Has AI Peaked?

Global markets are locked in a fierce debate between veterans and newcomers, tech bulls and traditionalists.

 

The global economy is strictly K-shaped:

• The Upper Arm: The "silicon-based industry" represented by semiconductors, electronics, and communications, where narratives, data, and earnings profiles continue to be revised upward.

• The Lower Arm: The "carbon-based industry" driven by real estate, infrastructure, and legacy consumer segments, where demand and earnings are a complete mess.

This divergence is the inevitable result of global macroeconomic realignments and supply chain restructuring.

 

For individuals, success lies in recognizing these macro currents rather than fighting the market. Following the recent sharp drop, tech stocks look attractive again. Investors are strongly encouraged to use this pullback to balance their portfolios back toward technology.

 

Assessing the AI industry trend comes down to just two core metrics:

1. Liquidity: Which dictates valuations and market financing for compute equities.

2. Capital Expenditures (Capex): Which serves as the primary engine for earnings growth.

As long as Cloud Service Provider (CSP) capex projections continue to be revised upward and the Fed does not pivot toward rate hikes, one should be very cautious about calling a peak in AI.

 

Current market estimates model 2026 CSP capex at $720 billion, scaling toward $900 billion for 2027. Under this timeline, the third quarter of this year will likely be the peak of capex growth velocity.

 

However, historical data over the trailing three years shows that the market has systematically underestimated actual capex growth by 30 to 40 percentage points. Morgan Stanley's estimate for 2027 AI capex has already reached $1.2 trillion to $1.4 trillion, leaving substantial room for upward revisions.

 

Capex velocity is fundamentally determined by the Return on Investment (ROI). ROI directly tracks the Annual Recurring Revenue (ARR) of foundational model developers, which in turn is tied to the scaling speed of model intelligence.

 

When the "AI bubble" narrative spiked late last year, the core catalyst was a slowdown in Q3 ARR. But as technology crossed critical performance thresholds in December, ARR inflected exponentially, capex expectations revised up again, and AI enjoyed a powerful Davis Double-Play of expanding valuations and surging earnings.

 

The ARR trajectory since the start of this year highlights this scaling:

• January: ARR hit $14 billion, recording a net monthly expansion of $5 billion.

• February: ARR reached $19 billion, maintaining a $5 billion monthly net add.

• March: ARR scaled to $29 billion, marking a $10 billion monthly net add.

• April: ARR accelerated to $40 billion, up $11 billion sequentially.

• May: ARR printed at $54 billion, logging a $14 billion net monthly expansion.

• Mid-June: ARR reached $62 billion, tracking toward a projected $70 billion month-end close—implying a net monthly expansion of $16 billion.

Anthropic’s latest run-rate ARR has scaled to $62 billion, marking a 15-fold expansion over the trailing twelve months. Concurrently, OpenAI’s run-rate ARR is tracking near $39 billion to $40 billion.

 

Together, these two pacesetters command an aggregate ARR profile of $100 billion. Given that their combined ARR stood at just $20 billion at the end of 2025, it has expanded 5-fold in under six months, with a high probability of breaching the $200 billion milestone by year-end.

 

A recent research report from analytical firm Exponential View confirms that AI monetization is crossing a critical inflection point. Real enterprise demand is beginning to provide concrete economic justification for the massive capital deployments directed into data centers and hardware architectures over recent years.

 

In the first quarter of 2026, global AI-generated revenue reached $25 billion. This marks the second consecutive quarter where top-line monetization surpassed the estimated depreciation costs tied to data center and silicon investments (which currently track around $21 billion).

 

Historically, the market questioned whether downstream customer demand could support the massive data center build-out. Because market leaders like OpenAI and Anthropic are private, demand-side metrics remained difficult to quantify, forcing the market to rely on supply-side proxies.

 

Now, by tracking over 1,000 enterprise AI spending allocations, Exponential View demonstrates that the AI complex generated $110 billion in top-line revenue over the trailing 12 months—a adoption pace roughly three times faster than historical IT waves like the internet, mobile apps, and early cloud computing.

 

Whether analyzing the exponential ARR scaling of foundational platforms or institutional tracking across broader corporate ecosystems, the data confirms that AI has crossed its critical adoption threshold. Current ARR profiles are beginning to align with upstream capex budgets.

 

At the frontier layer, advanced model iterations continue to accelerate. While previous boundaries or restrictions heightened market anxieties, the sequential lifting of these constraints offers a clear catalyst to monitor for further ARR acceleration.

 

During a recent shareholder presentation, Jensen Huang reiterated this structural mechanic: When AI becomes capable of executing useful work, token generation gains intrinsic value.

 

As tokens begin generating concrete bottom-line profits, the demand for compute capacity accelerates. Practical AI has arrived; the debate surrounding AI return on investment has been settled. Every industry vertical is now racing to deploy agentic AI architectures.

 

The recent technical pullback across tech heavyweights has been blamed on fears that expanding margins within the memory vertical will squeeze cash positions at major cloud providers, ultimately crimping forward AI capex expectations.

 

However, looking at the architecture of US tech heavyweights, market leadership has historically been driven by high-margin pricing power—as seen across Nvidia, Apple, Alphabet, ASML, and Microsoft.

 

While capital allocation was previously concentrated within downstream application layers and endpoints, economic returns have naturally consolidated within upstream infrastructure suppliers. As the technology lifecycle matures, pricing power will eventually recalibrate across the value chain according to standard economic laws.

 

Major cloud service provider capex budgets are unlikely to contract due to near-term memory pricing pressures, supported by two core dynamics:

 

First, out of the $2 trillion in backlog orders logged by major hyperscalers last quarter, roughly half stems directly from OpenAI and Anthropic commitments.

 

As long as these foundational platforms maintain their current ARR expansion, this massive contractual pipeline offers a durable floor for sustained capex growth.

 

Second, the structural ROI model has been thoroughly validated. Operational metrics indicate that 1 gigawatt (GW) of compute capacity can generate between $30 billion and $40 billion in annualized revenue. Given that the annualized cost structure of a 1GW data center sits near the $10 billion threshold, the underlying return profile remains highly compelling.

 

The upcoming July-to-August corporate earnings season will offer crucial visibility into CSP enterprise AI metrics, hyperscale cloud backlogs, and forward capex guidance—all of which are highly likely to beat consensus estimates and trigger upward revisions.

 

For allocators focused on mid-term free cash flow constraints, viewing the ecosystem through static metrics misses the broader financing dynamic.

 

Hyperscalers maintain frictionless access to debt and equity capital markets to fund infrastructure rollouts, while impending public listings from major foundational players will unlock alternative liquidity pools. As AI operations achieve greater scale toward 2028, operating cash flows will naturally realign.

 

Since the initial phase of the AI expansion cycle spanning 2023 through 2026, the market has routinely cycled through bearish narratives warning that outsized upstream profit concentration—first in GPUs, now in memory silicon—would force downstream capex cuts.

 

Each time, these anxieties have been systematically dismantled during blockbuster earnings seasons, allowing the core AI expansion narrative to advance.