10-minute read
The bill for AI’s infrastructure buildout keeps growing, but turning that investment into profits will take time. With borrowing costs still high, investors are questioning whether the industry can sustain its spending while waiting for those profits to grow. Those concerns have put AI-related stocks under pressure in recent weeks.
In this month's CIO Insights, we examine these concerns, what they mean for AI's longer-term prospects, and how investors can position for the opportunities ahead.
Key takeaways
- AI's longer-term growth opportunity remains substantial, though monetisation may take longer to materialise. Slower revenue growth at frontier AI companies and competition from lower-cost models have raised concerns about the industry's outlook. Yet AI usage continues to grow, even as falling costs put pressure on prices. Those same cost declines could also open up opportunities beyond coding, including finance, healthcare, robotics and personal agents. Developing these applications safely could take time, extending the gap between technological breakthroughs and commercial returns.
- Financing pressures could slow parts of the AI buildout, but stronger companies have room to keep investing. AI infrastructure spending is set to grow significantly, putting free cash flow under pressure as borrowing costs remain elevated. Major hyperscalers have substantial cash reserves and established businesses to support their investments, while some debt-dependent infrastructure providers face greater risks. Tighter financing conditions could therefore delay some projects and shift investment towards financially stronger companies, rather than derail the AI buildout altogether. Slower investment could also weigh on economic growth and put downward pressure on interest rates, helping ease funding pressures over time.
- Diversification can help investors benefit from AI’s long-term potential while managing the risks. We remain constructive on AI's growth prospects, though financing pressures could bring market volatility in the near-term. Broad US or global equity portfolios offer exposure to AI alongside other sectors, which can help cushion a decline if investors rotate away from AI stocks. A broad portfolio also provides exposure to businesses that can use AI to improve productivity over time, not just the ones building it. Investors should review their existing AI exposure, ensure they have sufficient diversification across other investments, and maintain an allocation they can stick with through periods of uncertainty.
(See our Glossary at the end for a breakdown of the terms used in this article.)
AI has further room to grow, but monetisation will take time
AI is already generating substantial revenues across the value chain, from chips and data centres to power supply and software. On the application side, coding agents have been an early commercial success. As of end-September, Anthropic’s annualised revenue was roughly US$70 billion1, while OpenAI’s approached US$50 billion2.
But investors are starting to question how quickly that growth can continue. Estimates suggest Anthropic’s monthly revenue growth slowed sharply in August and September – although part of the slowdown may be temporary, reflecting new restrictions on model distillation (where a more advanced model’s responses are used to train another model). Competition from cheaper Chinese frontier and other open-source models is also putting pressure on prices, making it harder to translate growing usage into revenues.
New applications bring further challenges. As AI expands into healthcare, finance, and robotics, mistakes can have more serious consequences, requiring stronger safeguards. Frontier labs are already placing limits on how their most capable models are used. Anthropic restricts advanced cybersecurity and biology capabilities to vetted organisations. OpenAI has strengthened automated reviews after its internal coding agents weakened safety monitoring, and is developing requirements to halt workloads that lack an adequate safety case.
These (very necessary!) safeguards would delay commercialisation, lengthening the gap between technological breakthroughs and returns on investment – two components of the AI investment flywheel we introduced in our 2026 Macro Outlook: Just the FACTs (see Exhibit 1 below).
Companies may therefore continue to commit substantial capital to AI infrastructure while waiting for new applications to generate returns. That gap could leave investors more sensitive to slower revenue growth or delayed product launches, triggering sharper market swings. So while AI could eventually serve a much wider range of customers and industries, revenues may arrive less smoothly than investors expect.
AI’s opportunity extends beyond coding
While these headwinds could weigh on revenue growth, slower growth doesn’t necessarily mean demand is weakening. As Exhibit 2 shows, businesses tracked by Ramp used 135% more Anthropic and OpenAI tokens between end-June and mid-October, while their spending rose 56%. In other words, businesses are using considerably more AI, even as they pay less per unit.
Those same cost declines could also make a wider range of applications commercially viable – from financial analysis and drug discovery to shopping agents and robotics. Exhibit 3 illustrates the scale of the potential opportunity across these areas. Businesses have a strong incentive to adopt these tools: McKinsey estimates that applying AI to existing work could generate around US$2.5 trillion in annual productivity gains globally.
These applications could give AI companies new sources of growth, building on their early successes in coding. Still, it will take time to build products that work reliably and for businesses to adopt them. The recent slowdown in revenue growth alone therefore doesn’t mean that AI is running out of commercial opportunities.
The longer-term question is how quickly these applications gain adoption, and how much of the value AI companies can capture. The more immediate question: can the companies funding the AI buildout sustain that spending while they wait for their investments to generate returns?
Why financing pressures may not derail the AI buildout
For now, the largest tech companies appear well-positioned to fund their expansion. But the amount of investment required continues to grow. Consensus estimates put combined capital expenditure by Amazon, Microsoft, Alphabet, Meta, Oracle and the major neocloud companies (specialist providers that rent out AI computing power) at around US$860 billion in 2026.
That figure is expected to rise to US$1.2 trillion in 2027 and more than US$1.3 trillion a year from 2028 through 2030, which adds up to more than US$5 trillion over the coming four years. Much of that spending must happen before the new infrastructure starts earning revenue, making the timing of AI’s commercial returns increasingly important.
All that spending is putting pressure on cash flows. As Exhibit 4 shows, consensus estimates suggest these companies’ combined free cash flow (FCF) – or the cash left from operations after capital spending – will turn negative in 2026 and remain negative in 2027. Analysts then expect a recovery in 2028 as spending growth moderates and revenues catch up.
That said, not all companies face the same pressure. Microsoft is forecast to maintain positive free cash flow over the coming years. But even those with expected shortfalls, such as Alphabet and Amazon, continue to bring in substantial cash from their established businesses to fund their investments.
They also have sizable cash reserves that can help cover the remaining gap. As Exhibit 5 shows, Alphabet holds around US$120 billion in cash after accounting for short-term debt, compared with its forecast FCF shortfall of around US$30 billion in 2027, while Amazon has a buffer of roughly US$105 billion. Alongside the option to reduce share buybacks or raise additional financing, these reserves give them room to sustain investment while waiting for revenues to catch up.
The neoclouds have much less to fall back on. IREN and Nebius each have net cash buffers of less than US$10 billion, while CoreWeave’s short-term debt already exceeds its cash holdings. Their expansion therefore depends more heavily on continued access to financing, leaving them exposed if lenders become cautious or commercial returns take longer to materialise.
Higher borrowing costs hit some AI companies harder than others
The challenge is that borrowing is already becoming more expensive. The US 10-year Treasury yield, a benchmark for borrowing costs, stands at around 5% at the time of writing, up from roughly 4% a year earlier. (For more on bond yields, see CIO Insights: Three forces behind higher bond yields.)
Meanwhile, companies with larger debt loads and more uncertain business models face greater financing risks. As Exhibit 6 shows, over the past 12 months, implied credit default swap (CDS) spreads, an indicator of perceived credit risk, have widened much more sharply for some of these neoclouds than for more established hyperscalers.
Over the past 12 months, spreads rose by more than 100 basis points (bps) for IREN and CoreWeave, compared with 19 bps for Alphabet, 6 bps for Microsoft and 4 bps for Amazon. Negative free cash flow and smaller cash buffers leave these firms more dependent on external funding, making further expansion harder to finance amid higher borrowing costs.
These concerns are also evident in the recent selloff in CoreWeave’s bonds. As a refresher, the company borrows to buy chips and build computing capacity, which it then rents to AI customers. That means spending heavily upfront and collecting revenues over time; the company has taken on around US$50 billion in debt to fund its spending3.
The selloff pushed yields on some of its bonds above 15%, reflecting investors’ concerns about its financial outlook. If CoreWeave sought to raise similar debt today, it would face substantially higher borrowing costs, making further growth harder to finance.
This creates another potential constraint on the AI investment flywheel. If revenues take longer to materialise while borrowing costs remain elevated, more indebted companies could be forced to slow expansion or consolidate.
Still, we believe demand for AI computing could remain strong through that adjustment, and financially stronger hyperscalers could benefit from reduced competition and take on some of that business. Financing pressures could therefore delay some projects and shift more of the buildout towards companies better able to fund it.
Could slower investment ease financing pressures?
Any slowdown in the investment cycle could also help ease financing pressures. If tighter financing curbs AI spending and weighs on economic growth, that could put downward pressure on Treasury yields, helping reduce borrowing costs and ease funding pressures.
To be sure, wider credit spreads could offset any relief for riskier borrowers. AI investment is also only one factor influencing interest rates. Persistent inflation and large US fiscal deficits could keep yields elevated, while the US midterm elections could shape tax and spending decisions that might further impact the trajectory of bond yields.
How to invest in AI’s potential while managing the risks
We remain constructive on AI’s long-term prospects, though financing pressures could make the AI investment cycle more uneven. Growing usage and the scope for applications beyond coding suggest there is still considerable room for adoption. But while those opportunities will take time to develop, the infrastructure for those AI applications needs funding now. That gap could mean periods of slower spending and greater market volatility, even as the technology continues to advance.
For investors, broad US or global equity portfolios offer a way to benefit from AI’s long-term growth while spreading investments across sectors. When AI stocks come under pressure, money may rotate towards other parts of the market – which can help cushion the impact for a diversified portfolio. A broad portfolio also gives you exposure to businesses that can use AI to improve productivity over time, not just the ones building it.
If you’re already invested, review how much AI exposure you hold through equities, and whether your other investments provide enough diversification. Build an allocation that’s in line with your goals – and that you can stick with through periods of uncertainty – and give AI’s longer-term opportunities time to develop.
Authors
Stephanie Leung, Chief Investment Officer
Stephanie and her team oversee the full spectrum of investment products and portfolios offered at StashAway. She brings more than two decades of investment expertise across multiple asset classes. Prior to joining StashAway in 2020, she managed investment portfolios at institutions such as Goldman Sachs and multi-billion dollar family offices in the region.
Justin Jimenez, Head of Macro and Investment Research
Justin brings 15 years of experience in economic and investment research to StashAway, where he contributes to shaping the investment office’s views on the global economy and financial markets. Before joining StashAway in 2022, he was an economist at Bloomberg, and holds degrees in international economics and finance from Columbia and UCLA.
Glossary
Hyperscaler
A large technology company that operates extensive data centre infrastructure to power its own services or provide cloud computing services to customers. Examples include Amazon, Microsoft, Alphabet and Meta.
Neoclouds
Specialist cloud computing companies that rent out AI computing power, typically using high-performance chips. Examples include CoreWeave, Nebius and IREN.
Frontier companies
The companies developing the most advanced AI models currently, like Anthropic and OpenAI. Also known as frontier labs.
Capital expenditure (capex)
The money a company spends on long-term physical assets, such as data centres, computer and network hardware, and power infrastructure.
Free cash flow (FCF)
The cash a company generates from its operations after paying for capital expenditure. It shows how much cash remains available to repay debt, invest or return to shareholders.
Credit default swap (CDS) spread
The cost of insuring against a borrower defaulting on its debt. Higher spreads generally indicate that investors perceive greater credit risk.
AI value chain
The different layers of business involved in AI, from chipmakers and data centres to power suppliers, software developers, and the businesses that use AI.
Token
The basic unit of text an AI model processes. AI providers typically charge per token, so token usage is a common measure of how much is being used.
References
- Lipschultz, B., Mascarenhas, N., Torrence, R., & Ghaffary, S. (2026). Anthropic Revenue Jumps to $16.8 Billion in Quarter Before IPO. Bloomberg. Retrieved from: https://www.bloomberg.com/news/articles/2026-10-14/anthropic-revenue-jumps-to-16-8-billion-in-quarter-before-ipo
- Hammond, G., & Morris, S. (2026). OpenAI annualised revenues $20bn less than previously signalled. Financial Times. Retrieved from: https://www.ft.com/content/b66a9858-f8fb-46cb-b506-44bfe26fca2a
- Soni, A., Rooprai, A., & Matthews, L. (2026). CoreWeave’s slump spotlights growing burden of debt behind AI boom. Reuters. Retrieved from: https://www.reuters.com/legal/transactional/coreweave-shares-bonds-slide-further-ai-debt-fears-mount-2026-10-15/
