The $600 Billion AI Bet: Why Cloud Giants Are Spending Big

finance Aug 8, 2026 7 min read By Pyae Phyo Kyaw

On 4 August 2026, The Register reported that Amazon, Microsoft, and Google are collectively committing roughly $595 billion to infrastructure capital expenditure in 2026 — a figure that has grown every quarter as AI demand keeps outstripping supply [1]. The number is staggering on its own: it is larger than the GDP of most countries. But the context makes it more remarkable. It comes right after a quarter in which all three cloud businesses grew at their fastest pace in years — AWS revenue up 36.7% to $42.2 billion, Google Cloud up 82% to $24.8 billion, and Azure up 43% [1][2][3][4][5]. Nearly $600 billion in combined AI infrastructure spending is the clearest signal yet of the scale of the AI buildout, with major implications for the entire supply chain and the broader economy. This post breaks down the numbers, explains why the spending is happening, and looks at what it means next.

Figure 1: The story at a glance. Roughly $595 billion in combined 2026 cloud capex is the clearest signal yet of the scale of the AI buildout. Source: The Register, company earnings reports.

What: the numbers behind the $595 billion bet

Second-quarter earnings season, which ran from 22 to 30 July, delivered a clean sweep for the three hyperscalers. AWS reported net sales of $42.2 billion, up 36.7% year over year — its fastest growth in 18 quarters — with operating income of $16.6 billion and an operating margin of 39.4% [2][3]. AWS is now a $169 billion annualized revenue run rate business [3]. Google Cloud reported $24.8 billion in revenue, up 82%, with operating income of $8.8 billion and a margin of 35.6% [4]. Microsoft's commercial cloud revenue reached $59.3 billion, up 27%, with Azure and related services growing 43% [5].

Table 1: Q2 2026 cloud results for the three hyperscalers. Sources: Amazon IR, About Amazon, TechCrunch, Microsoft IR.
Business	Q2 2026 revenue	YoY growth	Operating income	Operating margin
AWS (Amazon)	$42.2B	+36.7%	$16.6B	39.4%
Google Cloud (Alphabet)	$24.8B	+82%	$8.8B	35.6%
Microsoft commercial cloud	$59.3B	+27%	—	—
Azure (Microsoft)	—	+43%	—	—

The growth is not confined to the big three. Synergy Research Group estimates the enterprise cloud infrastructure market hit $143.4 billion in Q2, up 43% year over year — the highest growth rate in eight years — with trailing twelve-month revenue crossing $500 billion [6]. The three hyperscalers still dominate: AWS holds 28% of the market, Microsoft 20%, and Google 15% [6].

Why: AI demand is the engine

The spending is driven almost entirely by AI. AWS says its AI business now runs at an annual revenue run rate above $25 billion, growing triple digits, and its custom chip business (Trainium and Graviton) has crossed the same mark [3]. GenAI-specific cloud services across the market grew 165% year over year [6]. The Register quotes AWS executives saying "the demand still outpaces that investment" [1].

The backlogs tell the same story. AWS ended the quarter with $496 billion in contracted work not yet online, and Google Cloud's backlog grew to $514 billion [2][4]. These are signed commitments — customers have already agreed to pay for capacity that has not been built yet. Downstream, the demand is visible in the AI companies themselves: OpenAI now generates about $2 billion in monthly revenue, and Anthropic's revenue run rate is expected to reach $50 billion by mid-2026 [9].

Who: the three giants and their suppliers

All three companies raised their 2026 capex guidance during the earnings season [1].

Table 2: 2026 infrastructure capex commitments. Sources: The Register, company earnings reports.
Company	2026 capex guidance	Change from prior guidance
Amazon	~$220B	Raised from ~$200B
Alphabet (Google)	$195B-$205B	Raised from $180B-$190B
Microsoft	~$175B	Q1 FY27 to exceed $50B
Combined	~$595B	—

That money flows into a concentrated supply chain. Roughly 60% of infrastructure dollars go to chips and servers — mostly Nvidia GPUs — with 25% to facilities and power and 15% to networking and cooling [8]. The Register notes that "suppliers are prioritizing their very largest customers, which cloud providers are" [1]. The ripple effects are already visible: Nvidia expects cumulative AI GPU revenue to top $1 trillion by 2027, and memory maker Micron reported Q2 revenue of $23 billion, up 196% year over year [9].

When and where: timing and geography

The buildout is happening now, and it is concentrated in the United States. The US cloud market grew 49% in Q2, well above the worldwide average, and remains larger than the entire Asia-Pacific region [6]. Data centers are the physical expression of the capex: single clusters now need 500MW to 1GW of power, and data centers could consume 9-12% of US electricity by 2028 [8]. That is why hyperscalers are signing nuclear, gas, and geothermal power deals directly, and why liquid cooling has become a $15 billion-plus annual market as high-density AI racks run too hot for air [8].

Which bets: GPUs, custom chips, and power

Each hyperscaler is placing slightly different bets. Amazon is leaning on its own silicon — Trainium and Graviton — alongside Nvidia GPUs [3]. Google is pushing its TPUs and began recognizing revenue from TPU system sales to customer data centers in Q2 [4]. Microsoft is pairing Azure with its OpenAI partnership and extended the useful life of its data center buildings from 15 to 25 years, an accounting change that spreads the cost of the buildout over a longer period [1][5].

How: funding the buildout

The scale of the spending raises an obvious question: where does the money come from? The answer, so far, is mostly from the companies' own cash flows, supplemented by debt and equity. Alphabet raised $49.6 billion in equity and $20.3 billion in debt in Q2 alone [4]. Amazon's free cash flow flipped to an outflow of $7.6 billion over the trailing twelve months as AI infrastructure investment accelerated [2]. Microsoft said Q1 FY27 capex is expected to exceed $50 billion [1]. The funding is a bet that today's spending becomes tomorrow's revenue — and the backlogs suggest customers are already signing up for it.

The historical parallel: the 1999-2001 fiber boom

Every big buildout invites a comparison to the last one, and the obvious parallel here is the telecom fiber boom of 1999-2001, when carriers laid millions of miles of fiber on borrowed money — much of it "dark," or unused, for years [7][9].

Table 3: The 1999-2001 telecom boom vs the 2026 AI buildout. Sources: Michel Johannsen, CoBank, Value Add VC.
Dimension	1999-2001 telecom	2026 AI buildout
Counterparties	CLECs with little or no revenue	OpenAI, CoreWeave, Anthropic — loss-making but revenue-generating
Funding	Debt-financed startups	Hyperscalers with $100B+ annual free cash flow
Capacity	Excess dark fiber, unused for years	AI compute consumed today; no excess capacity
Risk instrument	Trade credit, bad-debt reversals	Equity and purchase commitments, capped losses

The differences matter. Today's counterparties — OpenAI, CoreWeave, Anthropic — are loss-making but not revenue-less, unlike the CLECs of 1999 that often had no revenue at all [7]. The funding is coming from hyperscalers with $100 billion-plus in annual free cash flow, not from debt-financed startups [8]. And unlike the dark fiber of 2000, AI compute is being consumed today [7]. That said, the vendor-financing loops are real: Nvidia has an up-to-$100 billion investment letter of intent with OpenAI and a $6.3 billion purchase backstop for CoreWeave's unsold cloud capacity [7]. The risk is not fraud but synchronization — all the counterparties failing at once [7].

What next: the future outlook

CoBank, which tracks the digital infrastructure economy, argues the cycle is "just beginning": US hyperscalers spent an estimated $235 billion in 2024, $400 billion in 2025, and are projected to exceed $700 billion in 2026 [9]. The Register's ~$595 billion figure covers only the three giants; add Meta, Oracle, and the neoclouds and the total is even larger [1][9].

The risks are real. Concentration is the biggest: a handful of companies are making the largest capital commitment in tech history, and a slowdown in AI demand would hit the entire supply chain. Power is a constraint — data centers are competing with homes and factories for electricity. And the fiber bottleneck is back: AI GPU racks need 10-36x more fiber than traditional CPU racks, and fiber prices hit a seven-year high in early 2026 [8].

For the broader economy, the capex is a double-edged sword. It is a powerful investment engine — jobs, construction, chip fabs, power plants — but it concentrates risk in a small number of balance sheets. The most likely outcome, as one analysis puts it, is not collapse but a sharp correction that wipes out weaker players while the hyperscalers keep building [8].

What to watch in the coming quarters:

Backlog conversion. AWS's $496 billion and Google Cloud's $514 billion in contracted work are the best leading indicators of whether the demand is real [2][4].
Capex guidance revisions. Every raise so far has been upward. A quarter where guidance stays flat would be a signal the cycle is maturing [1].
Power and fiber. Data center electricity demand and fiber prices are the physical constraints that will decide how fast the buildout can actually proceed [8].
Vendor financing terms. In 1999, loosening terms were the tell that the cycle was turning. Watch for the same in AI deals [7].

References

Disclaimer

Not financial advice. This content is for educational purposes only. Figures are as of 8 August 2026 and may be revised; markets move quickly. Always do your own research before making any investment decision.