In the first quarter of 2026, the four largest hyperscalers — Google, Microsoft, Amazon, and Meta — spent more than $130 billion combined on data centers and AI infrastructure. In ninety days. The numbers are auditable in their SEC filings and grounded in the Bloomberg coverage of every earnings cycle. The headlines treat this as a tech-industry story. It is mostly a marketing-industry story, and the second-order effects on email are going to be larger and stranger than anything we've planned for.
Here's the macro-to-micro version: the AI infrastructure email impact isn't some abstract future thing. The compute being poured into the ground right now is going to land in your HubSpot portal over the next 18 months in concrete ways. Some of it will be great for marketers. Some of it will be quietly destructive. Most of it will be invisible until it isn't.
To make this concrete: the Q1 2026 capex from the four largest cloud providers is funding roughly three categories of buildout, all of which matter for email:
That third bucket is the one most marketers haven't priced into their planning yet. Let's start there.
Here's an asymmetry worth sitting with: the AI infrastructure being built right now is being deployed on the receiving end before it's being deployed on the sending end. Mailbox providers have larger budgets, tighter user-facing feedback loops, and a clear business reason to invest. They're shipping inbox AI that does three things that materially change your job:
The implication: email is no longer a direct conversation with a human. It's a conversation mediated by an AI that's deciding what to surface, summarize, and suggest. Writing for that mediated environment is a different craft than writing for an inbox in 2019 — and it's part of why open rates post-MPP have stopped meaning what marketers think they mean.
Working forward from the capex, here are the specific changes I'd expect to see hit B2B email marketing over the next eighteen months. Some of these are happening now. The rest are the trajectory.
The dominant cost of personalization today is engineering and data, not compute. As inference gets cheaper, the math changes. Generating subject-line variants per recipient, scoring content blocks per persona, and dynamic body insertion all become economically defensible for mid-market senders, not just enterprise. Expect HubSpot, Marketo, and the rest to ship features that assume this cost curve.
The AI that watches your sender reputation, predicts engagement drops, and flags anomalies is currently sold as a premium add-on. With cheaper inference, this becomes table stakes — the way SPF/DKIM/DMARC checks for HubSpot sender reputation moved from "specialist feature" to "checkbox in setup" over five years. The vendors who don't ship it natively in 2027 will lose deals.
The autonomous-agent vision keeps getting deferred. What's actually shipping is generative AI inside the composer — better suggestions, smarter rewrites, faster localization. The capex makes these features fast and cheap enough to integrate into every editor by default. Expect this everywhere within twelve months.
"Best time to send" tools that pick one window per audience are going to look as quaint as desktop-only email previews. Cheap inference makes per-recipient modeling the default, which is the architecture our scheduler has used since 2019 and is now being adopted across the category — with the measurable proof points for send-time optimization now widely documented. The Campaign Orchestration release we shipped in February 2026 is a preview of where the rest of the market is headed.
Just as SEO emerged when search engines started ranking pages, "mailbox AI optimization" is the discipline emerging now. Writing copy that summarizes well, structuring CTAs that survive prioritization, formatting emails that reply-suggestions render correctly — this is the next layer of craft. The marketers who get good at it early will have a window of asymmetric advantage.
Three concrete budget moves worth making now, in priority order:
Worth saying clearly: a meaningful fraction of the $130B in AI capex is going to be revealed, in retrospect, as overbuilt. That's how every infrastructure boom works — railroads, fiber, cloud, AI. Overbuilding is the mechanism by which the cost curve drops far enough for normal businesses to use the technology.
For email marketers, this is unambiguously good news. You don't have to fund the buildout. You get to use the depreciated infrastructure once the cycle prints capacity into the ground. The job is positioning your stack and your team to take advantage of it as the cost curve drops, not predicting which hyperscaler wins.
The sending economics won't change much — SMTP delivery is already commoditized. What changes is the per-send economics of AI features layered on top of sending: personalization, send-time optimization, deliverability monitoring. Those drop in cost meaningfully as inference gets cheaper, which is how features that were enterprise-only move down-market.
Concerned, but not panicked. Gmail's summarization is real and affecting engagement now, but the fix is craft, not technology: write tighter copy, lead with the action, and design CTAs that survive summarization. Marketers who adapt will outperform those who don't, and the adaptation isn't expensive.
For generative tools, possibly — prices are dropping and features are improving fast enough that buying now means buying obsolescence. For infrastructure AI (deliverability, send-time, orchestration), buy now — the production track record matters more than the price curve, and the compounding benefit starts the day you turn it on.
In some categories, yes. In email specifically, no — the domain expertise required to build deliverability-grade AI is not in the AWS or Azure playbook, and the data depth required is unique to vendors who've been in the space. Expect consolidation among marketing-AI vendors, not displacement by hyperscalers.
The AI capex story is the kind of macro trend that's easy to ignore because it doesn't show up in your weekly metrics — until it does, all at once, when a mailbox provider ships a feature that reshapes your engagement curve overnight. The marketers who plan for the trajectory get to surf it. The ones who don't get washed.
If you want to see how recipient-level AI orchestration changes the math in your HubSpot portal, the free trial of Seventh Sense takes about 15 minutes to set up and starts modeling individual recipient behavior on day one. You'll see the kind of AI that's already crossed the gap from capex to results, in your own data.