Big Tech’s four largest companies are preparing to pour a combined $650 billion or more into AI infrastructure in 2026, a spending surge that dwarfs anything the technology industry has ever attempted. Alphabet, Amazon, Meta, and Microsoft have each laid out capital expenditure plans that, taken together, represent the largest single-year corporate investment cycle in history – one that is reshaping semiconductor supply chains, straining power grids, and forcing Wall Street to rethink how it values the world’s most profitable companies.

The numbers are staggering. Alphabet has guided $175 billion to $185 billion in 2026 capex, nearly doubling its 2025 spending. Amazon plans roughly $200 billion, with the bulk flowing into AWS data centers. Meta has earmarked $115 billion to $135 billion, and Microsoft is running at a $145 billion annualized rate. These commitments are not aspirational targets – they represent binding infrastructure contracts, GPU purchase agreements, and data center construction timelines already underway.

The question consuming investors, analysts, and the broader technology ecosystem is whether this spending will generate returns sufficient to justify its scale, or whether the AI infrastructure boom of 2026 will be remembered as the most expensive bet in corporate history.

The $650 Billion Breakdown: What Each Company Is Spending

Each of the Big Four hyperscalers has articulated a distinct rationale for its spending, but the underlying driver is identical: the race to build enough compute capacity to train and serve frontier AI models while meeting surging enterprise demand for cloud-based AI services.

Amazon leads in absolute spending at $200 billion in planned 2026 capex, representing a nearly 50% year-over-year increase. CEO Andy Jassy defended the outlay during the company’s Q4 2025 earnings call, noting that “AWS growing 24% — our fastest growth in 13 quarters — this growth is happening because we’re continuing to innovate at a rapid rate.” AWS posted $128.7 billion in full-year 2025 revenue, with Q4 alone hitting $35.6 billion (up 24% year-over-year). Amazon’s custom Trainium chips now represent a multi-billion-dollar run rate exceeding $10 billion, growing at triple-digit percentages annually.

Alphabet guided $175 billion to $185 billion, with roughly 60% allocated to servers and 40% to data centers and networking equipment. Google Cloud revenue hit $17.7 billion in Q4 2025, up 48% year-over-year, pushing the segment to an annual run rate above $70 billion. Total Alphabet revenue reached $113.8 billion in Q4 2025, with GAAP EPS of $2.82. The company’s cloud backlog surged 55% quarter-over-quarter to $240 billion, indicating strong enterprise demand for AI services through Vertex AI and Gemini-powered tools.

Microsoft is running at a $145 billion annualized capex rate, driven by Azure cloud platform expansion and its OpenAI partnership. Microsoft Cloud grew 26% in Q4 2025, though this trailed Google Cloud’s 48% growth rate. The company’s AI infrastructure push spans custom Maia chips, massive GPU procurement, and a sprawling global data center footprint.

Meta disclosed $115 billion to $135 billion in 2026 capex guidance, nearly doubling its $72 billion spent in 2025. At the midpoint of $125 billion, Meta’s annual capital spending exceeds the GDP of over 120 countries. CEO Mark Zuckerberg framed the investment as betting on “personal superintelligence” rather than the metaverse as the next platform shift.

2026 Big Tech AI Capex Compared to Historical Spending

Company	2024 Capex (Est.)	2025 Capex	2026 Capex Guidance	YoY Growth
Amazon	$53B	~$135B	$200B	~48%
Alphabet	$52B	~$91.4B	$175B–$185B	~97%
Microsoft	$44B	~$80B	$145B (run-rate)	~81%
Meta	$37B	$72B	$115B–$135B	~74%
Total	~$186B	~$378B	$635B–$665B	~72%

The acceleration is unprecedented. In 2025, combined Big Four capex was already estimated at over $378 billion, up 62% from initial analyst expectations of $250 billion to $280 billion. Morgan Stanley had initially projected $300 billion for 2025 before revising sharply upward. Now, 2026 guidance suggests another near-doubling, with the combined total approaching $650 billion.

👁 2026 Big Tech AI Capex Compared to Historical Spending

Where the Money Goes: Servers, Data Centers, and Custom Chips

The capital is flowing into three primary categories: GPU and custom chip procurement, physical data center construction, and networking and cooling infrastructure.

Nvidia remains the dominant supplier. The chipmaker posted $215.9 billion in total revenue for its fiscal year ending January 2026, up 65% from FY2025’s $130.5 billion. Data center revenue specifically reached $197.3 billion, up from $115.2 billion the prior year. Nvidia’s quarterly revenues progressed from $44.1 billion in Q1 to $68.1 billion in Q4, reflecting escalating demand from hyperscalers. CEO Jensen Huang noted during the Q4 earnings call that “hundreds of billions’ worth of capital expenditures now flow into AI, which eventually translates into growth, which translates directly to revenues.”

Nvidia’s market capitalization has reached approximately $4.83 trillion, making it the world’s most valuable company, powered almost entirely by data center AI chip demand. The company’s Blackwell and forthcoming Vera Rubin architectures are central to hyperscaler buildout plans.

Meanwhile, each hyperscaler is investing heavily in custom silicon to reduce Nvidia dependence. Amazon’s Trainium chips power a multi-billion-dollar business growing at triple-digit rates. Google’s TPU chips underpin Vertex AI and Gemini model training. Microsoft is developing its Maia custom accelerator, and Meta is building MTIA (Meta Training and Inference Accelerator) for its Llama model family. Despite these efforts, Nvidia’s data center revenue continues to grow faster than custom chip alternatives, suggesting that hyperscaler demand for third-party GPUs is increasing alongside proprietary chip adoption – not being replaced by it.

The Nuclear Energy Land Grab Powering AI Data Centers

Perhaps the most telling indicator of AI infrastructure scale is the hyperscaler rush to secure nuclear power. The electricity demands of training and running frontier AI models have pushed Big Tech companies into long-term energy deals that would have seemed inconceivable five years ago.

Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Three Mile Island’s Unit 1 reactor, which has been offline since 2019. The 837-megawatt facility will produce approximately 7 million MWh annually once operational in 2028, with Constellation investing $1.6 billion in upgrades including new turbines and control systems. The deal supports Microsoft’s goal of becoming carbon-negative by 2030 and qualifies for federal 45Y clean energy tax credits.

Amazon has pursued nuclear capacity through its agreement with Talen Energy at the Susquehanna nuclear plant, as well as investments in small modular reactor (SMR) technology. Google signed a deal with Kairos Power for SMR deployment, betting on next-generation reactor designs that can be co-located near data center clusters. These agreements signal that Big Tech views nuclear energy not merely as a sustainability play, but as a competitive necessity: without reliable, large-scale baseload power, AI data center expansion will hit physical limits.

The $1.4 trillion grid overhaul now underway across 51 U.S. utilities is directly linked to this demand. Data center electricity consumption is growing so rapidly that utilities are accelerating capital investment cycles to avoid grid destabilization in regions with dense hyperscaler presence.

Cloud Revenue Growth: Is AI Spending Paying Off?

The core question for investors is whether this capex surge is generating proportional revenue growth. The early data suggests a mixed picture.

👁 Cloud Revenue Growth: Is AI Spending Paying Off?

Google Cloud is the clearest success story, with revenue growing 48% year-over-year to $17.7 billion in Q4 2025, adding $2.5 billion in incremental revenue – more than Microsoft Cloud’s $2.4 billion incremental gain in the same period. Google Cloud’s backlog of $240 billion, up 55% quarter-over-quarter, signals sustained enterprise demand for AI workloads through Vertex AI, BigQuery ML, and Gemini-powered tools.

AWS posted its fastest growth in 13 quarters at 24% in Q4 2025, reaching $35.6 billion in quarterly revenue and $128.7 billion for the full year. Amazon’s custom Trainium chips are contributing meaningfully, with the silicon business running at a multi-billion-dollar rate growing at triple-digit percentages year-over-year.

Microsoft Cloud grew 26% in Q4 2025, a respectable rate but notably trailing Google Cloud’s 48% expansion. Microsoft’s Azure platform remains the second-largest cloud provider by market share, but the gap with Google Cloud is narrowing on a growth-rate basis.

Despite these strong topline numbers, analysts warn that revenue growth may not keep pace with capex acceleration. Free cash flow for Big Tech companies could decline by up to 90% in 2026 as capital spending outpaces the revenue it generates, according to Wall Street estimates. The critical question is whether this represents a temporary trough before AI-driven revenue accelerates, or an early indicator that infrastructure spending is running ahead of actual demand.

Cloud Revenue Performance: Q4 2025 Scorecard

Metric	AWS	Google Cloud	Microsoft Cloud	Meta AI
Q4 2025 Revenue	$35.6B	$17.7B	~$51.5B (est.)	Ad AI-driven
YoY Growth	24%	48%	26%	N/A
Full-Year 2025 Revenue	$128.7B	~$70B (run-rate)	~$200B (est.)	N/A
2026 Capex Guidance	$200B	$175B–$185B	$145B	$115B–$135B
Custom Chip Program	Trainium	TPU	Maia	MTIA
Key AI Product	Bedrock	Vertex AI	Azure OpenAI	Llama

Nvidia’s $197 Billion Data Center Windfall

No company has benefited more from the AI capex surge than Nvidia. The chipmaker’s data center segment generated $197.3 billion in FY2026 revenue (fiscal year ending January 2026), up from $115.2 billion in FY2025. Total company revenue hit $215.9 billion, a 65% year-over-year increase from $130.5 billion.

Jensen Huang has positioned Nvidia not merely as a chip vendor but as the enabler of a new computing paradigm. “The hundreds of billions’ worth of capital expenditures now flow into AI, which eventually translates into growth, which translates directly to revenues,” Huang said during Q4 FY2026 earnings. With a market cap of approximately $4.83 trillion, Nvidia is now the world’s most valuable company – a position built almost entirely on selling GPUs to the four hyperscalers and their enterprise customers.

The Nvidia Vera Rubin platform, unveiled at GTC 2026, represents the next generation of data center GPUs. With a 336-billion-transistor design and a claimed 5x performance improvement over Blackwell, Vera Rubin is expected to drive another cycle of hyperscaler procurement starting in late 2026 and 2027. Nvidia expects at least $500 billion in cumulative AI chip revenue by end of 2026, with a target of $1 trillion through 2027 from secured orders.

The company’s dominance creates a paradox for hyperscalers: while each is investing heavily in custom chips (Trainium, TPU, Maia, MTIA) to reduce Nvidia dependency, their aggregate GPU spend continues to grow because AI workload demand is expanding faster than custom chip capacity can scale. Nvidia’s revenue is growing despite the rise of alternatives, not in spite of it.

Wall Street’s Fear: The Free Cash Flow Cliff

For all the bullish rhetoric from Big Tech executives, Wall Street is increasingly uneasy about the sustainability of current spending levels. The central concern is that free cash flow could plummet by up to 90% across the Big Four in 2026 as capex dramatically outpaces revenue growth.

👁 Wall Street's Fear: The Free Cash Flow Cliff

“The existential question for these stocks is whether capex execution produces measurable improvements in revenue and engagement,” noted Wall Street analysts tracking the sector. “If $650 billion in annual spending produces results resembling Meta’s metaverse spending of 2021 to 2023 — with little visible return — the market will punish these companies severely.”

The comparison to Meta’s metaverse investment is particularly pointed. The company’s Reality Labs division accumulated tens of billions in operating losses between 2021 and 2024, with limited consumer adoption to show for it. Meta’s pivot to AI spending represents a rebranding of its capital allocation strategy, but the magnitude of investment raises similar execution risks.

Alphabet’s stock illustrated this tension. When the company announced its $175 billion to $185 billion capex guidance, shares initially nosedived more than 6% in after-hours trading. However, during the subsequent earnings call – where Google Cloud’s 48% growth and $240 billion backlog were detailed – the stock recovered, eventually closing down just 0.4%. The whiplash reflected the market’s conflicted stance: investors want AI growth but fear the cost of achieving it.

Dan Ives, a senior analyst at Wedbush Securities, has described 2026 as “the year AI spending must start showing returns.” He notes that while enterprise AI adoption is accelerating, the gap between infrastructure investment and revenue realization remains wide. “We are in the early innings of a multi-year AI infrastructure build,” Ives has argued, “but the market’s patience is not unlimited.”

The Custom Chip Arms Race: TPU vs Trainium vs Maia vs MTIA

One of the most significant subplots in the AI capex story is the race among hyperscalers to develop custom silicon that can reduce their dependence on Nvidia. Each company is pursuing a slightly different strategy, but the goal is identical: lower per-inference costs and increase control over AI hardware supply chains.

Google’s TPU program is the most mature. Google has been designing Tensor Processing Units since 2015, and TPUs now power the company’s largest AI workloads, including Gemini model training and inference across Google Search, YouTube, and Cloud. The latest TPU generation underpins Vertex AI and Google Cloud’s AI services, which grew 48% year-over-year to $17.7 billion in Q4 2025.

Amazon’s Trainium is the fastest-growing custom chip business by revenue metrics. The Trainium chip family, now running at a multi-billion-dollar annual rate exceeding $10 billion with triple-digit growth, is optimized for training large language models on AWS. Amazon is positioning Trainium as the default training chip for Bedrock and SageMaker customers, offering cost advantages over Nvidia GPUs for certain workloads.

Microsoft’s Maia accelerator is designed to complement the company’s massive Nvidia GPU fleet rather than replace it. Microsoft’s strategy centers on using Maia for specific Azure OpenAI Service workloads where custom hardware can improve efficiency, while continuing to rely on Nvidia for general-purpose training and inference.

Meta’s MTIA (Meta Training and Inference Accelerator) supports the company’s open-source Llama model family and its AI-powered advertising systems. Meta’s AI ad infrastructure – which uses machine learning to optimize ad targeting and content recommendations – is the primary revenue driver justifying the company’s $115 billion to $135 billion capex plan.

The Enterprise AI Demand Signal

Beyond hyperscaler spending, the broader enterprise market is providing strong demand signals that help justify the infrastructure buildout. According to enterprise surveys, 88% of companies report that AI adoption has driven revenue gains, with 30% seeing increases above 10%. Additionally, 86% of enterprises plan to increase their AI budgets in 2026, suggesting sustained demand for cloud-based AI services.

Google Cloud’s $240 billion backlog – up 55% quarter-over-quarter – is perhaps the single strongest data point supporting the capex thesis. Backlog represents contracted but not yet recognized revenue, primarily from multi-year enterprise cloud deals. A $240 billion backlog against $70 billion in annual run-rate revenue implies roughly 3.4 years of contracted future revenue, providing significant visibility into demand.

Andy Jassy echoed this demand narrative at Amazon, stating that AWS “expect[s] to invest about $200 billion in capital expenditures across Amazon in 2026, and anticipate strong long-term return on invested capital.” The emphasis on long-term return acknowledges that the payoff horizon for current spending extends well beyond 2026.

The enterprise demand pattern mirrors what happened during the cloud computing buildout of 2012 to 2018: hyperscalers invested aggressively ahead of demand, suffered periods of compressed margins, and eventually generated outsized returns as workload migration reached critical mass. The bull case for AI infrastructure spending assumes a similar trajectory, but at dramatically larger scale.

Power Grid Strain and Environmental Impact

The environmental and energy implications of $650 billion in AI infrastructure spending are becoming a first-order policy concern. Data center electricity consumption is growing so rapidly that it threatens to destabilize regional power grids, particularly in Northern Virginia, Central Texas, and the Pacific Northwest where hyperscaler density is highest.

👁 Power Grid Strain and Environmental Impact

The International Energy Agency has projected that global data center electricity consumption could reach 1,000 terawatt-hours (TWh) by 2026, equivalent to the total electricity consumption of Japan. AI training workloads, which can consume 10 to 100 times more electricity per compute cycle than traditional cloud workloads, are the primary driver of this acceleration.

The Senate GRID Act represents the most significant legislative response to date, aiming to regulate data center energy consumption and force transparency around power procurement. The bill has bipartisan support and would require large data center operators to disclose energy usage, water consumption, and carbon emissions on a facility-by-facility basis.

Nuclear energy deals – Microsoft’s Three Mile Island restart (837 MW, $1.6 billion investment), Amazon’s Susquehanna agreement, and Google’s Kairos Power SMR contract – represent Big Tech’s attempt to solve the power problem through dedicated clean energy procurement. But these facilities won’t come online until 2028 or later, creating a gap during which hyperscalers will rely heavily on natural gas generation to power their expanding fleets.

The Three Mile Island deal alone is expected to create 3,400 direct and indirect jobs, illustrating the broader economic impact of AI infrastructure development on local communities. Data center construction has become one of the fastest-growing segments of commercial real estate development in the United States.

The China Factor: Competing AI Infrastructure Buildouts

The U.S. AI infrastructure boom is not occurring in isolation. Chinese technology companies including Alibaba, Tencent, and ByteDance are pursuing their own AI infrastructure investments, though at substantially smaller scale due to U.S. export controls on advanced semiconductors.

The Nvidia H200 chip sales resumption to China has injected new complexity into the competitive landscape. While Chinese firms cannot access Nvidia’s most advanced GPUs (Blackwell and Vera Rubin architectures are restricted under Commerce Department rules), the availability of H200-class chips provides sufficient compute for training competitive AI models, as demonstrated by DeepSeek’s strong benchmark performance.

The geopolitical dimension adds urgency to the U.S. hyperscaler buildout. Each dollar spent on domestic AI infrastructure strengthens the argument that America’s AI lead is physical and industrial, not merely algorithmic. The $650 billion in 2026 capex represents, in aggregate, the largest peacetime industrial mobilization in U.S. history – a fact not lost on policymakers who view AI compute capacity as a matter of national security.

What $650 Billion Means for the Semiconductor Supply Chain

The downstream effects of Big Tech’s spending cascade through the entire semiconductor ecosystem. TSMC’s $35.71 billion Q1 2026 revenue and $56 billion capital expenditure plan are directly driven by AI chip demand from Nvidia, AMD, and hyperscaler custom silicon programs. The Broadcom AI revenue surge of 106% to $8.4 billion similarly reflects hyperscaler demand for custom AI accelerators and networking chips.

Supply chain constraints remain a significant risk. Advanced packaging capacity – the process of combining multiple chip dies into a single package for AI accelerators – is the primary bottleneck. TSMC’s CoWoS (Chip-on-Wafer-on-Substrate) packaging capacity has been running at near-100% utilization since 2024, and despite aggressive expansion, lead times for advanced AI chips remain measured in quarters rather than weeks.

Memory supply is equally constrained. High Bandwidth Memory (HBM), required for all frontier AI GPUs, remains in tight supply from Samsung, SK Hynix, and Micron. The combination of packaging and memory constraints means that even $650 billion in committed capex cannot be deployed instantaneously – the spending will stretch across multiple quarters as supply catches up to demand.

Five Predictions for the AI Infrastructure Market

Based on current trajectories and confirmed data, five key predictions emerge for the AI infrastructure market through 2027.

👁 Five Predictions for the AI Infrastructure Market

1. Combined Big Four capex will exceed $800 billion in 2027. With 2026 on track for $650 billion and enterprise AI adoption accelerating, the spending trajectory points toward continued escalation. Each company has signaled that current capex levels are not peak spending but rather the beginning of a multi-year infrastructure cycle. Nvidia’s $1 trillion cumulative revenue target through 2027 depends on this continued growth.

2. At least one hyperscaler will announce capex cuts by Q3 2026. The free cash flow pressure is real. If AI revenue growth decelerates even slightly, at least one of the four companies will revise its capex guidance downward, triggering a broad sector selloff. Meta, given its history of metaverse-related investor pushback, is the most likely candidate.

3. Custom chip revenue will surpass $50 billion across all hyperscalers by end of 2027. Amazon’s Trainium alone is growing at triple-digit rates from a $10 billion-plus base. Combined with Google TPU, Microsoft Maia, and Meta MTIA deployments, the custom chip market is approaching a scale where it genuinely threatens Nvidia’s monopoly position on specific workload categories, even as Nvidia’s overall revenue continues to grow.

4. Nuclear energy capacity dedicated to data centers will triple by 2030. Microsoft’s 837 MW Three Mile Island deal is just the beginning. As existing power grids prove insufficient for AI data center growth, nuclear – both traditional restarts and SMR deployments – will become the default power source for greenfield hyperscale facilities.

5. Google Cloud will overtake Azure in growth rate for at least three consecutive quarters. Google Cloud’s 48% growth versus Microsoft Cloud’s 26% in Q4 2025 reflects genuine market share shifts driven by Vertex AI and Gemini adoption. If this trajectory continues, Google will narrow the revenue gap with Azure significantly by 2027, even if Microsoft retains absolute scale advantage.

Expert Analysis: Is This Sustainable?

Industry experts are divided on the long-term sustainability of the current spending cycle.

Jensen Huang, Nvidia CEO, is predictably bullish: “The hundreds of billions’ worth of capital expenditures now flow into AI, which eventually translates into growth, which translates directly to revenues.” Huang’s confidence is backed by $215.9 billion in FY2026 revenue and $500 billion in cumulative secured orders.

Andy Jassy, Amazon CEO, points to demand metrics: “AWS growing 24% — our fastest growth in 13 quarters — this growth is happening because we’re continuing to innovate at a rapid rate. We expect to invest about $200 billion in capital expenditures across Amazon in 2026, and anticipate strong long-term return on invested capital.”

Dan Ives of Wedbush Securities offers a more cautious perspective: “We are in the early innings of a multi-year AI infrastructure build, but the market’s patience is not unlimited. 2026 is the year AI spending must start showing returns.”

Mark Zuckerberg has framed Meta’s spending as existential: the company is betting on “personal superintelligence” as the next platform shift, with $115 billion to $135 billion in capex representing the cost of staying competitive in a market where underspending could mean permanent relegation to second-tier status.

The tension between these perspectives defines the investment thesis for the entire sector. Bulls point to 48% Google Cloud growth, 24% AWS acceleration, $240 billion in cloud backlog, and 88% of enterprises reporting AI-driven revenue gains. Bears point to potential 90% free cash flow declines, the metaverse precedent, and the historical tendency for corporate capital cycles to overshoot.

How This Reshapes the Tech Industry Hierarchy

The AI infrastructure spending race is fundamentally altering the competitive landscape of the technology industry. Companies that can afford $100 billion-plus annual capex budgets are separating from those that cannot, creating a new tier of “infrastructure-scale” technology companies.

Nvidia, at $4.83 trillion in market capitalization, has ascended to the top of the global corporate hierarchy on the strength of selling picks and shovels to the AI gold rush. Marvell Technology and Broadcom have similarly benefited as secondary suppliers of custom AI chips and data center networking.

The spending asymmetry also has implications for the broader SaaS market. As hyperscalers build AI capabilities directly into their cloud platforms, traditional software companies face the risk of disintermediation. Why pay for a standalone AI analytics tool when Azure, AWS, or Google Cloud offers comparable functionality as a platform feature? The $650 billion infrastructure investment is not just building compute capacity – it is building competitive moats that could reshape the entire software industry.

Related Coverage

• The $1.4 Trillion Grid Overhaul: How 51 Utilities and $300B in AI Capex Are Reshaping America’s Power System
• Nvidia Vera Rubin Platform: Inside the 336B-Transistor Chip and 5x Blackwell Leap
• AWS vs Azure 2026: 31% vs 24% Market Share and a 75% Archive Cost Gap
• TSMC’s $35.71B Q1 2026 Revenue: Inside the 35% Surge and $56B Capex
• Microsoft’s $150 Billion AI Capex Gamble: Inside the Azure Surge
• Broadcom’s AI Revenue Surges 106% to $8.4 Billion
• Senate GRID Act Targets Data Center Energy Crisis

Frequently Asked Questions

How much are Big Tech companies spending on AI infrastructure in 2026?

The four largest hyperscalers – Amazon ($200 billion), Alphabet ($175 billion to $185 billion), Microsoft ($145 billion), and Meta ($115 billion to $135 billion) – have collectively guided between $635 billion and $665 billion in 2026 capital expenditure, with the vast majority directed toward AI data centers, GPU procurement, and custom chip development.

Why is Big Tech spending so much on AI?

Enterprise demand for cloud-based AI services is growing rapidly. Google Cloud grew 48% year-over-year to $17.7 billion in Q4 2025, AWS hit its fastest growth in 13 quarters at 24%, and Google Cloud’s $240 billion backlog indicates sustained multi-year demand. Companies are spending now to build capacity for anticipated future workloads, similar to the cloud infrastructure buildout of 2012 to 2018.

How does Nvidia benefit from the AI spending surge?

Nvidia posted $215.9 billion in total revenue for FY2026 (ending January 2026), with data center revenue specifically reaching $197.3 billion. The company’s market cap has reached approximately $4.83 trillion, making it the world’s most valuable company. Nvidia expects at least $500 billion in cumulative AI chip revenue by end of 2026.

Are Big Tech companies building their own AI chips?

Yes. Amazon’s Trainium exceeds $10 billion in annual revenue with triple-digit growth. Google builds TPUs, Microsoft is developing Maia, and Meta produces MTIA. However, Nvidia’s data center revenue continues to grow alongside custom chip adoption, indicating that total demand exceeds what custom chips alone can serve.

How are Big Tech companies powering AI data centers?

Hyperscalers are turning to nuclear energy. Microsoft signed a 20-year deal with Constellation Energy to restart Three Mile Island’s 837 MW reactor ($1.6 billion investment). Amazon has agreements involving the Susquehanna nuclear plant, and Google signed a contract with Kairos Power for small modular reactors. These facilities won’t come online until 2028 or later.

Could the AI spending boom turn into a bubble?

Wall Street analysts warn that free cash flow across the Big Four could decline by up to 90% in 2026 as capex outpaces revenue growth. The key risk is whether AI infrastructure investment resembles the successful cloud buildout of the 2010s or the unsuccessful metaverse spending that cost Meta billions with limited return. Enterprise adoption data suggests the former, but the scale of investment is unprecedented.