Tech Giants' CAPEX Comparison for 2026: Who is the Most Aggressive in the AI Computing Power Race — Google, Microsoft, Amazon, or Meta?

In early June 2026, Alphabet announced the completion of an equity financing totaling $84.75 billion — the largest single equity issuance in history worldwide, surpassing Petrobras' $70 billion record set in 2010. More notably, this financing was not an isolated case. Simultaneously, Alphabet raised its 2026 capital expenditure (CAPEX) guidance to $180 billion to $190 billion, and forecasted a significant increase in spending for 2027.

Alphabet’s move is just a snapshot of the AI infrastructure race among the four tech giants. This system compares Alphabet (Google), Microsoft, Amazon, and Meta’s 2026 CAPEX budgets, funding structures, and strategic logic, and dissects why a traditionally “asset-light” tech company faced nearly $190 billion in capital spending in 2026 and had to revert to equity financing tools unused for over two decades.

Alphabet’s Financing Puzzle: Where Did the $84.75 Billion Come From?

To understand Alphabet’s financing scale, first break down its structure. The financing consists of three parts:

Public offering: $34.8 billion. According to regulatory filings, $18 billion came from Class A common shares and Class C capital shares, and $16.75 billion from mandatory convertible preferred shares (issued via depositary receipts). The underwriting and pricing for this issuance were completed on June 2, with market demand exceeding supply several times — a rare occurrence for a large, mature company like Alphabet.

ATM Market Price Offering: $40 billion. This is a continuous mechanism without fixed issuance dates, expected to start in Q3 2026, whereby Alphabet will directly sell Class A and Class C shares to the open market as needed.

Berkshire Hathaway Private Placement: $10 billion. This is the most strategically significant component of the financing. Berkshire bought $5 billion worth of Class A shares at approximately $351.81 per share and $5 billion worth of Class C shares at about $348.20 per share, at a 6%-8% discount to the market closing price before the announcement. Notably, Berkshire has been gradually building its position in Alphabet since Q3 2025, holding about 58 million shares (roughly $17 billion) by the end of Q1 2026. With this $10 billion injection, its total holdings will rise to approximately $27 billion to $32 billion.

Why does Alphabet need such a large external financing? The core reason lies in the enormous scale of CAPEX. Even though Alphabet’s operating cash flow over the past 12 months reached about $174 billion, it is difficult to cover the $180 billion to $190 billion annual capital expenditure solely with internal cash flow. According to Capital Futures, before initiating equity financing, Alphabet had already raised over $85 billion through bond markets in the past year, in currencies including USD, EUR, GBP, JPY, CAD, and CHF, notably issuing a rare 100-year GBP bond. This is a typical “full-spectrum financing” strategy — debt to flatten long-term cost curves, equity to lock in immediate large capital, forming a multi-term, multi-currency funding system.

2026 CAPEX Panorama of the Four Tech Giants

The capital race for AI infrastructure in 2026 has evolved from phased investments to systemic strategic reinvestment. According to a June 2026 update forecast from Goldman Sachs, the four major hyperscale data center operators (Alphabet, Amazon, Microsoft, Meta) will have a combined CAPEX of $725 billion in 2026, up 77% from $410 billion in 2025. S&P Global’s estimates also point to over $700 billion.

The specific data and guidance are as follows:

| Company | 2026 CAPEX Guidance | Compared to 2025 (approx.) | Core Spending Focus | | --- | --- | --- | --- | | Amazon (AWS) | ~$200 billion | +50% | AI data centers, in-house chips (Trainium/Graviton), logistics and satellite networks | | Alphabet (Google) | $180–$190 billion | +100%+ | AI data centers, next-gen TPU R&D, Gemini model training | | Microsoft (Azure) | ~$190 billion (calendar year) | +130% (annualized) | GPU/CPU clusters, long-term infrastructure (>15 years), energy supply | | Meta | $115–$135 billion (later raised to $125–$145 billion) | +85% | Meta Superintelligence Labs, third-party cloud infrastructure, AI training clusters |

Sources: TrendForce May 2026 report, Goldman Sachs June 2026 update, S&P Global February 2026 analysis. Note: Microsoft’s figures vary depending on scope: TrendForce uses $190 billion calendar year, while S&P Global reports over $140 billion based on fiscal year (ending June), due to differences in definitions and accounting.

Amazon leads with a guidance of $200 billion. According to its February 2026 earnings report, CEO Andy Jassy stated that AWS is the main investment focus, with a significant portion allocated to the “Rainier” AI infrastructure project centered on self-developed Trainium2 chips. AWS has deployed nearly 500k Trainium2 chips, aiming for 30% of AI computing tasks to be handled by in-house chips by the end of 2026.

Microsoft’s $190 billion expenditure implicitly contains a key structural element: in Q2 of fiscal year 2026 (Q4 2025 calendar), about two-thirds of CAPEX is directed toward short-lived GPU and CPU devices, with the remaining one-third allocated to long-term infrastructure with a lifespan exceeding 15 years. This reflects Microsoft’s need to balance between two asset classes. More critically, Microsoft’s “business remaining performance obligations” (business RPO) has surged to about $625 billion, even after deducting $281 billion related to OpenAI, leaving $344 billion — more than twice the total order backlog of AWS. This indicates that Microsoft has already locked in future revenue certainty before making CAPEX investments.

Meta’s CAPEX range was revised upward in 2026 from the initial $115–$135 billion to $125–$145 billion. Spending mainly flows into its “Superintelligence Labs” and large-scale data centers supporting AI training. Compared to Amazon and Alphabet, Meta’s absolute CAPEX is smaller, but its year-over-year growth rate of about 85% makes it the most aggressive among the four giants.

Billion-Dollar Signal: Why Did Berkshire Hathaway Enter Now?

In Alphabet’s financing structure, Berkshire Hathaway’s $10 billion private placement is the most noteworthy. Known for staying “away from cutting-edge tech,” Berkshire’s portfolio has long been dominated by Apple, with its tech holdings viewed mainly as “consumer electronics” companies. The large additional stake in Alphabet — after its initial position in Q3 2025 and tripling in Q1 2026 — signals a substantial shift in investment logic.

What does Berkshire value?

First, Alphabet’s “moat” of cash flow remains solid. Google’s core search advertising provides a stable, high-margin cash base, YouTube continues to grow as an advertising engine, and Google Cloud is becoming a second growth curve fueled by AI — with Q1 2026 cloud revenue surpassing $20 billion, up 63% year-over-year, and AI solutions revenue growing nearly 800% YoY.

Second, while AI infrastructure CAPEX is “money-burning,” the assets generated have long-term value. Data centers, once built, typically have a lifespan of over 15 years. For Berkshire, which favors long-term holdings, Alphabet’s investment logic is akin to investing in railroads (BNSF) — building long-term infrastructure for freight transport versus building long-term compute infrastructure for AI.

Third, the risk-reward of buying at a discount is attractive. Berkshire locked in its position at about 6%-8% discount, acting as a “cornerstone investor” in the $84.75 billion financing, helping to ease market concerns about equity dilution and endorsing Alphabet’s financing.

Computing Power and Electricity: Two Major Constraints in the CAPEX Race

To understand the $725 billion CAPEX in 2026, two core constraints must be examined: computing supply and power capacity.

On the compute side, all giants are accelerating their shift toward self-developed chips (ASICs). Google’s TPU (Tensor Processing Unit) has reached its seventh generation, becoming the most important alternative path outside NVIDIA GPUs. Amazon’s Trainium series is also deployed at scale within AWS, with 1.4 million Trainium2 chips shipped by early 2026. Microsoft, while not publicly disclosing ASIC deployment at the same scale, allocated about two-thirds of its Q4 2025 CAPEX to short-lived GPU/CPU devices, with the rest dedicated to long-term infrastructure. The strategic value of self-developed chips lies not in immediate replacement of NVIDIA but in reducing dependency on a single supplier and achieving better unit cost at inference.

On the power side, AI data centers’ electricity consumption is now measured in gigawatts (GW). Microsoft added 1 GW of data center capacity in Q2 FY2026 (Q4 2025 calendar). TrendForce estimates that global data center power capacity will reach about 155 GW in 2026, up 29% year-over-year, with AI servers’ total power consumption surpassing that of general-purpose servers for the first time. This makes power supply — along with cooling systems and high-voltage direct current (HVDC) transmission — a key variable in the CAPEX race.

Long-Term Outlook: The Significance and Risks of $53 Trillion

Goldman Sachs’ June 2026 report raised the projected cumulative CAPEX of the four hyperscale data center operators from $4.5 trillion to $5.3 trillion for 2025–2030. If viewed as an economy, this expenditure exceeds the GDPs of over 200 countries including Japan, the UK, India, and France, ranking as the “world’s fourth-largest economy” after the US, China, and Germany.

This forecast also implies two analytical dimensions:

First, financing methods must diversify. According to Goldman Sachs, industry spending over the next five years (including data centers, power, and computing) could reach $7.6 trillion. Relying solely on operating cash flow is insufficient for such a scale. Alphabet’s record-breaking equity financing, along with Amazon, Microsoft, and Meta’s intensive bond and private credit market activities, point to a trend: hyperscale data center operators are shifting from “asset-light, self-financed” models to “heavy assets, multi-channel financing.”

Second, the return pathways are still incomplete. As mentioned, Microsoft has locked in $625 billion in obligations, but the future revenue visibility for Alphabet, Amazon, and Meta is comparatively lower. S&P Global’s analysis indicates that although all these operators are expanding spending, their ability to absorb such investments without significant credit stress varies. The current $725 billion CAPEX scale is close to one of the largest capital reinvestment cycles in tech history, with a potential payoff cycle of 5 to 10 years.

Conclusion

The 2026 AI infrastructure CAPEX race is no longer just a technical path competition but a three-dimensional contest involving compute, capital, and energy. Alphabet’s record $84.75 billion equity raise, Berkshire Hathaway’s $10 billion trust vote during this transition, Amazon’s leading $200 billion, Microsoft’s locked-in future revenue for massive CAPEX, and Meta’s aggressive growth all illustrate this trend. The combined $725 billion CAPEX among the four giants in 2026 is just the beginning of Goldman Sachs’ forecasted $5.3 trillion long-term investment cycle from 2025 to 2030.

For investors, this cycle offers both opportunities and risks: those who can balance compute supply, power capacity, and capital efficiency will likely secure a more advantageous position in the next AI platform cycle.