SRRF评论：不劳而获，但芯片会免费吗？

Will AI capital expenditure deliver supercharged profits and enhanced productivity gains? Or will increasingly debt-funded hyperscalers over-extend themselves causing compute costs to plummet?

Global equities rebounded over 2% in local currency terms over the month, but under the surface technology companies saw a strong recovery. The US technology sector rebounded over 6% over the month, while the NASDAQ and the Japanese TOPIX rallied 4%. Bond yields did a round trip, falling on expectations that inflation was falling, only to rise after a stronger than expected Consumer Price Index (CPI) print and more hawkish comments from Warsh at Jackson Hole. US bonds sold off the least, with 10-year yields only moving higher by 2 basis points over the month, whereas Australian bonds suffered the most, with 10-year yields rising 17 basis points over the month after higher-than-expected inflation and a more hawkish RBA press conference. Credit spreads were mostly subdued, with Australian and emerging market investment grade credit outperforming, along with US high yield which saw spreads come in over 20 basis points. In currencies, the Australian dollar rallied over 2% given the hawkish tilt whereas the Japanese yen sold off after combined BoJ and US intervention. Gold rallied almost 10% over the month, with broad commodities up over 7% in USD terms.

We’ve been vocal in our view that we have entered a higher inflation regime, where inflation is likely to be above central bank targets for most of the 2020s. Driven by persistent fiscal largesse, growing populism, protectionist industrial policy, and national security imperatives, central bank inflation targets look increasingly out of reach. In this supply-constrained environment, a massive productivity breakthrough would be required to help return to a more benign inflation environment. As hyperscalers pour hundreds of billions into digital infrastructure, clients are asking whether AI will lead to a transformative 1990s-style productivity boom, or a 2000s-style over-hyped unprofitable technology crash?

Hyperscalers are expected to spend almost US$800bn this year on AI capital expenditure (capex), growing to over US$1tn p.a. to the end of the decade. This means capex is likely to grow from 1.5% of US GDP, to potentially upwards of 2-3% of GDP. These companies have gone from asset light free-cash generating machines of the past decade, to increasingly negative free-cash flow asset heavy businesses of tomorrow. But, underpinning this capex supercycle is an ever-growing backlog of cloud demand, with future locked in contractual revenues rising over US$300bn over the last quarter alone, driven by the insatiable appetite for compute from the AI model providers.

Chart 1: Hyperscaler capex is expected to grow over US$1 trillion per year to 2030

However, the AI model providers, like OpenAI and Anthropic, do not have to pay for the future contractual obligations until the infrastructure is built and the compute is used. This timing mismatch has seen hyperscalers turn increasingly towards debt to fund the gap. The shift away from using free cashflow to fund the AI buildout towards debt financing increases the vulnerability of these companies if AI turns out to be less profitable than expected. While the overall level of debt is not yet concerning, given the strong balance sheets of these companies, it is a worrying sign. More controversial is the amount of leasing agreements, which is essentially off-balance sheet debt. If we include this the debt in the system is almost double what is being reported. Leverage has joined the party.

Chart 2: Capex is being funded more and more by corporate debt and off-balance sheet financing

Leverage by itself is not an omen of a bubble, or the sign one is about to pop, but it is the start of late cycle corporate behaviour. These companies still enjoy very strong revenue from their back book of business, and even though spreads on hyperscalers have widened given the influx of issuance, overall credit spreads remain extremely tight. Free cash flow from operations and cheap debt financing has not hindered their ability to increase capex markedly. Current cloud revenue growth has been very strong, which bodes well for the hyperscalers. Google saw cloud revenue grow at an unprecedented 81% year-on-year. We’re also seeing increased revenue from some of the AI model providers, with Anthropic’s revenue jumping from sub-US$10bn by the end of 2025, to around US$65bn today.

Chart 3: But AI investment is already translating into revenue

Even when combining Anthropic’s US$65 billion run rate with OpenAI's US$40 billion, the model creators cannot single-handedly fund US$1 trillion in hyperscaler obligations. Will they be able to pay the bill when it comes due? This is one of our biggest concerns. While the rapid increase in revenue from AI model providers is impressive, bridging the financial gap still requires a massive leap of faith, the standard precursor to any bubble. If cash flows disappoint, the market expects these labs to seek capital injections, new funding rounds, launch public offerings, or lean into sovereign partnerships (such as OpenAI's proposal to trade a 5% stake for government backing), assuming they can get a good valuation. There is also the chance that another model provider creates a breakthrough and starts taking market share and utilises the future compute capacity, just look at the chart below to see how fast Claude has increased its market share over the past year. But it’s not just US models that use hyperscaler compute, a domestic enterprise shifting to a Chinese open-weight model still has to run that model on a Western cloud server, which could eat away at any excess compute capacity.

Unfortunately, that’s a lot of maybes, as right now it is hard to predict or model the future given the breakneck speed this technology is evolving. In the worst case, AI capex gets pulled back, which crushes the downstream chip providers, the hyperscalers are left with a surplus of compute capacity with revenues that did not materialise, causing compute cost to plummet. Maybe that’s when we get our productivity boost, when the price of compute dwindles. Even without that dire scenario, token expenditure is already falling. Does this portend improved productivity due to falling AI costs, or weaker future earnings due to lower pricing power?

Chart 4: Adoption rates continue to climb, but token expenditure is falling

Although there are some worrying signs starting to appear, a hard crash is not our base case. Enterprise AI adoption continues to scale, traffic across top-tier models is resilient, and fresh contenders are regularly entering the arena, all while cloud infrastructure and frontier model providers report accelerating revenue. Furthermore, US corporate earnings are improving broadly across most sectors, proving this economic resilience extends beyond a handful of AI-linked names. Equity markets have already penalized hyperscalers for their historic capital spending splurge, compressing their valuations to decade lows. While it is far too early to declare an end to the AI investment cycle, the current environment demands a close watch to identify exactly when the AI capex tide might turn.

After a tumultuous July, where we saw technology companies sell off hard, we used the washout in positioning to rebuild positions in the US and technology companies in general. Our equity weight increased from 35% to 37.4% delta-adjusted, predominantly from an increase in US equities, including a 1% addition to both US technology sector and the Nasdaq via futures. We also sold our position in US utilities in favour of energy. We took profit on our 1% allocation to Latam equities given volatility is likely around the upcoming presidential election. Investor positioning in US equities went from a high of over 2 standard deviations over-owned, to below zero standard deviations by the end of July, which coincided with the now infamous Situational Awareness collapse. US growth remains robust, US PMIs are strong and our US recession probability model continues to drop even further. To us the risk remains higher inflation, either from an escalation of war or a more dovish Federal Reserve, rather than a growth downturn.

The fundamental backdrop remains strong. Our proprietary US EPS model points to circa 21% growth over the next 12 months, with ex-Mag 7 growth also improving. Earnings season has been robust, with broad-based beats, positive guidance revisions and margin expansion. The median stock continues to see improving earnings growth, and after adjusting for one-off private investment revaluations in Amazon and Alphabet, the rest of the index outgrew the Mag 7 for the first time since late 2022. We are also seeing more attractive valuations in the US, with most of the Mag 7 enjoying forward PE ratios not seen in the last 10-years. We continue to hold 2% in the Nasdaq and 1% in the US technology sector. We expect technology to continue rebounding after the technical washout before leadership potentially rotates back towards the broadening-out trade, where more domestic economy linked companies should benefit from the strong economy. We therefore expect the US equal-weight trade to be preferred in the coming months but tactically are long technology. We remain long 4% in the S&P equal weight as a more medium-term view.

To protect ourselves from a resumption of hostilities in Iran or any volatility around the US mid-term elections, we initiated a 5% notional put-spread on the S&P 500, expiring in late November, which will protect the fund for falls between 4% and 13% from current levels.

Credit spreads continue to be tight, but we do not expect a significant widening event or increase in defaults in the near future, so prefer to take carry given overall attractive all-in yields. US investment grade has underperformed given the abundance of hyperscaler issuance over the year. With this issuance now moving offshore, including new issues from Google in Australia, issuance headwinds may start to fade on the margin. Strong US earnings and economic backdrop suggest we can take carry, despite stretched valuations. Europe also benefits from an improved macro backdrop and improved corporate fundamentals. US securitised spreads have come in and are no longer as attractive, with US mortgage-backed security (MBS) spreads in the 11th percentile. Higher UST rates volatility can lengthen MBS assets making them more vulnerable. We continue to prefer Australian corporates given their wider spreads and higher quality, although hyperscaler issuance hitting the Australian market could keep spreads from contracting. This month we added 2% to US high yield and 2% to Australian BBB corporates.

We reduced our duration substantially after duration rallied at the start of the month, cutting our overall duration by 0.5 years to end the month at 2.0 years at the portfolio level. The majority of the cut came from selling our positions in the front-end of Australia. In July we added modestly to short-end Australian duration (in the two-to-three-year maturity range) on the view that the RBA was done tightening and that front-end yields would rally as markets priced an extended pause, with the next move a cut. With yields on Australian 3-year bonds coming in over 20 basis points from the high in July to levels in early-August, we used that as an opportunity to take profit before the RBA and before the CPI print, both which came in more hawkish and hotter than anticipated. We added back slightly to Australian duration towards the end of the month after yields repriced higher. We also reduced our exposure to front-end German bunds by 0.2 years and shifted 0.1 years into both UK and Japanese bonds. Our bond fair value models show most 10-year government bonds have moved from overvalued to neutral, with our cyclical models moving from neutral to undervalued, particularly in Germany, the UK and Japan.

We also made some changes to the US duration positioning. Firstly, we increased US duration exposure mid-month as weaker data, particularly lower inflation, reduced our expectations for a Fed tightening prior to the mid-terms in early November. Secondly, we have been running a yield curve steepening position for the past and reduced it by following the US Treasury announcement to increase off-the-run buy backs saw the curve reverse its steepening trend. We added back to the steeper at month end following Warsh’s more hawkish speech at the Jackson Hole conference which added some value back into shorter maturities as fed rate hike probabilities for September moves above 70%. We continue to hold US 2-year break-evens, which saw break-evens improve 17 basis points from the end of July.

We did not adjust our currency positions this month. While we continue to expect US growth to outpace other developed markets, inflation momentum appears to be stalling out, while the rates market seems reasonably priced with 1-2 hikes priced in over a year. From a rates differential perspective, we no longer see much upside to the USD, while positioning and mean reversion indicators are broadly neutral for the US dollar as well. However, we still like the USD from a portfolio hedging perspective and continue to hold around 8% in the portfolio. We have been moderately increasing our position in the Euro given the positive surprise in economic data for the region over the last few months and maintain a position of 4%. While the absolute level of growth remains low in the region, growth has surprised to the upside against consensus and proven to be more resilient than expectations. There remain structural challenges, but we do not see this impacting the Euro over the shorter term. We continue to hold a -2% position in the GBP given its poor longer-term outlook with restrictive fiscal policy and benign growth. While valuations for the JPY remain attractive, we do not see a strong catalyst for the Yen to rally and have no exposure. We have upgraded our view on the AUD given the potential for a more hawkish RBA but keep our foreign currency unchanged to help hedge against our increase in risk through both an increase in equities and credit over the month. We continue to hold 2% in gold and 3% in broad commodities.

Learn more about investing in Schroder Real Return Fund or Schroder Real Return Active ETF (ASX Code: GROW).

This document is issued by Schroder Investment Management Australia Limited (ABN 22 000 443 274, AFSL 226473) (Schroders). It is intended solely for wholesale clients (as defined under the Corporations Act 2001 (Cth)) and is not suitable for distribution to retail clients. This document does not contain and should not be taken as containing any financial product advice or financial product recommendations. This document does not take into consideration any recipient’s objectives, financial situation or needs. Before making any decision relating to a Schroders fund, you should obtain and read a copy of the product disclosure statement available at www.schroders.com.au or other relevant disclosure document for that fund and consider the appropriateness of the fund to your objectives, financial situation and needs. You should also refer to the target market determination for the fund at www.schroders.com.au. All investments carry risk, and the repayment of capital and performance in any of the funds named in this document are not guaranteed by Schroders or any company in the Schroders Group. The material contained in this document is not intended to provide, and should not be relied on for accounting, legal or tax advice. Schroders does not give any warranty as to the accuracy, reliability or completeness of information which is contained in this document. To the maximum extent permitted by law, Schroders, every company in the Schroders plc group, and their respective directors, officers, employees, consultants and agents exclude all liability (however arising) for any direct or indirect loss or damage that may be suffered by the recipient or any other person in connection with this document. Opinions, estimates and projections contained in this document reflect the opinions of the authors as at the date of this document and are subject to change without notice. “Forward-looking” information, such as forecasts or projections, are not guarantees of any future performance and there is no assurance that any forecast or projection will be realised. Past performance is not a reliable indicator of future performance. All references to securities, sectors, regions and/or countries are made for illustrative purposes only and are not to be construed as recommendations to buy, sell or hold. Telephone calls and other electronic communications with Schroders representatives may be recorded.