David Yakobovitch: Beneath the Stack

David Yakobovitch is the founder of DataPower Capital, a venture capital fund based in New York City, with investments in companies such as OpenAI, Anthropic, and ElevenLabs.

Today, the firm focuses on the foundational layer of the AI stack — the systems that power modern AI, from cloud infrastructure to automated systems — based on a core thesis: while most attention goes to applications, the real leverage sits deeper in the stack.

“OpenAI and Anthropic wouldn’t exist without the infrastructure beneath them,” he says. “That’s where the real opportunity is.”

Where It Started

Yakobovitch’s path into venture capital didn’t begin with investing — it began with building.

Long before DataPower Capital, he was already embedded in the startup ecosystem — developing early products and experimenting with ideas that, in some cases, would later resemble category-defining companies.

“I’ve always had this startup blood,” he says. “Even back in my early days, I wanted to either build companies or support entrepreneurs on their mission.”

That instinct led him into operating roles across multiple startups, where he worked across data science, product, and go-to-market — gaining firsthand experience in how companies are built and scaled.

Along the way, he began angel investing — writing early checks, spending time with founders, and developing conviction around the kinds of companies he wanted to support.

What began as a side activity eventually led to the creation of DataPower Capital.

Betting on the Layer Beneath AI

Yakobovitch’s early investments were in data developer tools — the infrastructure that enables modern software. At the time, this was still a relatively underexplored space.

Working closely with infrastructure companies — and later inside Google — gave him a different perspective on where the market was heading.

“We started by investing in data dev tools, the picks and shovels for building software,” he explains. “And what we saw over time was that the market was unifying very quickly in the cloud.”

This led to a broader realization: the real leverage in AI isn’t just in models or applications — it’s in the systems that make them possible.

In the previous generation of software, infrastructure was largely manual — engineers had to configure environments, manage compute, and deploy applications themselves using tools like Docker and Kubernetes. What changed was automation: cloud platforms began provisioning everything in real time, making modern AI systems possible at scale.

“Companies like OpenAI and Anthropic are built on top of this infrastructure layer,” Yakobovitch says. “They wouldn’t be possible without cloud systems, chips, and automated provisioning already in place.”

From Static Software to Always-On Intelligence

At the core of Yakobovitch’s worldview is a shift from static software to real-time, always-on systems.

“We’re moving into a world where devices are always on,” he says. “You’ll have chips everywhere — on your wrist, in your glasses, maybe even contact lenses — constantly surfacing real-time insights.”

In this environment, software is no longer something you open and close. It becomes ambient — always running, learning, and assisting.

As these systems integrate more deeply into everyday workflows, they begin to play a larger role in decision-making.

“Some of those decisions will be human-led, and others will be AI-led — or collaborative between the human and the AI,” he says.

Yakobovitch explored this shift early by launching HUMAIN, one of the first podcasts focused on responsible AI.

The Misunderstood Nature of AI Progress

One of the biggest misconceptions about AI, Yakobovitch argues, stems from early experiences — particularly around the ChatGPT moment in late 2022.

For some users, it felt magical. For others, unreliable.

“A lot of people tried it in 2022 or 2023 and made up their mind,” he says. “If it didn’t work well, they just assumed it was all hype.”

What many underestimated was how quickly the technology would improve.

Models that were only partially reliable just a few years ago — often in the range of 12–20% accuracy — have rapidly evolved to levels approaching 90–95% or higher, with significantly stronger reasoning, memory, and iteration capabilities.

The pace of progress, he argues, has been far faster than most expected — turning what once felt unreliable into systems that are now highly capable in real-world use.

That shift is already redefining productivity.

Earlier estimates from McKinsey & Company suggested AI might improve knowledge worker productivity by 10–30%.

“I was shouting at companies, saying: no — I think there are people who are going to see somewhere between 100% and 1600% or more productivity improvements,” he says.

The gap between perception and reality is closing — much faster than most expect.

Why Domain Experts Will Win

Despite the dominance of large AI labs, Yakobovitch does not expect value to concentrate entirely at the top of the stack.

Instead, he sees a different pattern emerging — one driven by domain expertise and proprietary data.

“When you’re a domain expert, you have a strong data moat,” he says. “You understand the nuances of your data in a way that general models don’t.”

This creates an opportunity for highly specialized companies to outperform general-purpose systems — not by building better models, but by applying them more effectively.

That dynamic is already playing out. Companies like Harvey AI in legal or ElevenLabs in voice have built strong positions by focusing deeply on a single domain — before expanding outward.

The pattern is consistent: narrow focus first, expansion later.

The New Bar for Founders

As AI reduces the cost and complexity of building software, it is fundamentally changing how companies are created — and what is expected from founders.

What once required large teams can now be done by individuals or small groups, often in days.

“Today, you can build something in a weekend that would have taken a team of engineers before,” Yakobovitch says.

As a result, early-stage founders are no longer judged solely on vision. They are expected to show working products, early users, and clear signs of demand much earlier in the process.

Speed has become a defining trait.

In modern AI companies, product development has shifted from slow, periodic releases to continuous shipping — with teams deploying updates weekly or even daily.

“The best teams move fast. They’re constantly iterating, constantly shipping, constantly pushing forward,” he says. 

The AI Cycle: Early, Mature, and Overhyped — All at Once

Yakobovitch describes the current state of AI as unfolding across multiple timelines.

“It’s a mix of all three,” he says. “Some parts of the market are early, some are mature, and some are already overhyped.”

Certain areas — such as infrastructure, compute, and reshoring — remain in early stages, requiring long-term investment and coordination.

For example, the U.S. currently produces only around 12% of the world’s chips, but projections suggest that could rise to as much as one-third of global supply over the next 10 to 15 years if reshoring efforts succeed.

Elsewhere, parts of the market are moving much faster.

Yakobovitch believes the AI application layer is still early — but accelerating rapidly, driven by dramatically lower barriers to building software.

In some cases, individuals with no technical background can now build functional products in a single weekend — something that would have previously required months of work and significant capital.

The result is a highly uneven landscape — where different parts of the AI stack evolve at completely different speeds, creating both dislocation and opportunity for founders and investors.

The Next Decade of AI

AI is no longer confined to software — it is beginning to reshape entire industries.

Yakobovitch points to recent examples where companies that failed to adapt were quickly displaced, losing relevance almost overnight as models like ChatGPT improved — in some cases wiping out entire categories.

“The companies that don’t adapt will struggle,” he says. “But the ones that recognize the shift and pivot can still win.”

This reflects a broader transition from traditional SaaS to AI-native companies.

AI-native products are faster to develop, cheaper to scale, and often deliver better outcomes — giving them a structural advantage over legacy software.

But this is not a simple replacement cycle. Companies that recognize the shift and make a hard pivot to AI can survive — and in some cases, even thrive — as seen with incumbents like Google, Adobe, and Salesforce.

Yakobovitch sees this not as a temporary wave, but as a structural transformation. Over the next decade, as much as 25–50% of today’s public companies could be replaced by AI-native businesses.

“For some, it’s frightening — especially if you’ve been in the industry for decades,” he says. “But for others, it’s the most exciting moment in years.”