Databricks co-founder says cloud, consistency needed for Korea's AI ambitions

Home > Business > Industry

Databricks co-founder says cloud, consistency needed for Korea's AI ambitions

Published: 17 Sep. 2025, 15:20 Updated: 19 Sep. 2025, 09:15

LEE JAE-LIM
[email protected]

Audio report: written by reporters, read by AI

Arsalan Tavakoli-Shiraji, Databricks co-founder and senior vice president of field engineering, speaks during an interview with the Korea JoongAng Daily at Signiel Seoul, southern Seoul, on Sept. 12. [PARK SANG-MOON]

[INTERVIEW]

Nearly every major Korean corporation today proclaims AI adoption as a top priority. Yet many are struggling to deliver results, largely because they lack the ability to fully utilize their data or enforce proper governance.

Databricks, a San Francisco-based data and AI company, says it can help close that gap. By embedding its platform into corporate systems such as enterprise resource planning (ERP), the company aims to enable manufacturers to make smarter decisions — from improving yield management to strengthening predictive maintenance and enhancing demand forecasting. These capabilities, says the firm, can translate into major competitive advantages for Korea’s manufacturing-driven economy.

"That’s the type of partnership we want to form in Korea," said Arsalan Tavakoli-Shiraji, Databricks co-founder and senior vice president of field engineering, in an interview with the Korea JoongAng Daily in eastern Seoul. Tavakoli was visiting to attend the Chief Information Officer Forum, an annual event organized by Databricks Korea since 2024 that brings together local data leaders to exchange insights on governance and AI adoption. The company opened its Korea office in 2022.

Databricks, which pioneered the data lakehouse architecture, already counts LG Electronics, Hanwha, KT and most recently Dongwon Group among its Korean partners. Its role has been to modernize data platforms and break down silos so enterprises can establish unified governance. Following the surge in interest around AI, the company in August launched Agent Bricks, a toolkit for building enterprise-grade AI agents, and earlier this month raised $1 billion to advance research in AI and agentic infrastructure.

Park Jong-sung , chief information officer at Dongwon Group, left, and Arsalan Tavakoli-Shiraji, Databricks co-founder and senior vice president of field engineering, pose for a commemorative photo after signing an agreement to deploy Databricks Data Intelligence Platform within the Korean firm. [DATABRICKS KOREA]

"At the end of the day, it comes down to two things: data and governance," Tavakoli emphasized. "If you don’t solve those, you don’t have a prayer of getting AI right. Companies need to focus on breaking down silos and putting governance in place so data becomes truly consumable."

The following are excerpts from the interview, edited for length and clarity.

Q. Is Agentic AI a destination or just a phase?
A. I don’t think it makes sense to focus on technology for technology’s sake. The companies that succeed are the ones that start with a business problem and then figure out the best way to deliver a solution. Sometimes that involves agentic AI, sometimes it’s traditional machine learning–based automation. But if you ask these companies what they’ve achieved, they won’t point to the number of agents they’ve deployed. Instead, they’ll tell you: "Here’s a process that used to cost this much and take this long, and here’s how we automated it." Success is measured in outcomes—whether that’s hyper-personalization for users, better semiconductor yields or entirely new capabilities enabled by AI.

Arsalan Tavakoli-Shiraji, Databricks co-founder and senior vice president of field engineering, poses for a photo after an interview with the Korea JoongAng Daily at Signiel Seoul, southern Seoul, on Sept. 12. [PARK SANG-MOON]

What are the most common bottlenecks companies face when adopting AI?
There are two major ones. First is data and governance. If you don’t solve those, you won’t get AI right. Breaking down silos and putting proper governance in place is essential. Second, companies often treat technology as the end goal. In reality, success depends as much on people and processes as it does on the technology itself. If organizations fail to empower their people or define governance and automation processes properly, 99 percent of the time they won’t get the results they’re hoping for.

Any Korea-specific challenges?
Yes, two stand out. One, a lot of data is still kept on-premises. Moving that data to the cloud is critical, but it’s also complicated by national security considerations. Two, Korea’s large conglomerates face the added complexity of managing data consistently across global operations. Many have done well, but a lot of our conversations here focus on helping them lay down the right processes and people structures to build that foundation.

Will chipmakers and infrastructure providers also need to anchor themselves to platforms like Databricks?
Every technology wave starts with a pyramid. At the bottom you have infrastructure — right now dominated by Nvidia. Above that are the foundation model providers, many of which generate revenue but burn cash. At the top sit the applications, which today are small but will eventually drive the majority of value as AI adoption scales.

Infrastructure will remain important, but barriers to entry are so high that only a few companies succeed there. Most of the money, though, will move into applications across verticals and domains. Nvidia has done more than just hardware — they have CUDA, for instance — but moving further up the stack is hard. That’s why their strategy, and ours, is to partner. We work closely with Nvidia to ensure their hardware powers what we deliver to customers, which in turn drives more demand for their infrastructure.

From a supply chain perspective, what bottlenecks can Databricks help address for sectors like semiconductors, automotive or shipbuilding?

Three areas stand out. First is yield management: even a 0.1 percent improvement in semiconductor yields has a huge impact. Second is predictive maintenance and digital twins, which let companies get ahead of failures. Third is demand forecasting — being able to anticipate demand so fabs and supply chains can produce and deliver the right products at the right time. These use cases apply not only in manufacturing but also in automotive, retail and consumer goods.

That is why companies come to us. When firms lay out their three-to-five year vision, they often ask us to lean in with resources, thought leadership, technology roadmaps, training, and enablement. That’s what real partnership looks like: embedding resources, tailoring the product roadmap and helping them execute.

What about plans for going public?
We don’t intend to remain private forever. [But] Being private in such a fast-moving space like AI gives us the flexibility to move quickly and place bold bets that may take 12 to 18 months to pay off. Public markets often bring more friction and less appetite for that kind of risk. So we will go public, but we’re not sacrificing anything right now by being private and it allows us to move quickly in a way that has helped us innovate and better address the market.

BY LEE JAE-LIM [[email protected]]