하이퍼스케일 시장 추적 조사(2025년 4분기) : 2025년 CAPEX는 65% 증가로 5,000억 달러를 돌파

Big tech's AI mania and the search for the elusive moat is driving capex higher, despite some alarming warning signs.

Snapshot view of 4Q25

The 4Q25 edition of MTN Consulting's Hyperscale Tracker covers a market that has never been bigger, faster-moving, or harder to read. Annualized revenues crossed $3 trillion. Capex cleared $500 billion for the first time, up 65% in a single year. Net profit margins hit a 15-year high. And yet free cash flow margins dropped to a 15-year low. Headcount growth, which ran at 7% per year over the prior six years, fell to zero. The full report analyzes what is driving each of these numbers, which companies are winning and losing, and what the data says about where this market goes next.

Why read this report

MTN Consulting has been tracking this market with quarterly deep dives since 4Q17, and our database begins in 4Q11. We have a long history of in-depth coverage of this market, and proprietary tools and consistent, comparable data across revenues, capex, R&D, profitability, employment, and balance sheet metrics for the world's leading hyperscalers (previously called "webscalers"). We cover Amazon, Alphabet, Microsoft, Meta, Apple, Oracle, Alibaba, Tencent, and more than a dozen other players, including newer entrants like Coreweave, Nebius, Kuaishou, and Xiaomi. No other analyst firm covers this market with this depth or our data coverage.

We have been calling this market a bubble for over a year and a half, and that view is reflected throughout the analysis. The full 4Q25 report includes detailed company-by-company breakouts, regional data, vendor revenue benchmarks, and three capex outlook scenarios for 2026-2028 ranging from $420 billion to $810 billion. If you are making investment, procurement, or strategic decisions that depend on where hyperscale spending goes from here, this report is built for you.

Hyperscale's AI-driven infrastructure buildout keeps breaking records. In 4Q25, the companies in our Hyperscale Tracker generated $875 billion (B) in single-quarter revenue (+14.2% YoY), with annualized revenues reaching $3.019 trillion (+13.2% YoY). Annualized capex hit $507B (+64.7% YoY), R&D reached $384B (+20.2% YoY), and cash holdings stood at $745B (+13.6% YoY) against $679B in debt (+23.2% YoY). Net PP&E surged 47.5% YoY to $1.368 trillion. Headcount growth was essentially flat at 0.0% YoY.

Notes: (1) This is MTN Consulting's 33rd quarterly assessment of the hyperscale market, part of a series we launched in 4Q17; our data and analysis spans the 1Q11-4Q25 timeframe, i.e. 60 quarters. (2) The companies in our study include several recent additions: CoreWeave (added in 2024), and Kuaishou, Nebius (Yandex spinoff), and Xiaomi (added in 3Q25). (3) The 4Q25 edition adopts a new term for the sector, swapping "webscale" with the more common "hyperscale".

Revenue: Topline grows at 14% YoY rate again, led by Meta, Alphabet, and Coreweave

Hyperscale revenues reached $875B in 4Q25, up 14.2% YoY, with annualized revenues rising to $3.019T (+13.2% YoY). The largest players remain Amazon, Apple, Alphabet, Microsoft, and Meta. Meta continues to deliver above 20% growth, driven by AI-powered improvements in advertisement ranking, targeting, and recommendations.

The fastest growth came from Coreweave (+110% YoY), the crypto-turned-neocloud player. Nebius, the Yandex spinoff, recorded an even higher rate but remains an outlier with limited geographic reach. Among established companies, Meta (+24%), HPE (absorbing Juniper, +18%), and Alphabet (+18%) saw the strongest top-line results, with GCP acceleration, YouTube advertising, and Gemini enterprise seats all contributing at Alphabet. At the other end, revenue weakness was seen at Baidu (weak legacy ad business), JD.Com (weak Chinese economy and price competition), and Alibaba (tough ecommerce competition). Fujitsu declined as well.

Advertising remains central for several firms. Meta is the most exposed, with ads still driving nearly all revenue, despite efforts to diversify into hardware and AI platforms. Alphabet's non-ad share has risen above 25%. Amazon is approaching 10% of revenues from ads. Ad-dependent companies face heightened risk given concerns about US consumer spending in 2026, which have grown more pronounced following the energy market disruption created by the Trump administration's actions in Iran. A further open question is whether scaled AI platforms will rely heavily on ads, given slow traction for paid subscription models outside early adopters.

Capex: AI arms race drives annualized capex past $500 billion for the first time

Annualized capex reached $507B in 4Q25, up 64.7% YoY, with capital intensity rising to 16.8% of revenues. For the first time in our database, hyperscale capital intensity exceeded that of the telco market. Tech-related (Network, IT, and software) capex accounts for nearly 60% of the total at $297B (+77.5% YoY), focused heavily on GPU servers and related power and cooling systems. The remaining $210B in "other capex" covers land and physical facility construction. The top four hyperscalers (Amazon, Alphabet, Microsoft, and Meta) accounted for 74.2% of all capex in the 1Q25-4Q25 period. Oracle and Meta lead in network and IT capex intensity among major players; Apple is at the bottom, having offloaded most infrastructure to partners.

R&D intensity settled at 12.7% of revenues in 4Q25. For years, R&D as a percentage of revenue tracked well above capex, reflecting a software-centric innovation model. That pattern has reversed: capex has surged to 17% of revenues while R&D has plateaued near 13%. The sector's focus has shifted from a code-first model to a hardware-first land grab. Meta leads in R&D intensity at 29% of revenues, followed by SAP and Oracle.

NVIDIA's data center revenues have scaled in close alignment with hyperscaler tech capex. Other vendors benefiting from the hyperscale buildout include AMD, Arista, Ciena, Cisco, Corning, and Wiwynn. Many of these are racing to expand capacity to meet the rising opportunity, but they also need to hedge against downside risk from a market collapse. Even without that, there will inevitably be a shift away from the market's current over-reliance on NVIDIA. As custom silicon and new chip options mature, a correction in the Network/IT capex line is possible, potentially freeing up capital for R&D or physical facility expansion.

Net PP&E per employee averaged $303K in 4Q25. The big AI model builders are outliers, with net PP&E per employee several times larger than the market average. Fujitsu, JD, SAP, and several others have moved away from infrastructure ownership and record less than $100K in net PP&E per employee. Coreweave and Nebius are at the opposite extreme. M&A spending remains far below capex levels. Traditional acquisitions are too slow a tool for the current arms race, with large deals taking a year or more to close. Capital is instead flowing into infrastructure capex, targeted IP purchases, and acquihires. The Meta-Scale AI transaction ($14B+ for a 49% stake) illustrates this trend, securing talent and data-labeling IP without the integration lag of a full buyout.

Hyperscale capex, negligible a decade ago, surpassed telco capex for the first time in 4Q24, and reached $500B in 2025. US deployments account for an outsized share, representing around 60% of global capex in recent years, though this ratio dipped slightly in 2025 and is likely to slip further as hyperscalers expand footprints in other regions. Chinese AI players' expansion abroad, including into belt-and-road markets, will support this transition. In the short term, GenAI-related spending remains heavily concentrated in the US.

Capex Outlook: Three scenarios for 2026-2028

Capex in 2025 ended at just over $500B. Three potential scenarios shape the outlook. The high case follows current official projections from key tech players, projecting capex above $800B in 2026, followed by two years in the $700-800B range. The slight decrease in 2027-28 assumes NVIDIA begins to lose its lock on chip deployments as captive alternatives developed by hyperscalers take hold, some pending projects are canceled or consolidated, public backlash against buildouts remains a factor, and power stays a constraint on expansion.

The base case projects capex of $550-600B for the next two years, followed by a downtick in 2028. The low case assumes a market correction begins in mid-2026, triggered by a mix of weak economic news from 1Q26 and, relatedly, soft financial results from the hyperscalers themselves. This scenario does not assume an economic depression, but does assume that Chinese companies continue to make breakthroughs at far lower levels of investment and without full access to the priciest chips, and that custom silicon produced by the hyperscalers themselves mature rapidly as an economic alternative to NVIDIA GPUs.

Profitability: Net margin hits record high while free cash flow margin stays at record low

The divergence between net margin and FCF margin has widened significantly. Net profitability reached a record 21.1% in 4Q25, the highest in our database going back to 2011. At the same time, FCF margin sits at 12.8%, the lowest on record, tied with the 3Q25 result. FCF is cash from operations minus capex and is generally a more reliable profit metric than net income. The gap reflects the brutal math of the AI arms race: because FCF is operating cash minus capex, the hyperscaler heavyweights risk cannibalizing their own liquidity with their immense capex spend. Microsoft, Alphabet, and Meta are reporting good net income, but are eating into cash reserves as they race to build more GPU clusters. Amazon and Oracle remain in the same cycle, trading liquid cash for hard infrastructure with long payoff horizons.

On a per-employee basis, Apple and Meta are leaders in FCF, generating $740K and $585K per employee respectively over the last 12 months. Apple's high figures reflect its cautious stance on AI-related capex. In terms of FCF margins by company, Apple leads at 28% and Meta follows at 23%. At the other end, Oracle, Baidu, and Alibaba all recorded negative FCF margins, reflecting heavy build-ahead phases where massive capex for AI infrastructure is being recognized immediately while resulting revenue lags. Margin compression at Alibaba and Tencent reflects a different struggle: navigating a weakened Chinese economy and intense competition while ramping up AI spend.

Regulatory fines and civil lawsuits represent a persistent, though minor, risk to profitability. Hyperscalers consistently treat this as a cost of doing business, often ignoring rulings, aggressively fighting them in court, and using public relations to minimize backlash, moving far from the earlier "don't be evil" philosophy.

Employment: Headcount growth stalls at 0.0% YoY as hyperscalers signal further cuts

Headcount growth across the hyperscale sector was flat at 0.0% YoY in 4Q25, a sharp contrast to the 2019-25 CAGR of 7.1%. Meta and Amazon have both announced layoffs entering 1Q26, and Microsoft has frozen hiring in several groups. Total headcount is a tricky metric in hyperscale, as the sector's employee base is influenced heavily by logistics, fulfillment, and delivery employees at companies like Amazon, Alibaba, and JD.Com. Among the more tech-centric hyperscalers, the direction is clearly downward.

The hyperscale business model is built around massive economies of scale, with investment concentrated in areas where the marginal cost of production can approach zero. Many hyperscaler executives would be happy to see headcount fall significantly and rely on AI platforms to run more of their operations over time. That is clearly the industry's direction. Revenue per employee and net PP&E per employee have both made sizable gains in the last 2-3 years, and those trends are likely to continue.

Regional Revenues: Americas dominant, Europe and MEA growing faster

The Americas accounted for 46.4% of hyperscale revenues in 2025, a share that has held broadly stable since 2011 but edged up slightly over the last three years. In absolute terms, Americas revenues reached $1.399 trillion for the year, up 13.0% from 2024. Asia Pacific was the second largest region at $974 billion (+11.6%), followed by Europe at $550 billion (+15.8%) and MEA at $96 billion (+16.1%).

In 4Q25 specifically, Europe and MEA were the fastest-growing regions on a YoY basis, up 18.5% and 19.0% respectively, while Asia Pacific lagged slightly at 11.9%. The outperformance of Europe and MEA likely reflects continued international expansion by the major US-based hyperscalers, as well as a lower base. Asia Pacific's more modest growth reflects a mix of macro headwinds in China and the competitive pressures facing Alibaba, Baidu, and JD.Com noted elsewhere in this report.

Companies covered:

• 25 in total: 21 active and reporting data publicly, 4 either inactive (ChinaCache), gone private (Twitter), or now part of other companies (Altaba and LinkedIn).
• The 21 active companies include: Alibaba, Alphabet, Amazon, Apple, Baidu, Cognizant, CoreWeave, eBay, Fujitsu, HPE, IBM, JD.COM, Kuaishou, Meta (FB), Microsoft, Nebius, Oracle, SAP, Tencent, Xiaomi, and Yandex
• Three new companies have been added in 2Q25: Kuaishou, Nebius, and Xiaomi. CoreWeave was added in 4Q24.
• Vendors addressed: AMD, Arista , Ciena, Cisco, HPE, Infinera, Intel, Juniper, Nokia, NVIDIA, and Wiwynn

Table of Contents

1. Report highlights

2. Outlook

3. Analysis

4. Key Stats

5. Company Drilldown

6. Company Benchmarking

7. Regional Breakouts

8. Raw Data

9. Exchange Rates

10. About