Australia outperformed this week. The ASX 300 rose 0.9% while the Nasdaq shed 4.5%, and around 6% over the past fortnight, with emerging markets dragged lower by Korea and its large chipmakers. The S&P 500 finished roughly flat, and markets with more defensive exposures, the UK as well as Australia, edged higher. Most of the damage came in a single week of accelerating pessimism about the AI trade, and some of the places where defence has shown up are less obvious than the usual suspects.

Part of what is supporting the local market is that interest-rate expectations have ebbed. The RBA's October minutes were hawkish, a unanimous 25 basis point hike to 4.60%, a Board leaning on capacity constraints, and an admission that the trade-off between inflation and full employment has deteriorated, but September's labour figures took some heat out of the story. Unemployment rose to 4.7% from 4.6%, its highest in almost five years, as 39,000 jobs were lost; full-time employment actually rose 34,000, so the damage was in part-time work. Trend employment growth of around 16,000 a month is well short of the roughly 30,000 needed to absorb population growth. Markets now price less than one full hike, around 20 basis points over the coming year, down from closer to 30, and the Australian dollar slipped towards 69 US cents.

The lightning rod for AI related nerves, especially in the local market, was the collapse of the proposed $40 billion-plus Firmus float, which now looks set to try again in the US. It crystallised a growing fear that the hyperscalers may struggle to monetise AI quickly enough to pay for the compute they need. The awkward part is that chip prices themselves continue to rise unabated; what investors are questioning is the valuation being placed on the longevity of the cycle. And it has become a credit problem. The hyperscalers, and the companies they have engaged to build out capacity, are raising a great deal of debt, much of it off balance sheet through convertibles, vendor financing and special-purpose vehicles, with chip depreciation schedules that assume a long useful life. As investors do their sums, credit spreads and credit default swaps have widened sharply, most of all on high-yield, data-centre-driven deals. CoreWeave fell hard in both its shares and its bonds, AirTrunk postponed its Singapore listing, and OpenAI reined in expectations for an imminent IPO. A more sceptical market is demanding a higher cost of capital.

Locally, the market held up because of the companies in the index and specifically lack of tech stocks and direct AI exposure. Health care, utilities and staples played their defensive role, but the banks also did well, and only materials and real estate were meaningfully weaker. The ASX is far less exposed to the AI thematic than US and Asian markets, for once, our two-pillar market structure was a defence. Company news was steady: Rio Tinto held guidance despite diesel cost pressure, Bank of Queensland reported in line, and Treasury Wine said demand was tracking slightly ahead of expectations. Housing credit was the soft spot, with AFG reporting September quarter lodgements down 4.5% on the June quarter.

The US earnings season matters more than usual this time. On consensus, earnings momentum in corporate America is broadening out, with blended growth of 29.0% expected from a wider cohort of stocks, but dig underneath and it remains narrow. Strip out Micron and Nvidia and index growth falls to 20.0%. The Magnificent 7 are expected to grow a whopping 33.2%, but take out Nvidia and Meta — the latter flattered by an easy year-ago comparison — and the remaining five slow to 7.5%, weighed down by how much they are spending on the very chips that are driving the headline. Energy's +127.6% is largely a price effect, financials are growing just 3.0%, and good results are being sold rather than bought.

The more interesting signal is outside the AI complex altogether. Goldman Sachs' S&P 500 ex-AI index, which strips out the roughly 45% of the index Goldman classes as AI enablers (chips, hyperscalers, cloud), leaving the 55% made up of health care, utilities, industrials and the like, has rallied over recent sessions precisely as the AI trade went the other way. That is close to complete diversification. Just as important, on Goldman's figures you are getting similar earnings growth from the ex-AI cohort for a much lower price-earnings multiple. One swallow doesn't make a summer, but the same rotation is showing up elsewhere. Global small caps fell only around 1% through the volatility, and the quality-growth small-cap managers we follow, hammered earlier in the year,  rose three to four per cent. In emerging markets, managers who have rotated out of chip exposure are also up. That is hard to model, because it is not necessarily driven by improving fundamentals; it is a substitution effect, as investors holding too much chip-related exposure look for something else that does roughly the same job.

Historical analogies are always a little dodgy, and this is not the dot-com boom. But after 2000 there was a similar rotation into unloved assets, and despite the recession that followed, many supposedly cyclical smaller companies rose progressively over the next few years because they started cheap and the market looked through the downturn. Marching to a different tune may yet deliver more diversification than expected.

Diversified portfolios have also done better than might be expected, and the weaker Australian dollar has helped. Still within its 69–73 US cent range and drifting towards the bottom of it, the dollar is a silver lining for Australian investors, who as a peer group hold an increasing share of their portfolios in international equities. More encouragingly, it is starting to behave cyclically again — perhaps reflecting demand for our materials or how traders see the local currency — restoring some of the defensive quality that looked noisy and unreliable earlier in the year.

All of this feeds a question our internal investment committee has been debating: what diversification really means now. Years ago the main lever was the growth–defensive split. More recently the diversification came from within asset classes, how much US exposure relative to the rest of the world, and duration versus credit within bonds. Today the picture is more nuanced still. Short of an overt growth–defensive call, which is fraught with timing risk, advisers may wish to consider looking under the bonnet at the kind of exposures discussed above.

We also have to look out of the front window. In this week's conversation, Andrew Hunt's message was that there is no need yet to run for the hills. His latest research looks beyond the balance-sheet side of the debt-to-GDP conundrum, France has riots over too little spending while its bond market objects to too much, to the cash-flow side, and to what tends to trigger a bond-market upset. His answer is the behaviour of the dominant lender, which in this era means foreign investors. The US relies on capital inflows, mainly from Asia's exporting nations recycling their surpluses, so the strength of Asian currencies, including a steadily rising, managed renminbi, is the marker to watch. Because that adjustment is managed, there is a case for hope: the rotation could continue in a measured way, without an imminent downswing.

The AI trade has become a credit story, and in owning something different finally paid off. Next week brings more of the US earnings season — Meta and Tesla report on 21 October — ahead of the Fed on 28 October and the RBA on 3 November.