Looking ahead to 2026: why hyperscalers can’t slow spending without losing the AI war 

The artificial intelligence revolution has forced a brutal choice upon the world’s biggest cloud providers.

Amazon, Microsoft, Google, and Meta have committed an unprecedented $400-plus billion annually to AI infrastructure, a spending surge that has reshaped markets, accelerated semiconductor shortages, and created entirely new vendor dynamics.

Yet as 2026 approaches, executives and investors face an uncomfortable reality: maintaining this pace risks profitability if AI adoption lags, but pulling back virtually guarantees losing the competitive race.

The industry is caught between two equally perilous scenarios, each carrying distinct risks for valuations and market position.

The magnitude of the bet: Numbers that defy convention

The scale of hyperscalers spending is almost incomprehensible.

Alphabet alone has revised its 2025 capital expenditure guidance upward three times, reaching $91–$93 billion, compared to just $52.5 billion in 2024.

Microsoft spent a staggering $34.9 billion in capital expenditures in a single quarter, a 74% year-over-year jump, and executives have signaled that fiscal 2026 capex will grow even faster.

Amazon raised its 2025 capex guidance to $125 billion, representing a 61% increase year-over-year.

Meta has boosted its capex guidance to $70 billion, with CEO Mark Zuckerberg explicitly stating that “making a significantly larger investment here is very likely to be profitable”.​

Across the four largest hyperscalers, combined capex is expected to approach $600 billion in 2026, up roughly 36% year-over-year, according to analyst estimates.

This represents capital intensity levels, capex as a percentage of revenue, that have reached historically unthinkable levels, with some hyperscalers dedicating 45–57% of revenues to infrastructure spending.

For context, global data center capital expenditures surged 59% year-over-year in the third quarter of 2025 alone, marking the eighth consecutive quarter of double-digit growth.

Goldman Sachs projects total hyperscalers capex from 2025 through 2027 will reach $1.15 trillion, more than double the $477 billion spent from 2022 through 2024.​

The unstated assumption: Monetisation must materialise

Behind these staggering figures lies a single, untested premise that dominates market thinking: today’s massive infrastructure outlays will translate into durable, asymmetric revenue growth.

Yet this assumption rests on a troubling gap: enterprise adoption is accelerating, but where is the end-user demand?

Joshua Mahony, Chief Market Analyst at Scope Markets, distilled the central tension:

“Today’s mega-cap AI valuations assume that the current surge in AI spending is not a one-off infrastructure build, but the start of a highly profitable, self-reinforcing industry,” the analyst told Invezz.

One in which businesses continue to spend more on AI products and software because AI itself is driving revenue growth. So far, the business-to-business spending has been clear for all to see, but the end-user spending remains questionable.

“The focus around circular spending highlights market concerns that specific AI spending from the end user has yet to fully emerge,” Joshua Mahony added.

This observation cuts to the heart of 2026’s central risk.

The cloud providers are spending on the assumption that they will harvest vast revenues from a wave of AI-powered applications, many of which don’t yet exist at scale.

Cloud providers are experiencing robust growth in AI-related services, but the conversion rates remain troubling.

AI-related services are expected to deliver only about $25 billion in revenue in 2025, roughly 10% of what hyperscalers are spending on infrastructure.

That disconnect highlights a fundamental gap: only about 25% of AI initiatives have delivered their expected ROI to date, and fewer than 20% have been scaled across entire enterprises.​

As technology strategist Jac Arbour, CEO of J.M. Arbour Wealth Management, warned:

The biggest untested assumption in the 2026 AI narrative is that today’s valuations are justified by fundamentals that have yet to materialise.

“The AI tech and startup ecosystem is exuberantly priced and structurally fragile, largely because early-cycle hype has overwhelmed realistic expectations for revenue and profitability,” Jac Arbour said while speaking with Invezz.

The critical question: What happens in 2026?

Mahony’s full analysis frames the stakes with precision:

“By 2026, investors will need to see tangible earnings that justify those investments and demonstrate that rising AI infrastructure spend is sustainable.”

Significant risks remain, including energy constraints, Chinese competition, data-centre capacity, hardware depreciation, and stretched valuations.

“But if AI can deliver sustained earnings growth that also includes revenues from everyday consumers and businesses outside of the tech space, it will go a long way to overshadowing any of those wider concerns,” the analyst added.

This is the inflection point. The hyperscalers are betting that by 2026, the infrastructure they’re building today will have catalysed a wave of profitable AI services and enterprise adoption that validates their expenditure.

If that thesis holds, if end-user demand emerges, if enterprise spending accelerates beyond B2B conversations, if consumer adoption of AI-powered products drives material revenue growth, then the capex binge will be seen as prescient and undervalued.

Conversely, if 2026 arrives without evidence of that monetization, market sentiment could reverse with dramatic speed.

The strategic inertia trap: Costs of caution

Pulling back on capex spending in 2026 carries its own set of catastrophic risks, and executives understand this all too well.

The competitive dynamics of AI infrastructure have become almost Darwinian.

Whoever builds the largest, most efficient data centers first gains asymmetric advantages: priority access to the latest NVIDIA GPUs, faster model training and iteration cycles, exclusive partnerships with enterprise customers, and the ability to set pricing for AI services from a position of strength.

Delays in expanding capacity translate into supply constraints that directly impede business growth.

Microsoft executives have already acknowledged that supply constraints will likely persist into the first half of fiscal 2026.

Pausing investment now means surrendering these advantages to rivals and starting from a disadvantaged position once spending resumes. Lead times on GPUs and servers are notoriously long.

The semiconductor supply chain remains tight, with NVIDIA Blackwell Ultra ramping and custom accelerators across hyperscalers competing for limited foundry capacity at TSMC.

Any hyperscaler that steps back risks being unable to reacquire capacity quickly once the market realises AI demand was genuine.​

Moreover, the developer and startup ecosystems are already gravitating toward the platforms with the most abundant compute.

Startups choose to build on Azure, Google Cloud, or AWS based partly on the perceived capacity and stability of each platform.

That mindshare has real economic value; it drives switching costs, locks in customers, and creates network effects that compound over time.

Ceding that advantage is not reversible in a quarter or two.

Energy as a strategic constraint: The new frontier

The energy dimension of the capex dilemma has emerged as an entirely new layer of complexity.

AI data centers are voracious consumers of power, and securing reliable, 24/7 clean energy has become a competitive necessity.

Google has signed a landmark deal with Kairos Power to deploy 500 megawatts of advanced nuclear capacity by 2035, with the first facility, Hermes 2, coming online in Tennessee by 2030.

Google has also restarted the Duane Arnold Nuclear Power Plant in Iowa, which was closed in 2020, with a target restart of 2028–2029.

Microsoft’s electricity demand for AI data centers is projected to surge over 600% by 2030, creating infrastructure challenges and local community opposition in regions where facilities are sited.​

Google’s recent $4.75 billion acquisition of Intersect Power underscores how critical energy infrastructure has become to AI strategy.

These energy deals are not discretionary add-ons; they are prerequisites for scaling AI operations.

The cost and execution risk are substantial, and delays in securing power create cascading effects across the entire capex plan.​

Valuation fragility and venture ecosystem risk

The second-order consequences of hyperscaler capex decisions ripple through the entire venture capital ecosystem.

Global AI startups raised $83.6 billion in the first half of 2025 alone, capturing 57.9% of all venture capital funding.

Yet this capital is concentrated dangerously: in Q2 2025, nearly $40 billion of the $91 billion in global VC funding went to just 16 companies that raised $500 million or more.​

If hyperscalers signal a pullback on capex spending, or worse, if they maintain spending but fail to monetise it effectively, sentiment could reverse rapidly.

Startup valuations rest largely on the assumption that hyperscaler infrastructure spending will create a thriving ecosystem of AI products, services, and derivatives.

If that thesis breaks, a significant portion of venture capital could unwind, particularly among companies with weaker capital structures.​

Geopolitical pressure and the export control wild card

The competitive landscape is further complicated by geopolitical dynamics.

US export controls on advanced semiconductors have constrained China’s domestic AI capabilities, but recent policy shifts threaten to relax those restrictions.

The Trump administration has signaled approval to sell Nvidia H200 chips to China, which could fundamentally alter the global AI competitive balance.

If China gains access to cutting-edge chips, hyperscalers in the West face a new competitive threat and a less certain return on their massive infrastructure investments.​

Additionally, the concentration of GPU supply from Nvidia and foundry capacity at TSMC creates systemic vulnerability.

Any disruption to either supply chain, whether from geopolitical escalation, manufacturing delays, or natural disaster, would cripple hyperscaler expansion plans.

This unknown has real optionality value; it amplifies the risk of committing massive capital if returns could be undermined by forces beyond any single company’s control.

The 2026 inflection point

For investors and executives, 2026 is a watershed year.

The industry will need to demonstrate tangible evidence that AI infrastructure spending is delivering measurable business outcomes, precisely what Mahony highlighted as the make-or-break threshold for valuations and confidence.

Arbour’s assessment captures the broader market dynamic:

New leaders, often small, obscure, or overlooked names, can emerge in unlikely industries. The next giants will be born, and some current ones will disappoint. Diversification and selective exposure matter more than ever.

The hyperscalers have chosen to maintain spending and double down on the AI bet.

That choice reflects a judgment that the downside of falling behind exceeds the downside of overspending. But that wager is not yet settled, and 2026 will test whether that confidence is justified.

The answer will likely determine not only the fate of individual tech stocks but also the shape of the entire technology-driven economy for the decade ahead.

The post Looking ahead to 2026: why hyperscalers can't slow spending without losing the AI war  appeared first on Invezz