Why Infrastructure Is the New Moat in MarTech

Before AI, the hard part of building a marketing technology company was the user interface. Design systems, drag-and-drop builders, pixel-perfect dashboards — these took years and large frontend teams to get right. The UI was the product, and the product was the moat.

That era is ending. According to Kashish Gupta, Co-CEO of Hightouch — the $1.2B composable customer data platform — AI has collapsed the difficulty of building interfaces while dramatically increasing the difficulty of building reliable infrastructure underneath them.

The thesis: every challenge is now an infrastructure challenge

When asked what he believes about MarTech that most people would disagree with, Kashish was direct.

“I think now shipping user interfaces is somewhat trivial and every challenge is an infrastructure challenge,” he said. “Making this work for hundreds of millions of consumers, making all the data warehouse pipelines work, making all the reliability and all that stuff work is really hard. So the challenge has shifted.”

This isn’t abstract contrarianism. Hightouch has watched it play out in their own product development. Designers are now shipping frontend code. PMs will be next. The people who can contribute to the interface layer has expanded dramatically because AI coding tools have lowered the barrier. But the infrastructure layer — data warehouse pipelines operating at the scale of hundreds of millions of consumers, real-time integrations across dozens of advertising platforms, feedback loop processing — remains stubbornly difficult.

What actually constitutes the hard infrastructure

The conversation revealed specific infrastructure challenges that don’t get easier with better AI models:

Running ads simultaneously across Facebook, TikTok, Snapchat, and Google — then collecting real-time performance feedback and automatically iterating on content — is a pipeline problem, not an AI problem. “Running that ad and getting the feedback back is not trivial,” Kashish explained. “We basically believe in building more of the right side because it’s going to be verticalized versus the generic platform.”

The “right side of the equation,” in his framing, is everything that happens after you have data: activation, distribution, feedback collection, and iteration. Most tools focus on the ingestion side — getting data in. The hard part is using it at scale in the real world.

Then there’s the heterogeneity problem. Each customer has a different database setup, different image assets, different engagement tools they want to integrate with. Engineers at Hightouch discovered that “agents would have loved for everything to just be a tool call” but in practice, “you have to give it a significantly narrower scope and really clear instructions of which tools to use when.” Infrastructure that handles this variety reliably is the moat.

The consolidation prediction

If UI is no longer the moat, many MarTech companies built primarily around interface quality will struggle to defend their position. Kashish’s prediction follows directly from his thesis.

“I do think there’s going to be a lot of consolidation in the space.”

The companies that survive will be the ones with deep infrastructure — the data pipelines, the integration networks, the feedback loops operating at consumer scale. Companies whose primary value was a polished dashboard or a slick drag-and-drop builder will find that competitive advantage evaporating as AI makes similar interfaces buildable in weeks.

This aligns with a broader pattern across SaaS: when AI collapses the cost of building any individual feature, the surviving companies are the ones sitting on proprietary data, hard-to-replicate integrations, or infrastructure that took years of production use to harden.

The build-vs-buy test

Kashish’s infrastructure thesis sharpened during a direct challenge. When presented with the possibility of recreating Hightouch’s capabilities by chaining together free AI pipeline tools — connecting a data source through an agentic process, then feeding it into a content generation tool — he didn’t dismiss the ingestion side.

“This does seem really useful. It’s very different from what I’m doing. It’s more on the ingestion side.”

But the activation side is where the real complexity lives. Building an always-on agent that iterates on content across multiple ad platforms, collects performance feedback, and automatically adjusts — that, he argued, requires infrastructure that most B2C brands can’t build in-house.

“At that point, you’re a tech company. Most B2C brands don’t have those kind of resources.”

FAQ

Why is building a marketing technology UI no longer a competitive advantage?

AI coding tools have dramatically lowered the barrier to building polished user interfaces. At Hightouch, designers now ship frontend code and PMs are expected to follow. The technical challenge has shifted from interface quality to infrastructure reliability — data warehouse pipelines, multi-platform integrations, and real-time feedback loops operating at the scale of hundreds of millions of consumers.

What is the hardest infrastructure challenge in MarTech today?

Running marketing campaigns simultaneously across Facebook, TikTok, Snapchat, and Google — then collecting real-time performance feedback and automatically iterating on content — requires integration infrastructure that doesn’t get easier with better AI models. Each customer has different database setups, different assets, and different tools, creating heterogeneity that demands hardened, production-tested systems.

How does Hightouch use composable CDP architecture for marketing?

Hightouch connects to a company’s own cloud data warehouse — Snowflake, BigQuery, Databricks — and layers segmentation, journey orchestration, real-time personalization, and AI modules on top. Companies keep their data in their own VPC. The marketing team gets self-serve access through Hightouch’s interface without duplicating data or relying on data engineering teams for every request.

Why will MarTech consolidation accelerate because of AI?

When AI makes it trivial to build polished user interfaces, companies whose primary competitive advantage was UI quality lose their moat. The surviving MarTech companies will be those with deep infrastructure — proprietary data pipelines, multi-platform integration networks, and feedback loops hardened through years of production use at consumer scale. Companies built primarily around dashboard quality will struggle.

Should B2C brands build or buy their marketing AI infrastructure?

Most B2C brands lack the engineering resources to build always-on iterative advertising systems that run across multiple platforms, collect real-time feedback, and automatically adjust campaigns. Building this in-house effectively means becoming a technology company. Hightouch’s argument: the ingestion side (getting data in) is increasingly commoditized, but the activation side (using data for marketing at scale) requires specialized infrastructure.

What does the shift from imperative to declarative marketing mean?

Current marketing is imperative: marketers manually configure “if customer does X, send message Y on channel Z.” Declarative marketing has the marketer state an end goal — “increase tennis product engagement” — and AI determines the best channel, timing, audience, and content. Kashish Gupta believes this shift is inevitable but building the infrastructure to make it reliable is the real challenge.

How are AI coding tools changing MarTech product development?

At Hightouch, designers are now shipping production frontend code using AI coding tools, and PMs are expected to follow. The company gives team members choice in which AI tools they use rather than mandating a single platform. This has expanded who can contribute to the interface layer while leaving infrastructure work — data pipelines, integrations, reliability — as the primary engineering bottleneck.

What infrastructure do AI marketing agents actually need to work at scale?

Production AI marketing agents need: a semantic layer providing structured metadata about the data warehouse, a verification system (smaller LLMs checking larger ones for action fabrication), multi-platform ad integrations for simultaneous campaign deployment, real-time feedback collection, and deterministic content assembly pipelines. Each customer’s unique setup adds heterogeneity that requires flexible but reliable infrastructure.

Why do most companies focus on data ingestion instead of data activation?

Data ingestion — getting data into a warehouse — is increasingly commoditized through AI-powered ETL tools. But the harder and more valuable side is activation: running marketing against that data set across multiple channels with real-time feedback. Most tools focus on ingestion because it’s a more tractable engineering problem. Activation requires deep integrations with advertising platforms, content systems, and feedback loops.