Man Institute Something Has Got to Give

AI has driven extreme dispersion between a soaring semiconductor sector and an under-pressure software industry. We expect this divergence to persist until the commercial returns from AI investment become clearer The hyperscalers are approaching a free cash flow inflection point, with AI capital expenditure now consuming over 90% of operating cash flows. Investors would welcome spending discipline, but it would be a direct headwind for semiconductor demand Semiconductor valuations are priced for perfection, leaving little margin for error. Upcoming AI IPOs should provide important clarity on the durability of AI spending plans Global equity markets remain at or near all-time highs, having shrugged off the Iran war and rising bond yields in the first half of the year. Most of the bullishness continues to be driven by optimism around AI. However, AI has also sparked extreme dispersion between a soaring semiconductor sector and an under-pressure software industry. As systematic investors, we’ll be closely watching how this plays out in the second half of the year. AI fervour has allowed the semiconductor complex to drive nearly two-thirds of year-to-date S&P 500 performance, and much of the 80% gain in South Korea’s KOSPI Index. On the other side, we are seeing the 'SaaSpocalypse' – the broad sell-off across software-as-a-service (SaaS) stocks – drag down the likes of Microsoft and the wider software sector, while most of the free cash flow (FCF) generated by the hyperscalers has disappeared. Our previous comments on the hyperscalers were based on the expectation that rising capital expenditure would offset gains in operating cash flow, and total FCF would fall to approximately US$200 billion in 2025 and 2026 before surging to new highs in 2027 and beyond. After the last two quarters of earnings reports, we can see that capital expenditure is now projected to consume more than 90% of operating cash flows, and we’ll have to wait until 2029 for a new high in this group’s FCF. Oracle, after more than 30 years of positive FCF, is now expected to have negative FCF for five consecutive years.

Source: Bloomberg, as at 27 May 2025. Within the S&P 500 and MSCI World, the cumulative weight of these five companies plus NVIDIA remains broadly unchanged at 26.3% and 18.6%, respectively. But there has been significant dispersion within this group, with NVIDIA the biggest beneficiary and Alphabet close behind based on optimism around their AI-monetisation and proprietary custom chip business. Meanwhile, Microsoft, Meta, and Oracle have struggled due to increased capital spending and other company-specific concerns. We will come back to the opportunity for the hyperscalers, but first let us compare the biggest winners and losers in this market to identify the risks on the other side of the ledger.

Source: Bloomberg and S&P 500, as at 2 May 2026. From a valuation perspective, the case for software is not obvious. The aggregate price-to-sales ratio is approximately seven times, or under six times excluding Microsoft and Palantir, which is below the levels seen over the past five years, but still meaningfully above longer-term historical levels. We strongly doubt the industry is going away, and there are likely some significant opportunities at present, but ongoing valuation compression is certainly possible until evidence of a growth recovery emerges.

Source: Bloomberg and S&P 500, as at 2 May 2026. However, the valuation story for semiconductors is a little clearer, and perhaps not in the most flattering way. The aggregate price-to-sales ratio for semiconductors traded between two and four times from 2005-2020, including or excluding NVIDIA. It rose sharply during the 2020 to 2021 bull run before coming back to four times in late 2022. Since then, that ratio has risen sharply to over 16 times (over 14 times excluding NVIDIA ). Clearly, this is pricing in unprecedented future growth for semiconductors, but would seem to leave little to no margin for error when the AI infrastructure build-out slows down. Of course, we do not know when that will happen, and if it happens in 2030 or beyond, these could be entirely reasonable valuations. But it does feel as though the semiconductor complex might be priced for perfection and will need some time to grow into these valuations.

Source: Bloomberg and S&P 500, as at 2 May 2026.

So let us go back to the opportunity for the hyperscalers – which also represents the key risk for the semiconductor sector. For most of the last 15 years, the likes of Microsoft, Alphabet and Meta were extraordinary cash flow generating machines that were also asset-light and could self-fund their growth, generating exceptional returns for shareholders. AI-related capital expenditure has grown sharply over the last few years and has continued to exceed expectations but we may be approaching a significant turning point. We have arrived at a point where, if capital spending exceeds current expectations, the hyperscalers could move into significantly negative free cashflow. For now, we believe that is an unlikely outcome. As an example, in valuation terms, Oracle is well below its highs as the market digests its dilutive capital-raising requirements. We may also gain more insight into the potential for AI spending through the expected IPOs of SpaceX, OpenAI, and possibly Anthropic. But with global interest rates at local highs, the opportunity cost of capital is real, and we believe spending discipline would be well received by investors. However, that could cause a reckoning in semiconductor valuations. Put another way, we believe that the pronounced dispersion we have witnessed is likely to persist as we all await what the second chapter of the AI boom will bring.