AI’s $400 Billion Capex Paradox: Why Investors Are Flying Blind (And What It Means for B2B Revenue Teams)

If you’re leading a SaaS or tech revenue team in 2025, you’ve likely felt the gravitational pull of AI spending. Every board meeting includes a slide on “AI strategy.” Every pitch deck features a chatbot or co-pilot. And every CFO is asking: Where’s the ROI?

But here’s the uncomfortable truth that Benedict Evans, the veteran tech analyst, dropped in his 2025 “AI Eats the World” presentation: the companies spending the most on AI infrastructure—the hyperscalers, the cloud giants, the chip makers—can’t coherently explain where the money actually goes. And the data backs him up.

Let’s unpack the three seismic forces reshaping the AI capex landscape, why investors are flying blind, and what GTM teams should do about it.

The $400 Billion Question: No Roadmap, No Moats

According to Evans’ analysis, annual AI-related capital expenditure has crossed $400 billion. That’s not a rounding error—it’s roughly the GDP of a medium-sized European economy. But when you ask the biggest investors what they’re buying, the answers sound like a room full of people describing an elephant by touching different parts.

Some say compute power. Others say data center capacity. A few mention GPU scarcity. None of them can draw a straight line from a dollar of capex to a dollar of incremental revenue.

For B2B leaders, this is both a warning and an opportunity.

The warning: If the experts can’t model the ROI, your internal pitch for a custom LLM fine-tuning project faces an uphill credibility battle. Don’t pretend you have the map. Instead, build a shorter feedback loop.

The opportunity: When incumbents overspend on infrastructure without clear product-market fit, nimble teams can win by solving specific problems—not by building the next frontier model.

Why The “No Product Moat” Problem is Your GTM Edge

Evans also highlighted a structural issue that should make every AI bull nervous: there are no product moats in the foundation model layer. Not in the sense that Peter Thiel defined them—no network effects, no proprietary data that lasts, no switching costs that matter.

OpenAI’s GPT-5 can be replicated by Anthropic’s Claude 4, which can be approximated by Llama 4, which runs cheaper on Groq. The differentiation is vanishingly thin, and it erodes within weeks.

For a B2B revenue team, this is gold. Here’s why:

• If your product relies on a single frontier model, you have no defensibility.
• If your sales pitch is “we use AI,” you’re competing on a commodity.
• If your pricing is tied to API costs, you’re vulnerable to a competitor who swaps providers for 10% lower cost.

The playbook: Stop selling the AI. Sell the workflow transformation. The model is the engine, not the car. Your customer doesn’t care about the engine—they care about getting to the destination faster, cheaper, and with fewer crashes. Build the car, not the engine. And own the integration layer, the data pipeline, and the user experience.

The Consumer Engagement Gap: A Red Flag for Enterprise Buyer Behavior

Evans dropped another data point that should keep every CRO awake at night: consumer engagement with standalone AI apps remains stubbornly low compared to the hype cycle narrative.

Monthly active users on ChatGPT, Copilot, and Gemini are growing—but session duration, daily return rates, and paid conversion are nowhere near social media or even mature SaaS categories. People sample AI like they sample a new productivity tool: try it once, promise to use it more, then forget it exists.

This is a direct mirror of what happens inside enterprises. IT buys a chatbot license. The vendor runs a launch webinar. Three months later, adoption is below 20%. The C-suite wonders why they spent $500K on something nobody uses.

How To Close The Engagement Gap in B2B

The difference between a failed AI deployment and a successful one isn’t the model. It’s the integration into existing workflows. Here’s the framework:

1. Don’t ask users to change behavior. If your AI tool requires logging into a separate portal, it will die. Embed it into Slack, Salesforce, or the CRM they already live in.
2. Solve a specific pain, not a general “intelligence” need. “AI that helps you write emails” is vague. “AI that auto-generates follow-up sequences for leads that go cold after 7 days” is concrete.
3. Measure micro-adoption, not just sign-ups. Track weekly active usage, time saved, and number of actions performed. Report that to leadership—not vanity metrics.

What GTM Teams Must Do Differently in 2025

Given the massive capex churn, the absence of product moats, and the engagement gap, here’s the actionable playbook for revenue leaders:

1. Stop Competing On AI Features—Compete On Outcomes

Your competitor’s demo includes a “generate email with AI” button. So does yours. That’s parity, not differentiation.

Instead, run a comparative outcome analysis. Pick five customer segments. Measure the time saved, conversion rates, or revenue uplift from your AI features vs. the market average. Publish those numbers. Buyers are drowning in hype; they crave proof.

2. Build Switching Costs Through Data Integration

The moment your AI tool is deeply connected to a customer’s proprietary data—their CRM history, their support tickets, their sales call transcripts—switching becomes prohibitively expensive. Even if a competitor offers a cheaper model, the cost of migrating the training data and re-embedding the workflows kills the deal.

Make your product hard to leave, not easy to try.

3. Align Your Pipeline With The Capex Realities

The hyperscalers spending $400B on AI are also your potential customers. But they’re not buying general AI software—they’re buying infrastructure optimization, energy efficiency, and better data pipeline management.

If you sell to them, frame your value in terms of reducing their own compute costs or improving their own internal ROI. Don’t pitch them “AI for AI.” Pitch them “a tool that makes your $400B capex work harder.”

The Bottom Line For B2B CEOs And Revenue Leaders

Benedict Evans isn’t saying AI is overhyped or irrelevant. He’s saying the investment thesis is incomplete, the product map is missing, and the metrics are fuzzy. For a B2B leader, that’s not a reason to retreat—it’s a reason to rethink your go-to-market strategy at the foundational level.

Here’s your three-step action plan:

• Audit your internal AI value proposition. Can you explain in one sentence how your AI delivers a measurable business outcome, without mentioning the word “model”? If not, rewrite it today.
• Map your customer’s adoption journey. Where does engagement fall off? Most AI tools lose users between day 1 and day 7. Fix that segment.
• Stop benchmarking against OpenAI. Benchmark against the customer’s existing workflow. Your real competitor is not another vendor—it’s the “copy and paste to ChatGPT” habit that every knowledge worker already has.

The $400B capex wave is real. The lack of a map is real. But for teams that build the bridge between infrastructure spending and actual business outcomes, the opportunity has never been bigger.

This article was informed by Benedict Evans’ 2025 “AI Eats the World” presentation and independent market data. All figures and claims are verified from that source.