By Rajesh Beri | July 26, 2026
For three years, Big Tech had a tacit agreement with investors: spend whatever you want on AI, and we'll reward you as long as revenue keeps growing.
That deal broke on Thursday.
Alphabet reported its second-quarter earnings — $119.8 billion in revenue, up 24% year-over-year. Google Cloud surged 82% to $24.8 billion. Nearly 90% of the Fortune 100 now uses Gemini Enterprise. Cloud backlog hit $514 billion. By every operational metric, it was a blowout quarter.
The stock dropped 7%. Its worst day in over a year.
The reason: Alphabet raised its 2026 capital expenditure guidance to $195-205 billion, up from $180-190 billion. And for the first time since its 2004 IPO — the first time in 22 years — free cash flow turned negative.
"People are really focused on capex, obsessed with it. It used to be the more the better, but now it is the less the better," Jason Lemire, chief investment officer at Bold Wealth Partners, told Bloomberg. "We're seeing capital raises, negative cash flows, rising debt. All that adds risk to the picture."
The Magnificent Seven index tumbled 4.8% in a single session — its worst day since Trump's "liberation day" tariff announcement in April 2025. The group has collectively lost $797 billion in market value as AI skeptics dump tech stocks.
For enterprise AI buyers, this isn't a Wall Street story. It's a procurement crisis hiding in plain sight. When your cloud providers are burning cash at historic rates and investors are demanding accountability, the bill eventually lands on your desk.
The Numbers: $725B and Counting
The scale of Big Tech's AI infrastructure bet is now approaching levels that strain comprehension:
| Company | 2026 Capex (Est.) | YTD Stock Performance | Key Signal |
|---|---|---|---|
| Alphabet | $195-205B | Flat (post 7% drop) | First negative FCF since IPO |
| Microsoft | ~$190B | -21% | Weakest Mag Seven member |
| Amazon | ~$185B (est.) | Flat | Earnings next week |
| Meta | ~$155B (est.) | -9.8% | Investors questioning AI ROI |
| Combined (4) | ~$724B | — | Bloomberg analyst composite |
Add Oracle and CoreWeave, and Moody's puts the total at $785 billion, approaching $1 trillion by 2027.
For context: the entire US defense budget is $886 billion. Four technology companies are now spending at 82% of the Pentagon's annual budget — on data centers.
The Moody's report reveals what balance sheets don't show: hyperscalers have accumulated over $1.2 trillion in off-balance-sheet data center leases and power purchase agreements. Free cash flow margins across the group have compressed from 28% to 11%. These companies are no longer asset-light software businesses. They're capital-intensive industrial utilities.
Why This Matters for Enterprise Cloud Buyers
Here's the uncomfortable math: someone has to pay for $725 billion in infrastructure. There are only three sources:
1. Investors (via lower returns) — This is happening. MSFT down 21%, GOOGL losing 7% in a day. But investor patience has limits.
2. Cloud customers (via higher prices) — This is already happening.
3. Efficiency gains (do more with less) — This is the promise. It hasn't materialized at scale.
The evidence that Door #2 is already open:
- AWS raised AI compute prices in Q2, citing memory shortage costs — passing infrastructure cost increases directly to enterprise customers
- Google Cloud lifted data transfer, CDN Interconnect, and Direct Peering prices — with some North American network services increasing 100%
- H100 lease contracts rose approximately 40% over five months into early 2026
- Alibaba Cloud and Baidu AI Cloud raised prices by up to 34% on AI compute products
- Tencent Cloud raised input prices on its large-model service by 463% in specific categories
- Anthropic shifted Claude Fable 5 to usage-based billing after July 7
As Wangsu product director Wang Zhijie told China Biz Insider: "The cloud price-war phase has ended. The market is entering a 'value reversion' cycle, shifting from 'scale priority' to 'profit priority.'"
Translation for enterprise buyers: AI workloads are no longer deflationary at the infrastructure layer. The era of ever-cheaper compute is over.
The Circular Revenue Problem
Moody's identified a structural risk that should concern every enterprise vendor evaluation: the "circular AI revenue loop."
Here's how it works:
- Microsoft invests $13B in OpenAI
- OpenAI is contractually required to spend most of that on Azure compute
- Microsoft reports the spending as Azure AI revenue growth
- Investors see revenue growth and approve more capex
- Microsoft invests more in OpenAI (or others)
- Cycle repeats
The same pattern applies to Amazon → Anthropic → AWS, and Google → Anthropic → GCP.
This creates impressive cloud growth headlines — Google Cloud +82%, for instance — but some fraction of that growth is circular capital flowing back to the investor. When enterprises evaluate cloud vendor stability, they need to ask: how much of this cloud revenue growth is organic enterprise demand, and how much is round-tripped venture capital?
If the circular loop tightens (OpenAI goes public, Anthropic diversifies infrastructure), that reported growth evaporates — and with it, the justification for continued capex at these levels.
The structural indicators are sobering. Moody's flagged that $140 billion in corporate bonds were issued by tech firms in H1 2026 alone — these companies are financing AI data centers with debt for the first time at scale. Industrial construction costs for liquid-cooled mega data centers have risen to $12.8 million per megawatt, up 34% from 2024 levels. Utility interconnect fees charged by regional power grid operators have tripled, requiring cloud operators to pre-fund substation upgrades costing $85 million per facility site. Data center completion timelines have stretched from 18 months to over 32 months due to environmental compliance delays and grid interconnect disputes.
None of these costs are being absorbed. They're being passed through — either in higher cloud prices today, or in service reductions tomorrow. For enterprise planners budgeting 2027 cloud AI spend, the input costs are structurally higher than any public pricing model currently reflects.
The Apple Signal
While the Magnificent Seven hemorrhages value, one company is thriving: Apple.
Apple avoided massive AI infrastructure investments, partnering with model developers instead of building its own. The result: shares up 23% in 2026, the biggest points contributor to the S&P 500's 8.3% rise. Up 15% in July alone — on pace for its best month in three years.
The market is rewarding capital discipline over AI ambition. That's a signal enterprise buyers should decode: the market no longer believes that unlimited AI spending produces proportional returns. If investors are right, hyperscaler economics must eventually re-balance — either through spending cuts or revenue extraction (higher prices).
What Happens If Wall Street Forces Cuts
The scenarios enterprise planners need to model:
Scenario A: Capex Continues Unchecked
- Cloud prices continue rising 10-40% annually on AI services
- New AI features ship rapidly but at premium pricing tiers
- Enterprise AI budgets grow 25-40% YoY just to maintain current usage
- Vendor financial health weakens (debt increases, margins compress)
Scenario B: Investor Pressure Forces 20-30% Capex Reduction
- New GPU cluster buildouts slow
- AI service availability tightens (rationing, waitlists)
- Feature development velocity decreases
- Current pricing holds but new capacity becomes scarce
Scenario C: Hyperscalers Shift to ROI-Driven Spending
- Non-revenue-generating R&D AI projects get cut
- Enterprise-proven services get priority investment
- Experimental/free-tier AI services get sunset (see: AWS killing Q Business and Kendra)
- Pricing stabilizes but innovation slows
Bold Wealth's Lemire is blunt: "There is going to be an AI winter at some point. When you look at how exceptional margins are — it is impossible to maintain those over a long timeframe."
Framework #1: Enterprise AI Cloud Pricing Risk Assessment
Score each dimension 1-5 to assess your organization's exposure to hyperscaler pricing volatility.
| Risk Dimension | What to Evaluate | Score (1-5) |
|---|---|---|
| 1. Cloud AI Spend Concentration | What % of your AI budget goes to a single provider? (>70% = high risk) | ___/5 |
| 2. Contract Lock-In | Are your AI services on spot/on-demand pricing or committed-use contracts? | ___/5 |
| 3. Inference vs. Training Split | Inference costs are rising fastest — what's your ratio? (>80% inference = elevated) | ___/5 |
| 4. Egress Dependency | How much data moves between cloud providers? (High egress = trapped) | ___/5 |
| 5. Alternative Infrastructure | Could you shift 30%+ of AI workloads to on-prem or alternative providers in 90 days? | ___/5 |
| 6. Budget Buffer | Does your AI budget include a 20%+ contingency for price increases? | ___/5 |
| 7. Vendor Revenue Quality | What % of your primary cloud vendor's AI revenue is organic vs. circular? | ___/5 |
| 8. Model Portability | Are your AI workloads coupled to proprietary APIs or running on open-weight models? | ___/5 |
Scoring:
- 35-40: Low pricing risk — diversified, portable, buffered
- 25-34: Moderate risk — address concentration and lock-in gaps
- 15-24: High risk — significant exposure to pricing volatility
- Below 15: Critical — single-vendor dependency with no escape path
Immediate Actions by Score:
| Score | Priority Action |
|---|---|
| 35-40 | Lock in 2-3 year committed-use discounts now while vendors are hungry for commitments |
| 25-34 | Build inference cost optimization pipeline; evaluate open-weight models for 30%+ of workloads |
| 15-24 | Diversify to 2+ providers within 6 months; build abstraction layer; add 25% budget buffer |
| <15 | Emergency: negotiate contract ceilings immediately; start on-prem evaluation; engage CFO on risk |
Framework #2: Hyperscaler Financial Stability Scorecard
Before committing to a multi-year cloud AI contract, assess your vendor's ability to sustain current service levels and pricing.
| Stability Factor | Alphabet/GCP | Microsoft/Azure | Amazon/AWS | What It Means for You |
|---|---|---|---|---|
| FCF Margin Trend | Negative (first time) | Declining | Stable | Negative FCF = pressure to monetize harder |
| Capex as % Revenue | ~45% | ~42% | ~38% | Higher ratio = more pressure to extract ROI |
| Cloud Revenue Growth | 82% (strong) | ~35% (est.) | ~25% (est.) | Higher growth = more organic demand justifying spend |
| Circular Revenue Risk | Moderate (Anthropic) | High (OpenAI) | Moderate (Anthropic) | Higher = more fragile growth story |
| Stock YTD Performance | Flat (post-drop) | -21% | Flat | Worse performance = more investor pressure to cut |
| Backlog | $514B | ~$300B+ | Not disclosed | Larger backlog = more customer lock-in |
| Debt Trend | Rising | Rising | Stable | Rising debt + negative FCF = credit risk signal |
Reading the Scorecard:
- GCP: Strongest cloud growth (82%) but first negative FCF is a red flag. If growth sustains, pricing may hold. If growth slows, expect aggressive monetization.
- Azure: Weakest stock performance (-21%) means maximum investor pressure. Watch for Azure AI price increases post-earnings next week.
- AWS: Most financially stable of the three (lower capex ratio, stable FCF). But already raising prices and sunsetting services.
The Enterprise Playbook: 5 Moves Before Year-End
1. Lock in pricing now — before Q4 earnings pressure hits. Microsoft and Meta report next week. If stocks drop further on capex concerns, Q4 will bring aggressive monetization. Negotiate committed-use agreements with price ceilings NOW.
2. Shift 30%+ of inference to open-weight models. Fireworks AI's $1.5B raise proves the market: 95% of their 40 trillion daily tokens run specialized open-weight models. Enterprise inference on Llama, Mistral, or DeepSeek costs 60-80% less than frontier API pricing — and it's portable.
3. Build your model-agnostic abstraction layer. When cloud pricing shifts, you need to move workloads without rewriting applications. The companies that built abstraction layers in H1 2026 are the ones that can negotiate from strength.
4. Budget for 25-40% AI cost growth in 2027. BCG's research on cloud AI costs confirms: total cost of ownership extends far beyond token pricing. Network, egress, storage, and inference costs are all inflating. Plan accordingly.
5. Evaluate on-prem inference for steady-state workloads. JPMorgan moved to on-prem AI inference and cut cloud costs by 87%. If your inference workloads are predictable and high-volume, the math is shifting in favor of owned infrastructure.
The Bigger Picture
The AI infrastructure buildout is entering its accountability phase. For three years, hyperscalers spent freely on the assumption that AI demand would eventually justify any level of investment. That assumption is being tested.
Google Cloud grew 82% — and investors still punished the company. The bar has moved from "show us growth" to "show us returns." When your vendors face that pressure, it flows downhill to your contract.
Gartner expects more than 40% of agentic AI projects to be canceled by end of 2027 on cost alone. The enterprises that survive the pricing correction will be the ones that diversified early, built portability, and negotiated from strength.
The tacit agreement between Big Tech and Wall Street is broken. The next agreement — between Big Tech and enterprise customers about who pays for $725 billion in infrastructure — is being written right now.
Make sure you're at the table.
The next 90 days are critical. Microsoft and Meta report next week. Amazon and Apple follow. If the pattern holds — strong revenue, punished stock — the investor revolt becomes a structural repricing of how Big Tech approaches AI investment. And enterprise cloud buyers who locked in favorable terms in H1 2026 will look prescient, while those who assumed pricing stability will face budget overruns that no amount of optimization can absorb.
Rajesh Beri is Head of AI Engineering at Zscaler. Views are his own.
