HOW MEETING.AI LEARNED TO BUILD FOR THE WORLDWhen a local AI startup goes head-to-head with global competitors that have raised hundreds of millions - while it hasn't raised a dollar in years - it's tempting to focus on the gap in war chests. But the more interesting fact is that there's still room to scale despite it, as Meeting.ai has found. The question the company now grapples with isn't how to raise more money. It's how to make an AI product feel fast and reliable wherever its users are. Kurniawan ran into that question on an otherwise ordinary day, and answering it changed how he thinks about where to place his bets. Growth might come when you're least preparedKurniawan wasn't expecting his experimental Instagram Reel in February to go viral, but it did. The video racked up millions of views within hours, eventually surpassing well over 10 million. Daily sign-ups on his platform jumped 100x from around 100 to 200 to over 10,000 overnight. Meeting.ai wasn't ready to cater to that kind of surge in traffic, so its system was quickly overloaded. Kurniawan estimates that around 3,000 users signed up and tried to use the platform, but it failed them. There wasn't enough compute. That's the thing with running an AI startup. An ecommerce company can serve hundreds, if not thousands, of users off a single server. But a platform like Meeting.ai can only host about eight at a time, he says. AI workloads are simply more demanding. "We ended up throwing money at the problem," Kurniawan recalls. In cloud hosting, there are two ways to add capacity: vertical scaling, that is, making existing machines bigger, or horizontal scaling, which means adding more machines. Horizontal scaling is cheaper and more resilient in the long run, but it requires advanced planning. Kurniawan had no choice but to pay for the expensive fix before he started investing in the more cost-effective solution. That month, the company's cloud bill doubled from US$15,000 to US$30,000. Its revenue doubled, too, so it wasn't a big problem. But for Kurniawan, the message was clear: A better plan needed to be put in place. Good thing he had an ace up his sleeve. He had existing relationships with multiple providers, whom he relied on for support during those critical hours. So, when he needed an emergency quota increase, he called his contacts instead of filling out some forms, which would have been a much slower process. Kurniawan has made the case to maintain such relationships, expanding them as he grows his platform. "There's no one provider that will provide you everything," he says. Kurniawan is now considering closing sign-ups and putting up a waitlist if traffic were to surpass compute capacity again in the future. That, in his view, is a better scenario than letting someone in for a broken experience. "At least there's still goodwill from people," he notes, comparing it to a full restaurant turning someone away politely, versus one that seats you anyway and serves a bad meal. Build the network around the userMeeting.ai wasn't planning for a global expansion when it started. "Initially, we didn't think of Meeting.ai for a global market," Kurniawan says. "We wanted it to be for Indonesia." Back then, he was limited by an outdated belief that his product couldn't compete globally, especially with rivals from places like the US, China, and India. When the company saw its first overseas transaction, Kurniawan brushed it off as pure luck. Meeting.ai took a year after release to double down on its global positioning. It only makes sense with the kind of numbers it was seeing. Now, the company counts four notable markets, in order: Indonesia, the US, Malaysia, and South Korea. Moreover, 60% of Meeting.ai's revenue comes from overseas, versus 40% from Indonesia. South Korean users are reportedly 4x more willing to pay than Indonesian users, with notably strong retention. But the numbers worth following go both directions - not just in terms of where the customer acquisition cost is low, but also where the cost for compute is competitive. Currently, Meeting.ai spends around US$20,000 per month on infrastructure, making it one of the company's biggest expenditures. Kurniawan uses data centers in the US, Johor, and Singapore, based on where the costs and available capacity make the most sense for its workloads. It's not just about price. Serving users across several markets means weighing compute costs against latency, reliability, and available capacity. A local team, a global productMeeting.ai's team of nine people - two in marketing, one in finance, and the rest engineering - is based in Indonesia. When asked whether he planned to hire members in other markets where users are present, Kurniawan simply says: "We haven't seen the need." This combination - a team concentrated in one country, customers scattered around the world, and servers spread across markets - only works because the underlying digital links make distance less relevant to day-to-day operations. The upside of this arrangement, according to Kurniawan, is speed and shared context. A small team in proximity can execute ideas without the friction of coordinating across offices. "You might gain some capability if you expand the team, but you lose contact," he adds. "That's a trade-off we've made." What makes that trade-off viable is AI doing the work previously done by extra headcount. Meeting.ai used to employ nearly 40 people, but it's now more productive with just nine, says Kurniawan. The team - himself included - runs much of its work through AI agents, juggling several coding agents in parallel across multiple open monitors so no single task slows it down.  That same lean approach shapes how Meeting.ai competes without outside capital. "We have a cost advantage from a talent standpoint," Kurniawan shares. His salary, although competitive by Indonesian standards, remains far beneath the US$100,000 annual baseline a US engineer commands. Kurniawan knows that some fights aren't winnable from Jakarta: Enterprise sales rewards proximity, and a founder with an Indonesian passport can't book a same-day flight to the US to win clients. Securing the visa alone takes weeks. "That part, we're going to lose," he says. So, Meeting.ai is playing in a battlefield where it could win: product-led growth. It has rolled out features that its competitors have yet to introduce, and they've proven to be quite popular. The platform's AI-generated visual notes have created a buzz, and its recently launched instant presentation drove conversions to double overnight upon release. "The product is the marketing," Kurniawan says, making the case that a strong product - not a bigger balance sheet - lets a Jakarta-based team compete with rivals raising in the hundreds of millions. Meeting.ai's case shows that the next stage of AI competition isn't determined only by who develops the most powerful model. It depends on how broadly businesses and workers can access those capabilities and incorporate them into everyday work. For a geographically dispersed market such as Indonesia, AI competitiveness is an ecosystem challenge: developing talent and digital skills, making computing and cloud infrastructure accessible, supporting innovation, and ensuring that reliable high-speed connectivity, like 5G networks, reaches businesses and communities beyond well-established tech hubs. |
