque-to-price link is unusually clear.

Patience, in this context, is not a temperament. It is a structural condition: the ability to hold a long-duration, illiquid, or contrarian position without forced exit, because the portfolio has the background support to wait for a thesis to be proven out. When the allocation marginal buyer is steady and diverse, patience is cheap. When it becomes variable or concentrated, patience becomes a premium — one the portfolio must fund from somewhere else.

APRA's annual superannuation statistics for 2024–25 give the throughput plainly. Total contributions: A$209.4 billion. Benefit payments: A$132.2 billion. Net contribution flows: A$70.9 billion. Those are not decorative figures. They are the throughput of a machine that quietly converts wages into markets.

It is tempting, at this point, to overstate the claim. Superannuation is not the invisible hand that sets the price of every global asset. But compulsory super is a reliable source of incremental risk-bearing capacity. When that capacity is steady, it makes patience cheap. When it becomes thinner or more contested, other forces take turns at the microphone: offshore flows, corporate programmes, leverage, and systematic strategies that are steady until they are not.

Liquidity, in these moments, stops being a concept and becomes an event.

The paycheque does not merely buy groceries. In Australia, by design, it buys the market too. Not directly, not as a heroic act of retail optimism, but as a rhythm. A habit. A law translated into payroll software.

You can see how this works as a chain of ordinary decisions. Wages are paid. Contributions are made. Portfolios are rebalanced. Risk is taken. Benchmarks exert gravity. Index weights concentrate. In stress, correlations rise. In calm, patience looks like genius. In both cases, the contribution file keeps arriving, so long as work keeps paying.

And work, increasingly, is what AI is about to rearrange.

───

Exhibit 1. From Paycheque to Price

Markets feel discretionary at the margin, but the biggest buyer is often automatic. When wages feed legislated super contributions, contributions become allocation, and allocation becomes the persistent bid that shapes prices. AI matters twice: it can disrupt the wage base, and it can concentrate the index. Illustrative.

───

I. When work becomes less steady, investing becomes less steady

Consider a person in her forties. She has skills. She is, in the current language, augmented. Her job survives. But her life becomes a patchwork. A contract extension here. A retraining course there. Two gaps of three weeks between projects that are not called unemployment but feel like it. A second role picked up to smooth cashflow. Employment looks stable on paper. Income becomes spikier in real life. Saving becomes more defensive. Planning becomes more tentative. Contributions still happen, but they arrive in bursts.

Nothing dramatic needs to occur for the contribution file to change. Stability only needs to erode at the edges.

There is already evidence that those edges are eroding. White-collar employment in Australia grew by just 0.8 per cent in 2024–25, 43,900 workers, the slowest rate in a decade, according to Deloitte Access Economics. In the same period, the Wage Price Index for professional and financial services ranked among the softest sectors in the economy, lagging the national headline. A single year of softened growth is also consistent with post-pandemic labour market normalisation following an unusually hot cycle, and that alternative deserves acknowledgment. What makes this reading more than noise is the coincidence of two structural changes arriving simultaneously: white-collar hiring cooling at precisely the moment the SG ratchet has switched off. These are not confirmations of the thesis. They are the thesis beginning to show itself in ordinary numbers.

There is a further structural pressure that receives almost no attention in investment commentary. The Super Guarantee rate reached its legislated ceiling of twelve per cent in July 2025. The decade-long tailwind that inflated employer contribution growth — successive annual rate increases that mechanically boosted inflows above the organic growth of wages and employment — is now spent. From this point, employer contributions grow only as fast as wages and employment grow. The engine that has been flattering the pipe’s organic growth rate is switched off.

At the system level, aggregation smooths a great deal of individual volatility. Millions of pay cycles blur into a steady-looking line. The point is not that flows become chaotic overnight. The point is that the margin for patience narrows when net cashflow slows, and the structural tailwinds that have been sustaining that cashflow — rising SG rates and a post-pandemic employment surge — have both run their course simultaneously.

Variable flows do not create a disaster. They change the cost of patience, the structural condition that allows a portfolio to hold what the analysis supports rather than what the liquidity allows.

When inflows are steady, patience is cheap. Illiquid allocations feel forgiving. When inflows become variable or structurally softer, rebalancing constraints tighten, liquidity demands from benefit payments compete for the same cash, and the holding period for long-duration positions effectively compresses. In that world, the portfolio’s job is not only to compound. It is to remain coherent when the cash arrives unevenly.

───

II. When the file funds its own compression

There is a second process operating on the other end of the chain, and it runs in the same direction as the first.

The contribution file does not choose what to buy. It moves through default allocations into balanced and growth options, most of which carry meaningful exposure to market-cap-weighted indices. Those indices, in turn, do not choose what to own. They own what is largest. And what is largest, in 2025, is a small number of technology platforms capturing the rental income of artificial intelligence at a scale the index has not seen since the late 1990s.

As their market capitalisations grow, their weight in the index grows with them. The contribution file, designed as a mechanism for broad participation in economic growth, becomes increasingly a mechanism for concentrated exposure to a small number of AI beneficiaries. The passive investor who believes they hold a diversified slice of the economy holds something different: a market-cap-weighted bet on the platforms capturing AI rents, surrounded by a diminishing tail of everything else. Their portfolio is diversified by name count. By risk source, it is something else entirely.

Here the loop closes. When a large technology platform replaces a team of junior analysts with a language model, the productivity gain flows to the platform’s shareholders. Those junior analysts’ super contributions, flowing through their default allocation into the index, are partly funding that same platform’s growing weight. The pipe finances the concentration of the very asset class it is buying. The workers whose wages built the pipe are also, without knowing it, funding the instruments that are shrinking the pipe.

For portfolios, this concentration creates two risks that tend to arrive together. The first is valuation risk: historically, extreme index concentration has resolved through multiple compression rather than orderly rotation. Japan in 1989, technology in 2000. The mechanics differ each time. The arithmetic is similar. The second is correlation risk: when the few companies dominating the index move together, as they do in stress, the diversification benefit of holding the index disappears at precisely the moment it is most needed. A broad index that is actually a concentrated portfolio behaves like a concentrated portfolio when it matters.

The investor who is aware of this has two choices that are hard and one that is easy. They can build genuinely diversified exposure that accepts meaningful tracking error relative to a concentrated benchmark. They can hold the concentration explicitly, with a view on its durability, rather than through the fiction of passive diversification. Or they can do what most investors do: accept the index weight as given, call it diversification, and discover the problem when it is too late to be early.

───

III. Who captures the gains decides who does the buying

AI may boost productivity. It may raise GDP. But productivity is not the same thing as broadly shared purchasing power. A society can become more productive while the wage share stagnates. It can generate more output while fewer households accumulate the surplus that funds the bid.

This claim is debated, and it should be. Some economists argue that AI will complement skilled labour and widen the wage distribution upward. Others, notably Daron Acemoglu and colleagues, argue that the task-substitution effect for knowledge work is more likely to compress labour’s share than previous waves of automation, because AI targets the cognitive tasks that had previously sheltered the professional middle class. The honest position is that this is uncertain, and the outcome depends heavily on adoption speed, regulatory response, and the ability of labour markets to absorb displaced workers. What is not uncertain is the direction of the mechanism, even if its magnitude is unknown.

If AI’s gains accrue disproportionately to capital owners, the bid shifts away from households. It does not vanish. It relocates, and the relocation matters, because different surplus-holders buy differently.

Corporate buybacks are a corporate bid and tend to be pro-cyclical: strongest when confidence is already high, weakest when caution is rational. Sovereign pools and public funds can be patient, but they are policy-constrained, and those constraints tighten suddenly when public mood shifts. Systematic strategies provide steady demand until the moment they become liquidity-sensitive, at which point they pull in the same direction as the market is already moving. Concentrated private wealth can be long-term, but ownership becomes narrower, and behaviour can be more narrative-driven, which changes market microstructure in ways that are hard to model and easy to feel.

The broad, wage-linked bid had a stabilising quality not because it was wise, but because it was habitual and diverse. It funded the market from millions of distinct economic circumstances simultaneously. When that diversity contracts, the remaining buyers share more assumptions, more exposures, and more reasons to move in the same direction at the same time. A thinner market is not necessarily a weaker market. It is a less forgiving one.

───

Exhibit 2. Three Labour Shapes and Their Market Fingerprints

There isn’t one AI outcome, and there won’t be one market consequence. The shape of labour change decides whether the bid stays steady, becomes erratic, or fades, and the watchpoints show which path you’re on before prices do. Illustrative.

───

What to Watch

The instruments that tell you whether the pipe is changing

If this thesis is right, these are the tells. Not market narratives, not conference chatter, but the small, repeatable measures that change before price does.

The most useful signals are not in equity markets. They are in labour market data, specifically in what is happening to employment stability, wage distribution, and churn in knowledge-intensive occupations. ABS Wage Price Index releases, read by sector rather than headline, show whether professional and financial services wages are tracking above or below the economy. Deloitte Access Economics employment forecasts track white-collar headcount growth. These are the early readings of a process that shows up in the contribution file years later. Aggregate unemployment is a lagging and politicised indicator. Sector-level wage growth and white-collar employment trends are not.

Superannuation contribution data provides the second instrument, and it requires a specific adjustment to be useful. Employer contribution growth has been inflated for a decade by successive SG rate increases. With the rate now at its legislated ceiling, any further growth in employer contributions reflects only wage growth and employment growth. Strip out the SG rate effect, which APRA’s quarterly releases allow, and you can see the organic growth rate of the pipe directly. A sustained deceleration in that organic rate, particularly if accompanied by acceleration in benefit payments as the retiree cohort expands, narrows the net flow that funds the market’s patience.

The third instrument is index concentration, specifically the spread between equal-weight and cap-weight index returns. When this spread is wide and persistent, when the index is rising while most of its constituents are flat or falling, the structural buyer is increasingly funding a concentrated bet rather than broad participation. Advance/decline lines and sector dispersion tell the same story in different registers. A market where index and breadth diverge for an extended period is one where the assumptions embedded in passive allocation are most likely to be tested.

The fourth instrument is specific to the mechanism this essay describes: wage growth divergence between professional services and AI-adjacent occupations. If software engineers, AI researchers, and platform managers command accelerating wage premiums while junior lawyers, junior analysts, and junior accountants see wage growth moderate, in absolute terms and relative to their predecessors a decade earlier, that divergence is the earliest labour market signal that AI task substitution is operating in the knowledge-work cohort that funds the contribution file. It will not appear in a single ABS release. It requires watching the distribution, not the mean.

───

The contribution file will keep arriving. For a long time it may look exactly the same. It will still be boring. It will still be reconciliation and throughput and operational competence. In that sense, AI may not change it at all, not this year, perhaps not for several.

What AI may change is not the file’s appearance but the water table beneath it. The volume and steadiness of the rain. By the time the change is visible in the file, it has been operating in labour markets for years. By the time it is visible in labour markets, it has been operating in task composition and wage distribution for years before that.

The unsettling possibility is not that GDP falls. It is that GDP rises, that productivity improves, that corporate earnings grow, that the platforms at the top of the index compound beautifully, while the bid that has financed the market’s patience becomes narrower and less diverse. Markets have shown they can tolerate that divergence for a long time. What they cannot tolerate is the assumption that it is not occurring.

The first sign will not be in the price. It will be in the file, a little lighter in places that no one, at first, thinks to look.

───

General information only. This article does not consider your objectives, financial situation, or needs, and is not financial advice. Numbers cited are factual or sourced; no forward-looking figures should be taken as forecasts. Prepared for wholesale and professional investors.

The Paycheque That Buys the Market