MC INTERVIEW Roblox’s Anupam Singh on building for India’s tier-2, tier-3 users as AI reshapes the platform
In an interview with Moneycontrol, Roblox SVP of Engineering Anupam Singh discusses the platform’s efforts to improve the user experience in India, AI’s potential to make game creation more accessible, and its focus on surfacing games that drive long-term retention.
Vikas SN
September 09, 2026 / 10:33 IST
Anupam Singh, SVP of Engineering, Roblox
Roblox, the popular user-generated gaming platform, has spent years trying to make game creation accessible to anyone. Now, generative AI is opening up new possibilities for how games are created on the platform.
India-origin Silicon Valley veteran Anupam Singh is at the centre of these efforts.
A computer science graduate of Pune University, Singh leads Roblox’s AI and infrastructure initiatives as Senior Vice President of Engineering, reporting directly to co-founder and CEO David Baszucki.
India is among Roblox’s top and fastest-growing markets, making it a crucial region for the creator-led platform that now reaches 123 million daily active users globally.
In an exclusive interview with Moneycontrol, Singh said Roblox is stepping up investments to make the platform work well on intermittent networks and lower-powered phones, ensuring a “great experience” even for users in tier-two and tier-three cities.
This includes upgrading its existing Mumbai datacentre and potentially evaluating more infrastructure investments in India as the platform scales.
In the interview, Singh also spoke about Roblox’s AI strategy, surfacing games that drive long-term retention rather than short-term engagement, and AI’s potential to help creators build complex games. Edited excerpts:
When you joined Roblox nearly five years ago, AI was far from central to the conversation. How has Roblox evolved since then, and how has AI changed the platform?
The reason I joined Roblox was because of AI. There is so much interaction on our platform that understanding what users will do next was a big opportunity.
At the time, it wasn’t called AI. It was called machine learning. But those were the large models of their time. Today, they wouldn’t be considered large models anymore.
Even then, we had one of the most sophisticated text filters running on BERT models in 2021, as well as our own recommendation model. Recommendations are the OG of machine learning.
None of us had thought that sentence completion would become such a big use case. That was the transformative effect of Google’s original Transformer paper, followed, of course, by chat.
For us, the biggest moment has been the ability to do 3D and 4D creation. When we saw Google’s Transformer architecture, we said, “Aha, this can also be applied to graphics.” We’ve been on that journey for the last three years.
India is a major growth market for Roblox. What are you doing to improve the user experience?
Our focus in India is making sure every user, even if they are in a tier-two or tier-three city, gets a great experience. We’ve done a lot of work to make Roblox perform well on intermittent networks and lower-powered phones.
We put a datacentre in Mumbai a few years ago, which immediately improved the user experience by lowering latency. We continue to enhance and expand it.
We are now evaluating GPUs in these datacentres as more workloads shift to AI. We’re looking at how to expand GPU access globally so our users can have a great AI experience. But building GPU capacity is both fascinating and challenging.
Since GPU capacity is constrained, we spend a lot of time distilling models so that we can run them at scale.
Would you look at more data centres in India?
We’re constantly evaluating it. One interesting thing about India is that, for some cities, the ping time to our Singapore data centre is actually better than to other cities in India. From Chennai or Bengaluru, for example, you can get very good connectivity to Singapore, where we already have a very large data centre.
Roblox has embraced generative AI across the platform. What is your broader AI strategy, and how do your recent launches fit into it?
There are four big areas of AI for us.
The first is 4D creation, focused on creating fully functional objects. If you want to create a car, for example, you want it to have a steering wheel, wheels that work, and doors that open.
The second is video. We recently announced Roblox Reality where you can switch on photo-realism with a single knob and upscale any game. One of the hardest and most expensive things in gaming is creating photo-realistic environments, from the leaves on trees, grass, mountains, sun and clouds.
The third is coding. But our twist on coding is different from the coding models you know. We recently announced Build, where you can create a game with a single prompt. We want to make that experience extremely elegant for the user.
The fourth one is NPCs (non-player characters). These are AI characters that you can add into your game to make it more interesting.
Across all four areas, we’re building our own models. Our strategy is a mix of off-the-shelf and open-open- example, is our own foundational AI model that generates 3D models and environments from text
World models are emerging as the next frontier in AI, with models like Google Genie and others. What is Roblox’s play there?
If there’s one word to remember about world models, it’s control.
You’ve seen video models where you give a prompt, get a video, and feel excited the first time you see it. But most of the time, the video will have its own mind and you can’t control it.
For a creator, imagine having a movie in your head, only to have the model do whatever it wants. That’s a classic hallucination problem. You need a game engine to control that video. The engine understands the objects and their physical properties, so if a car hits a tree, it stops.
So our approach is a hybrid one. Pixels are controlled by the video model but the actual object control is from the creator.
That’s why few world models have moved beyond good demos. Turning them into production games or tools is much harder.
One of the biggest challenges in creator platforms is the power-law effect, with a small percentage of creators capturing most of the engagement and revenue. Will generative AI help flatten that curve, or will it give the top creators an even bigger advantage?
My bet is that AI will flatten the curve. When our founder built Roblox 20 years ago, the idea was that anyone should be able to build a game. We would take care of the infrastructure, safety, distribution and monetisation.
Fast forward 20 years later, the promise remains the same, but AI makes it possible to create much more complex and beautiful games. You might have an amazing idea in your head, but previously you needed to hire 20 people to bring it to life. With AI, you should be able to do it with one friend, build the game and iterate on it without worrying about the infrastructure.
Wouldn’t discovery then become the new bottleneck? What will help creators stand out, and how is Roblox helping them reach a larger audience?
Discovery is going to become more complex. But even in our internal tests, we’ve found that making a game truly interesting still requires the human mind. I can build a racing game, and you can build a racing game, but the gameplay will be a differentiator.
We’re also investing heavily in AI for discovery so we can better understand what each player finds interesting and entertaining. We monitor a lot of signals because we don’t want games that capture your attention for two or five minutes and then lose it. We want games that drive long-term retention.
Roblox has faced significant scrutiny over safety in recent years. As AI makes it easier to create more content, how are you approaching the challenge of moderating it at scale?
We’ve invested heavily in safety, from our early days to today, and we’ve open-
We have models that continuously monitor images and objects inside games in real time. Will moderation become more complex? It will, certainly. We have a Sentinel model for text, our own voice safety model, and real-time moderation models, and we’re continuously enhancing them to keep pace with that complexity.
We have multiple publishing gates and tightly control distribution, so we’re careful about how much traffic and engagement we give to any new game.
For younger audiences, we have a moderated content library that creators have to qualify for before their content is shown to them.
Vikas SN covers Big Tech, streaming, social media and gaming industry
first published: Sep 9, 2026 10:31 am
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
Watch
Trending news
‘Only Rs 5-8k left’: 29-year-old man earning Rs 1.2 lakh/month says lifestyle inflation is stressing him
- ‘Investing about an extra Rs 30,000’: Techie says getting laid off from Meta made him richer
- ‘I just knew discipline’: 29-year-old woman buys her own plot after saving in piggy banks for 9 years
- Piyush Goyal joins viral 80s trend with AI-generated retro avatar: ‘Vibing to the 80s trend’
- Sachin Tendulkar adds Rs 70 crore land to real estate portfolio: List of properties the ‘God of Cricket’ owns
- Artificial Trailer OUT: Andrew Garfield plays OpenAI founder Sam Altman; explores dramatic ouster and return
- Hanuman Ansh director Vishal Chaturvedi, actor Shobhinav and team meet Uttar Pradesh CM Yogi Adityanath
- Father cobbler, mother labourer but son built Rs 350-crore company: How Ashok Khade went from poverty to Padma Shri
- She started her business while pregnant at 26, her beauty brand made Rs 3 crore in its first year
- Apple iPhone 17 Pro price drops to Rs 76,000 with exchange and last chance before Cosmic Orange gets discontinued
- Apple iPhone 18 Pro Max launch set for tonight: Here’s everything from price in India, updated processor, camera and design we know so far
- After high cholesterol and fatty liver diagnosis, Om Swami added tomato soup to his diet; see recipe
- Want to check how healthy you are? Take this 13-point self-assessment test
- Meet Luke Carpenter-Bundhoo, the ecologist who found a mysterious ‘vegan Dracula’ fish 900 miles too far north
- Quote of the day by Taylor Swift: “To me, Fearless is not the absense of fear. It’s…”—a powerful lesson on facing your fears
`);
}
if(res.stay_updated){
$(“.stay-updated-ajax”).html(res.stay_updated);
}
}catch(e){ console.log(e); }
});
},2000); // Reduced from 8s to 2s for better Core Web Vitals (LCP/UX)
});
