Artificial intelligence has forced the biggest cloud service providers to make a difficult choice.
Amazon, Microsoft Google and Meta are all spending an incredible $400 billion or more annually on AI infrastructure. This has changed the market, caused a semiconductor shortage, and affected vendor dynamics.
Executives and investors are faced with an unpleasant reality as the year 2026 draws near: maintaining current pace could result in a loss of profitability, while reversing course would almost guarantee losing out on the race for competitiveness.
There are two scenarios that pose equal risks to the industry, both of which carry distinct risk for market valuations.
The size of the wager: numbers that are not conventional
It is difficult to comprehend the scale of spending by hyperscalers.
Alphabet has increased its capital spending guidance for 2025 by three times. It now stands at $91-$93bn, up from $52.5bn in 2024.
Microsoft has spent $34.9 billion on capital expenditures during a quarter. This is a jump of 74% from the previous year. Executives have indicated that capex in fiscal 2026 will be even higher.
Amazon has raised its capex guidance for 2025 to $125 billion. This represents a 61% year-over-year increase.
Meta’s capex guidance has been raised to $70 billion. CEO Mark Zuckerberg stated that “making an investment of this magnitude is likely to prove profitable”.
Analyst estimates predict that the combined capital expenditures of four hyperscalers will reach $600 billion by 2026. This represents a 36% increase from last year.
The capital intensity, or capex, as a percent of revenue has reached previously unthinkable heights. Some hyperscalers dedicate 45-57% to their revenues for infrastructure.
In 2025, the global capital expenditures for data centers grew by 59% in just the third quarter, the eighth quarter with double-digit growth.
Goldman Sachs estimates that the total capex for hyperscalers from 2025 to 2027 is expected to reach 1,15 trillion dollars, which is more than twice as much as what was spent in 2022-2024.
Unstated assumptions: Money must be realised
These staggering numbers are based on a single untested assumption that is dominant in the market: massive infrastructure investments today will lead to a durable and asymmetric growth of revenue.
This assumption is flawed because it ignores a critical gap. Enterprise adoption is increasing, but what about end user demand?
Joshua Mahony Chief Market Analyst, Scope Markets, summarized the tension at its core:
The analyst said that today’s valuations of mega-cap AI companies assume the surge in AI expenditure isn’t a single-off build but rather the beginning of a profitable and self-reinforcing sector.
Businesses continue to invest in AI software and products because AI drives revenue growth. The business-tobusiness expenditure is clear, but end-users’ spending still remains a mystery.
Joshua Mahony said that the focus on circular spending is a sign of market concern, as it highlights the fact that end-user AI expenditure has not yet been fully revealed.
The central risk for 2026 is a direct result of this observation.
Cloud providers spend on the assumption they’ll reap huge revenues from an influx of AI-powered apps, most of which aren’t available yet at scale.
The conversion rate of AI services is still low, but cloud providers continue to grow.
AI services will only generate about $25 billion of revenue by 2025. This is roughly 10% less than what hyperscalers spend on infrastructure.
This disconnect highlights an important gap. Only about 25 percent of AI projects have achieved their ROI expectations to date. Less than 20 percent have been implemented across the entire enterprise.
J.M. Arbour Wealth Management CEO Jac Arbour is a technology strategist. Arbour Wealth Management warned that:
In the narrative of 2026, one of the biggest assumptions is that current valuations will be justified by future fundamentals.
Jac Arbour, speaking to ICD said that the AI startup and tech ecosystem was exuberantly valued and structurally vulnerable. This is largely due to early-cycle hype having overshadowed realistic revenue and profit expectations.
It is important to know what will happen in 2026.
Mahony’s complete analysis frames the stakes in a precise way:
By 2026, investors must see tangible returns that will justify their investments and show that the rising AI infrastructure spending is sustainable.
There are still many risks, such as energy shortages, Chinese competition in data centres, depreciated hardware, and overvalued valuations.
The analyst said that if AI could deliver sustained growth in earnings, which included revenues from consumers and companies outside the tech sector, this would go a very long way towards overcoming any wider concerns.
The inflection is here. Hyperscalers bet that the infrastructure that they are building will catalyse a wave profitable AI services by 2026 and that enterprise adoption of AI technology that justifies their investment.
The capex boom will appear prescient and underestimated if this thesis is true, if the end-users demand appears, if spending increases beyond B2B discussions, if consumers adopt AI-powered product, which drives revenue growth.
If 2026 comes without any evidence of monetization then the market’s sentiment may change dramatically.
Strategic inertia: The cost of caution
Executives understand that cutting back capex in 2026 comes with its own catastrophic risks.
AI Infrastructure has become more competitive than ever before.
The first to build the most powerful and efficient data center gains an asymmetric advantage: access to NVIDIA’s latest GPUs and faster training cycles for models, as well as exclusive partnership with enterprises and pricing power.
Supply constraints directly affect business growth when delays in increasing capacity are delayed.
Microsoft has already confirmed that the supply shortages will continue into fiscal year 2026’s first half.
Stopping investment means giving up these advantages and being at a disadvantage when spending is resumed. The lead times for GPUs and server are notoriously lengthy.
The supply chain for semiconductors remains constrained, as NVIDIA Blackwell Ultra and hyperscalers’ custom accelerators compete with limited capacity foundry at TSMC.
Hyperscalers who step back risk being unable re-acquire their capacity once the market realizes that AI demand is real.
The developer ecosystem and the startup ecosystems already gravitate towards the platforms that have the highest computing power.
The perceived stability and capacity of Azure, Google Cloud or AWS are some factors that influence the decision.
This mindshare is of real value to the economy. It drives down switching costs and locks in clients, while creating network effects which compound with time.
It is impossible to regain that advantage in just a quarter.
A new frontier: Energy as a constraint in strategic planning
Energy is a new dimension to the capital expenditure dilemma.
AI data centres are huge consumers of energy, so securing clean, reliable power 24/7 has become an essential competitive requirement.
Google signed an historic agreement with Kairos Power for the deployment of 500 megawatts in advanced nuclear power by 2035. The first facility will be Hermes 2 which is expected to come online by Tennessee within 2030.
Google also restarted Duane-Arnold Nuclear Power Plant, Iowa. The plant was shut down in 2020 and the target for a restart is 2028-2029.
Microsoft’s demand for electricity to power AI data centres is expected to increase by over 600% in 2030. This will create infrastructure problems and community opposition where the facilities are located.
Google’s acquisition of Intersect Power for $4.75 Billion by Google shows how important energy infrastructure is to AI strategy.
The energy contracts aren’t optional add-ons, they are essential for AI scaling.
Costs and risks are high, and any delays in the acquisition of power can have cascading consequences on your entire capital expenditure plan.
Value fragility, venture risk and ecosystem risk
Second-order effects of Hyperscaler Capex Decisions ripple throughout the Venture Capital Ecosystem.
In the first half 2025, global AI startups will have raised an estimated $83.6 billion. This represents 57.9% all venture capital financing.
This capital, however, is dangerously concentrated: In Q2 2025 nearly 40 billion dollars of the $91 million in global VC financing went to only 16 companies who raised more than $500,000,000.
Hyperscalers who signal a reduction in capex or, worse still, those who maintain their spending without monetising it, could cause a rapid shift of sentiment.
Startup valuations are based on the assumption hyperscale infrastructure will lead to a flourishing ecosystem of AI services and products.
If this thesis fails, venture capital investments could be slashed, especially amongst companies with lower capital structure.
Export control and geopolitical pressure
Geopolitical factors complicate the competitive landscape.
Recent policy changes threaten to ease the restrictions on China’s AI capability.
Trump has approved the sale of Nvidia chips to China. This could change fundamentally the AI competitive landscape in global.
Hyperscalers who invest in massive infrastructure face new competition and less certainty if China has access to the latest chips.
TSMC’s foundry and Nvidia’s GPU production are concentrated, creating a systemic vulnerability.
Hyperscalers’ expansion plans would be crippled by any disruption in either supply chain. This could come from geopolitical tensions, delays in manufacturing, or even natural disasters.
The unknown is a real asset. It increases the risks of investing massive amounts if forces outside a company’s control could undermine returns.
Inflection point 2026
Investors and business executives will find 2026 to be a pivotal year.
Mahony emphasized that the industry must demonstrate concrete evidence of AI infrastructure investments delivering tangible business results. This is what will determine whether valuations are justified and confidence can be restored.
Arbour’s analysis captures the wider market dynamics:
In unlikely industries, new leaders can appear, with names that are often obscure or unrecognized. Some current giants may disappoint, while others will become the next big thing. More than ever, diversification and selective exposure are important.
Hyperscalers are choosing to keep spending the same amount and increase their AI wager.
This choice is a reflection of a decision that believes the negative consequences of being behind outweigh the negative consequences of spending too much. This bet is still not settled. The year 2026 will determine if this confidence was justified.
Answering this question will determine the future of not just individual technology stocks, but the whole tech-driven economy in the coming decade.
The post Looking forward to 2026: Why hyperscalers cannot slow down spending without losing AI War may be updated as new information becomes available.
This site is for entertainment only. Click here to read more
