Goldman Sachs Warns: $5.3 Trillion AI Capex Pushing Credit Markets Toward Saturation
Alina Collins
Goldman Sachs projects hyperscalers will spend a cumulative $5.3 trillion on AI and data centers from 2025 to 2030, warning that credit markets may hit saturation — while enterprises are already capping AI spending, squeezing this credit-driven supercycle from both ends.
Where does $5.3 trillion come from?
Goldman estimates hyperscaler AI and data-center capex will reach $5.3 trillion cumulatively over 2025–2030. This means → it is not a one-off bet but a six-year financing marathon spanning multiple capital markets.
Morgan Stanley breaks the numbers down further: data-center construction alone needs nearly $2.9 trillion by 2028. Of that, $1.4 trillion is internal cash flow, $200 billion corporate bonds, $150 billion asset-backed securities, roughly $800 billion private credit and JV debt, and about $350 billion from other capital sources.
In plain terms = the cloud giants' own cash covers less than half. The rest is borrowed. AI infrastructure investment is substantially credit-driven.
Why would credit markets "saturate"?
A data center is not a single asset — it bundles land, power access, network links, buildings, cooling, and AI servers. Financing needs spill into infrastructure funds, real-estate funds, private credit, and corporate bonds across multiple markets.
Goldman warns that a handful of hyperscalers cannot issue unlimited public debt. Investors are already flagging issuer-concentration risk. In plain terms = the borrowers are the same few companies, and the market fears all the eggs are in one basket.
Goldman analysts add that AI capex is growing faster than actual data-center buildout. Future bottlenecks may shift from model demand to financing capacity, power supply, and project execution. This means → if a systemic correction hits, the chain of losses will be far more complex than during the dot-com bubble.
Why are enterprise users hitting the brakes?
Uber burned through its entire 2026 AI budget in one quarter, then capped employee spending on a single AI tool at $1,500 per month in tokens. President Andrew Macdonald said: "It's getting harder to justify AI token spending."
Walmart capped token usage on its internal AI assistant. Global CTO Suresh Kumar said usage on its Code Puppy coding platform "spiked dramatically" and it was time to "step back and reassess." After Anthropic switched to per-token billing, software firm Workato saw daily costs surge sevenfold — CIO Carter Busse called it "we created a monster."
This reflects a core tension on the enterprise side: AI usage is exploding, but the causal link between spending and actual product improvement is growing blurrier.
Will upstream valuations take a hit?
Anthropic and OpenAI both plan to go public later this year at valuations approaching one trillion dollars. But the trend of enterprises cutting AI spending is constraining revenue-growth expectations for both companies.
Major AI platforms are steering users toward cheaper, non-frontier models. GitHub COO Kyle Daigle said Microsoft has proactively discussed pricing changes with clients, stressing that "not every task needs a frontier model." Microsoft, Amazon, and Google have also rolled out tools that automatically route requests to the most cost-efficient model.
Some enterprises are turning to open-source models run on local servers or personal devices to reduce payments to AI labs and cloud providers. This means → upstream pricing power is being eroded bit by bit by enterprise cost-cutting.
What is the core risk of this supercycle?
NYU emeritus professor Gary Marcus called Goldman's language "a terrifying sentence," warning: "Hyperscalers are not getting their $5.3 trillion back unless they extract it from taxpayers via massive government subsidies."
Deloitte's global generative-AI lead Costi Perricos noted: "Compute costs are now on the radar of CFOs and boardrooms. Consumers and enterprises have been told AI is cheap or free — it is anything but." OpenAI CEO Sam Altman acknowledged this month that cost has become a "major issue" for customers this year — a topic barely mentioned last year.
Put simply = the investment side is charging ahead on credit while the demand side is hitting the brakes — whether $5.3 trillion in credit-driven investment can find enough willingness to pay to sustain returns is the central test of whether this cycle lands softly.
Content is for reference only, not financial advice.