Record profits at TSMC and ASML, but Gartner warns $3.5 trillion of AI spending is at risk
The companies that make AI chips have never earned more. A new forecast says the businesses buying all that hardware may struggle to earn it back.
This week gave two very different pictures of the AI boom. On one side, the companies that make AI chips reported their best results ever. On the other, research firm Gartner warned that much of the money being spent on AI data centres may never earn a proper return.
Both things can be true at once, and understanding why helps explain where the AI industry is heading.
The chipmakers are booming
TSMC, the Taiwanese company that manufactures chips for Nvidia, Apple and many others, reported third-quarter revenue of about US$46.7 billion, up roughly 51% from a year earlier. Net income rose about 69%. Advanced chips made with its newest processes, 7 nanometres and smaller, made up 79% of its wafer revenue. TSMC expects next quarter’s revenue to be even higher, between US$52.3 billion and US$54.3 billion.
Dutch company ASML, which makes the machines needed to produce advanced chips, also had a record quarter: €11.7 billion in sales and €5 billion in net income. Its chief executive, Christophe Fouquet, said some customers are already discussing orders for 2029 and even 2030.
Gartner’s warning
Gartner looked at the other side of the deal: the companies buying all this hardware. Its forecast, published on October 14, includes some striking numbers:
- AI providers are expected to spend nearly $1 trillion on AI infrastructure in 2026, and about $6 trillion more from 2027 to 2030.
- To cover those costs and earn a reasonable return, data-centre owners would need about $8.5 trillion in revenue by 2030.
- Gartner expects AI revenue linked to those data centres to be no more than $3.6 trillion.
- As a result, more than $3.5 trillion of spending could face “impairment”, meaning its value may have to be written down.
Gartner analyst Will Sommer said enterprise and consumer AI revenue “will be insufficient to support profitable returns” on the planned spending. The firm also predicts that more than 80% of “neocloud” providers, the newer companies that rent out AI computing power, will stop operating independently by 2030, through mergers, takeovers or closures.
In a gold rush, the people selling shovels often do better than the people digging.
How both stories fit together
Chipmakers get paid when hardware is ordered. The companies buying that hardware only make money later, if customers pay enough for AI services. Right now, demand for chips is so high that TSMC says it cannot make enough. But Gartner’s point is that revenue from AI products is not yet growing fast enough to justify everything being built.
It is important to note that Gartner’s figures are forecasts, not losses that have already happened. AI revenue could grow faster than expected, and big cloud companies such as Microsoft, Google and Amazon have other businesses to absorb risk.
Why it matters for you
- For investors: AI chip results are strong, but share prices already expect a lot. Forecasts like Gartner’s can make markets nervous.
- For businesses: if you rent AI computing from a smaller provider, think about what happens if that provider is bought or shuts down.
- For everyone else: fierce competition to earn back these investments could mean cheaper and better AI tools, at least for a while.
