NextFin News - Asian software and platform stocks have vaulted from near the bottom of the AI preference pile to the region's most attractive AI trade in just one month, after a quarter of investors in Bank of America's latest Asia fund manager survey named the category their top AI pick for the next 12 months. The shift, detailed in a Sept. 15 note by BofA strategists including Kaspar Lam, marks the fastest theme rotation in the AI trade this year - and it signals that Asia's investors have moved into the "show me" phase of artificial intelligence, where revenue conversion matters more than capital expenditure announcements.
Twenty-five percent of respondents picked software and platforms as their top AI trade in Asia, the highest share among choices that included memory chips and data centers. In August, the same theme ranked near the bottom, ahead of only AI compute. That is a move from last to first in roughly 30 days, a velocity that reflects less a change in the technology than a change in what investors are now asking AI companies to prove.
The Pivot in a Month
The rotation fits a broader pattern visible across Bank of America's survey work this summer, and it is best understood as a migration from the capex ledger to the revenue ledger. In the August Asia survey, 64% of fund managers said they needed clearer evidence of AI monetization and actual revenue generation before adding to AI-linked equity exposure - far ahead of other conviction triggers such as a re-acceleration in hyperscaler capital expenditure, which drew 18%, or stronger earnings revisions at 14%. Interest rate cuts scored zero. Money is no longer being deployed on the promise of infrastructure buildout; it is being held back until the revenue shows up on the income statement.
At the same time, hedging activity surged. A combined 59% of Asian respondents in August said they were actively protecting against downside in the AI trade over the next six to 12 months by rotating into value, cyclical and defensive sectors - more than double the 28% recorded in July. The camp of investors remaining unhedged and purely overweight AI collapsed from 28% to 9% in a single month. Capital moved into domestic cash generators: retail and e-commerce recorded the largest month-on-month allocation jump at 27 percentage points, followed by utilities, up 19 points, energy, up 17, and consumer staples and banks, each up 15. Healthcare and pharmaceuticals gained 12 points, media and entertainment 10, and telecom 9.
The backdrop for this rotation is a market that has run out of buyers. In the August global fund manager survey, average cash holdings fell to 3.5% of assets - the sixth-lowest reading since the survey began in 1998 - while global equity allocation climbed to a net 56% overweight, the highest since November 2021. Bank of America's Bull & Bear indicator reached 9.3, above the 8.0 level that historically triggers a contrarian sell signal. Michael Hartnett, the bank's chief investment strategist, put the implication plainly in the note:
"Current positioning does not recommend further buying of risk assets; it is a phase where one should either withdraw for now or seek tactical rotation within risk assets."
The software-and-platform pivot is exactly that kind of tactical rotation: staying inside the AI theme while moving to its most cash-generative layer. It is also a textbook example of what happens when a crowded trade runs into a positioning wall. With cash at multi-decade lows and equity exposure at a five-year high, the marginal buyer for hardware names had already been exhausted; the rotation had to go somewhere, and it went to the part of the AI stack that had been left behind.
Why Software, Not Silicon
The mechanism behind the rotation is straightforward once you follow the cash. For the first phase of the AI buildout, the bottlenecks - and therefore the profits - sat in hardware: graphics processors, high-bandwidth memory, advanced packaging, data center construction, power and cooling. Fund managers piled into the markets whose export cycles are tied to the semiconductor supply chain. Taiwan and Japan topped preference rankings in the August Asia read, with Japan holding its position as the most-preferred country, backed by a 50% net overweight, on the strength of semiconductor manufacturing equipment and precision components. Semiconductors were the most preferred sector in Asia ex-Japan, with a net 50% of fund managers overweight.
But hardware is a cyclical, capex-led trade. It pays off while hyperscalers are ordering, and it punishes investors the moment orders slow. Software and platforms sit at the point where AI usage is actually monetized. When a company pays for AI-powered software seats, cloud API calls or platform subscriptions, that revenue is recognizable and recurring - the very proof that 64% of Asian managers said they were waiting for. The rotation is not a rejection of AI; it is a migration from the part of the trade that depends on someone else's budget approval to the part that books the invoice.
There is also a positioning logic. By late summer, institutional ownership of large-cap memory and storage names had reached elevated levels relative to the broader market, while ownership of software names remained comparatively light. A crowded trade has less room to run; an under-owned one has room to be re-rated. When a quarter of survey respondents suddenly point in the same direction, the price impact is amplified by how few were positioned there before.
In August, BofA's read of the AI value chain placed power and energy, data-center infrastructure and connectivity ahead of memory, with software and platforms sitting above only AI compute. One month later, software and platforms lead. The infrastructure layer has not become less important - the power and data-center buildout continues - but the marginal dollar is asking a different question: not "who is building the pipes?" but "who is collecting the tolls?"
What Changed in 30 Days
Three catalysts converged to flip the ranking. First, the semiconductor cycle itself showed signs of fatigue. Regional optimism about the South Korea-Taiwan semiconductor export cycle fell to 27% in August from 60% in July - one of the sharpest one-month reversals in the theme this year. Second, the September global survey showed investors turning most bearish since June, rotating out of equities, Japan and healthcare into cash, the dollar and telecoms; global equity allocation fell to a net 13% overweight from 37% the prior month, the lowest since July 2025. In a de-risking environment, software platforms with contracted, recurring revenue look more defensive than hardware suppliers dependent on the next capex order. Third, and most important, the market's AI question changed: not "who is building the infrastructure?" but "who is actually getting paid for AI?"
The speed of the reversal also reflects how fast consensus can exhaust itself. A theme that goes from under-owned to consensus favorite in a month has done the work of a normal rotation in a fraction of the time - which means it can also unwind quickly if the revenue evidence fails to arrive. The same survey machinery that crowned software in September was showing record bullishness a month earlier: the contrast between the August global reading and the September global reading is itself the clearest evidence that positioning, not fundamentals, is driving the short-term moves.
There is a second-order implication that most commentary is missing. The rotation changes the discount rate the market applies to AI winners. In the capex phase, the winners were priced on unit growth - how many chips, how many racks, how many megawatts. In the monetization phase, the winners are priced on revenue durability - contract length, retention, gross margin. That is a different valuation framework, and it favors companies with visible recurring revenue over companies with visible order books. The market is not simply rotating sectors; it is rotating the metric by which AI value is judged.
Cyclical or Structural?
This is the judgment the market has to get right, and the answer is not the same for both halves of the AI trade. The hardware leg - memory, data centers, power, networking - is cyclical. It is driven by a capex super-cycle that will eventually mean-revert as capacity catches up with demand. History offers analogs: the fiber-optic buildout of the late 1990s, the smartphone supply chain of the early 2010s, and the cloud infrastructure wave of 2017-2018 each rewarded suppliers early and punished them once orders normalized. The evidence that this leg is cyclical is already visible - semiconductor export optimism halving in a month, allocation to tech hardware cut by 20 percentage points, and a global equity allocation that swung 24 points in four weeks.
The software and platform leg is more structural - but only conditionally. A structural shift requires a durable change in how enterprises buy and use technology, and there is evidence for it: AI usage is moving from experimentation to embedded workflows, and recurring software revenue is the natural place for that monetization to appear. But the condition is that the revenue must actually materialize. If hyperscaler capital expenditure slows before enterprise AI spending takes off - a gap that opens whenever the discount rate on future AI earnings rises faster than current AI revenue - software platforms lose the infrastructure tailwind before their own monetization has scaled. That is the narrow path the trade has to walk.
The cleanest way to state the call: the hardware leg is a cyclical wave riding a structural shift; the software leg is the structural shift waiting to be confirmed by cyclical revenue data. Investors treating both as the same trade will get the timing wrong. The hardware names will see their earnings peak first; the software names will see their earnings inflect later, and the gap between those two inflection points is where this trade will be won or lost.
The Counter-Thesis
The strongest case against the rotation is that it is late-cycle chasing. Software and platform valuations had already rerated higher on AI expectations through 2025 and early 2026, and a move into them as global equity sentiment hits its most bearish reading since June looks less like a contrarian pivot and more like a crowded exit door. If the August global survey is any guide - with AI hyperscaler capex named the most likely source of a systemic credit event, and a disorderly rise in bond yields the second-largest fear at 27% - then a credit or rates shock would hit growth-sensitive software multiples first and hardest. The counter-thesis, in short: investors are rotating into the one part of the AI trade that is most exposed to the discount rate, just as the discount rate is becoming the market's central worry.
That argument has force, but it misses the asymmetry. Hardware multiples are supported by visible order books; software multiples are supported by visible revenue. In a world where the market's dominant question is monetization, visible revenue is the more durable support. The counter-thesis would be proven right if enterprise AI software revenue growth fails to accelerate over the next two quarters while hyperscaler capex continues to climb - that combination would show the software layer is not capturing the value its valuations assume. That is the specific divergence to watch in the coming earnings season, and it is a falsifiable test, not a vague warning.
There is a weaker version of the counter-thesis that deserves dismissal. Some will argue that 25% is not a majority, so the signal is weak. That misses the point of a plurality in a fragmented survey: the question is not whether software has majority support, but whether it has the highest share among the choices, and it does. A quarter of a large pool of Asian allocators is still a meaningful flow signal, especially when it comes from a base that was under-owned.
What to Watch
The forward path splits by horizon. In the short term, the rotation's momentum depends on the September and October earnings prints from Asian and global software platforms - any sign that AI is lifting billings will extend the trade. In the medium term, the key signal is hyperscaler capital expenditure guidance: a re-acceleration would pull attention back to the hardware layer, while a flattening would confirm the shift toward revenue-generating software. In the long term, the structural question is whether AI becomes a durable line item in enterprise software budgets or remains a feature bundled into existing contracts - the difference between a new revenue stream and a margin story.
Three scenarios frame the path. The base case is a bifurcated AI market in which hardware names consolidate as the capex cycle matures while platforms with demonstrable AI revenue continue to re-rate. The upside case is that software monetization accelerates faster than expected, validating the 25% plurality and pulling the category toward majority support. The downside case is a credit or rates shock that compresses software multiples before revenue can catch up, sending investors back to cash and defensives.
Specific signals to track: enterprise software billings growth in the next two earnings seasons, hyperscaler capex guidance for the coming fiscal year, and the spread between software and semiconductor performance in Asian indices. If billings accelerate while capex guidance flattens, the software trade extends. If capex re-accelerates while billings disappoint, the rotation reverses.
The Bottom Line
The AI trade is being repriced from "who is building it" to "who is getting paid for it." A quarter of Asian fund managers now see software and platforms as the best AI bet in the region - up from near last place a month ago - and the speed of that pivot says more about the market's demand for proof than about any change in the technology itself. The investors who get this cycle right will be those who recognize that AI's first act was about infrastructure, and its second act is about income statements.
Explore more exclusive insights at nextfin.ai.
