The AI trade is broadening. The Magnificent Seven committed $725 billion in capital expenditure for 2026 to build infrastructure nobody else could afford. Now the S&P 493 is capturing the value — at a pace the telecom bust never achieved. The staked position: the S&P 493 outperforms the Mag 7 by the end of 2027.
The Magnificent Seven committed $725 billion in capital expenditure for 2026 — the largest single-year infrastructure investment by any group of companies in history. OpenAI alone is projected by Deutsche Bank to accumulate $143 billion in negative free cash flow from 2024 through 2029. This journal has covered that story — The Audit, The Margin, The S-1, The Negative Margin — and it remains true. The builders are destroying capital at a pace that rhymes with the telecom bust of 2001, when hundreds of billions in fiber-optic cable were laid and 95 percent of it went dark within four years.
But here is where the analogy breaks, and where this entry stakes a position the journal has not yet taken: the customers are capturing value at a speed the telecom cycle never produced. The fiber went dark because nobody had applications that needed it. The AI compute is being consumed as fast as it is built, and the companies consuming it are not startups burning venture capital. They are the other 493 companies in the S&P 500, and they are printing money.
The Customer Evidence
JPMorgan Chase deployed hundreds of AI use cases across more than 150,000 employees and reported two billion dollars a year in quantifiable savings. Walmart cut per-unit fulfillment costs by 20 percent and shipping costs by 30 percent through AI-driven logistics. Daimler Trucks, working through SAP's AI tools, watched its bid win rate climb from 10 percent to 40 percent — an additional 70 million euros in annual revenue from a single operational improvement. Duolingo increased content output tenfold while expanding gross margins by 190 basis points. IBM reached a $4.5 billion annual run-rate in productivity savings driven by AI and automation.
These are not pilot programs. One in four S&P 500 companies reported at least one quantifiable AI impact in the first quarter of 2026, up from 13 percent a year earlier. In financial services, 40 percent of firms reported measurable gains — nearly triple the rate from 2025. Deloitte's latest enterprise AI survey found the majority of large organizations now have at least one AI workload in production. The customer-side story is not a forecast. It is a set of audited earnings statements.
Why This Is Not Telecom
The 1990s fiber-optic buildout and the 2020s AI buildout share a capital structure: a small number of infrastructure companies spending far beyond what their own revenues can justify, betting that demand will grow into the capacity they are creating. In telecom, it did not. The first internet applications — email, static web pages, early e-commerce — needed a fraction of the bandwidth that had been laid. By 2001, only 5 percent of installed fiber was lit. Breakeven took ten to fifteen years. WorldCom's destruction of $180 billion in shareholder value was the headstone, but the real failure was demand that never materialized at the pace the supply required.
The AI cycle has produced the same supply-side excess. But the demand side is structurally different. AI compute, unlike dark fiber, has an immediate use case in every company that processes language, images, or structured data — which is every company. The per-employee spend on AI infrastructure rose 50 percent in the past year to $2,068, according to the Atlanta Federal Reserve. Finance-sector firms reported average payback periods of eight months on agentic AI systems. The fiber sat in the ground because the applications did not exist yet. The compute is being consumed because the applications already do.
The distribution of that value is what makes this cycle unusual. PwC found that 74 percent of AI's economic value is captured by the top 20 percent of companies deploying it, with leaders generating 7.2 times more gains than laggards. This concentration is consistent with the early stages of every major technology adoption — but it means that the winners are identifiable now, and they are not the companies building the infrastructure.
The Convergence
The earnings gap between the Magnificent Seven and the rest of the S&P 500 has been narrowing since 2024, when it stood at roughly 30 percentage points in earnings-per-share growth. By 2025 it had compressed to about 7. Goldman Sachs projects 4 percentage points by the end of 2026. BlackRock's Investment Institute sees 3 percentage points by 2027.
But the headline numbers still obscure the speed of the shift. Strip out NVIDIA — whose $81.6 billion quarter masks the deceleration of the other six — and the remaining Mag 6 grew earnings just 6.4 percent in Q1 2026. The S&P 493 grew at 10 percent. The crossover may have already happened.
All eleven S&P 500 sectors posted positive earnings growth simultaneously for the first time since 2021. The S&P 500's net profit margin reached 13.4 percent, the highest since FactSet began tracking the metric in 2009. This is not rotation into a few replacement darlings. It is broad-based earnings growth powered by companies deploying AI tools built by others at others' expense.
The Position
The staked claim: the S&P 493 outperforms the Magnificent Seven in total return by the end of 2027.
The mechanism is not a crash. The Mag 7 will continue to grow revenues and may avoid the catastrophic unwind that killed WorldCom and Global Crossing. The mechanism is a repricing. Markets have valued the Mag 7 as technology companies — high-multiple, high-growth, asset-light. But $725 billion in annual capital expenditure is not asset-light. It is the capital intensity of a pipeline operator, a railroad, or a regulated utility. The market will eventually price these companies for what they have become: the capital-intensive infrastructure layer of the AI economy, earning utility-like returns on massive fixed assets, while the companies that run software on that infrastructure earn the software-like margins.
The falsifiable conditions: if enterprise AI production deployments plateau below 85 percent, the demand thesis weakens and the builders' capital destruction may not be offset by customer gains. If the Mag 7 earnings gap re-widens in the second half of 2026, the convergence thesis fails on its own timeline. If NVIDIA's dominance persists and the other six resume high growth, the ex-NVIDIA decomposition is misleading.
But the central bet is that the AI capex cycle follows the same iron law as every prior infrastructure buildout: the builders overspend, the users capture the surplus, and the market eventually prices the difference. The telecom builders did not survive long enough to see the value they created. The AI builders are better capitalized and will likely survive — but survival is not outperformance. The companies deploying AI into mortgage origination, supply chains, insurance claims, and content production do not need to spend $725 billion. They need to spend $2,068 per employee. That is the diffusion.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)