Every headline about Flam’s Flam AI Series B funding this week has led with the same hook: Shah Rukh Khan just wrote a cheque into an Indian AI startup. It’s a good hook. It’s also not the interesting part.
The interesting part is buried three paragraphs deep in most of the coverage — a 26-billion-parameter model called Falcon that Flam claims can respond in 30 milliseconds, folded into a pipeline that reportedly returns a full answer in under two seconds. If that number holds up under real-world load, it’s a bigger story than the celebrity investor. Here’s what actually happened, what to make of the tech, and where the claims need a second look before you repeat them.
What Actually Happened
Flam, an AI interactive-content company headquartered in San Francisco with engineering teams in Bengaluru, closed a $40 million Series B on September 15, led by QED Investors. The round also drew Claypond Capital, Goldman Sachs alum-turned-investor Martin Chavez, Datadog co-founder Olivier Pomel, e-commerce veteran Venky Harinarayan, and actor Shah Rukh Khan, with existing backers RTP Global and Dovetail doubling down.
Founded in 2021 by BITS Pilani alumni Shourya Agarwal, Malhar Patil, and Amit Gaiki, Flam builds AI tools that turn static content — video, product shots, brand experiences — into interactive formats. The company says it has signed more than 100 enterprise customers in six quarters, including Google, and is growing 70–80% quarter-on-quarter. The new capital, per the company, goes toward model R&D, expanding its product suite, and scaling enterprise sales internationally, including a new push into Southeast Africa through a partner in Nigeria.
None of that is in dispute — it’s consistent across YourStory, Business Standard, and TechCabal’s coverage. What’s less examined is the stack underneath it.
Inside the Flam AI Series B Funding Story: The Tech Nobody’s Explaining
Flam says it runs five proprietary AI models across three product lines, and this is where the round gets genuinely interesting for a developer audience.
Flicks, its interactive-video format, swaps a person, product, or scene mid-playback using a patented compression technique, with a claimed 50-millisecond time-to-first-buffer. Airboards, its 3D format, streams high-fidelity content through a browser camera interface — no app install — loading in a claimed 300 milliseconds. Fable generates image-to-video assets as native alpha-channel video, useful for overlay-style ad creative.
The most technically loaded piece is Visual Agents — AI characters designed to hold real-time conversations rather than chat replies. Three models sit behind it: Fantom handles facial expression through identity preservation and motion transfer, Finesse handles multilingual voice, and Falcon, a 26-billion-parameter Mixture-of-Experts (MoE) language model, is the reasoning layer, claiming a 30-millisecond time-to-first-token. Stack all three together and Flam says the full round-trip — hearing a query, thinking, speaking back — lands under two seconds.
For context: a 26B MoE model at that claimed latency would be genuinely fast for a video-call-style agent, where most systems still feel like a phone call with a satellite delay. As Flam founder Shourya Agarwal put it, “the demand was always there. The technology wasn’t.”
How It Stacks Up Against India’s Other AI Bets
India’s AI-model landscape has mostly been about text and voice — Sarvam’s 30B and 105B models, Krutrim’s Indic-language stack, Two Platforms’ enterprise agents. Flam is one of the few Indian-origin teams building proprietary models specifically for real-time visual and conversational interaction, a category dominated globally by well-funded labs, not startups. That’s the actual white space QED’s Nigel Morris was pointing at when he said most “frontier AI content companies” are optimising formats that already exist rather than building something new.
The Reality Check Before You Believe the Hype
To be clear about sourcing: the 30ms, 50ms, and 300ms figures come from Flam’s own funding announcement, not an independent benchmark, third-party audit, or published technical paper. The 70–80% quarterly growth figure and “100+ enterprise customers” claim are also self-reported. None of that makes them false — but treat them as company claims until Flam publishes verifiable latency benchmarks or a customer discloses real usage numbers. The startup’s valuation from this round hasn’t been disclosed in any source we found, so avoid quoting one.
Why This Matters for Indian Developers and Startups
Beyond the funding headline, Flam is a useful data point for India’s AI builder community: a homegrown team competing on real-time multimodal inference — not just chatbots — and getting backed by investors who’ve built infrastructure companies themselves (Chavez ran Goldman’s tech division; Pomel co-founded Datadog). If Falcon’s latency claims hold up in independent testing, it’s a genuine signal that Indian AI teams are shipping serious systems infrastructure, not just wrappers around someone else’s API.
FAQs
How much did Flam raise in its Series B round?
Flam raised $40 million in its Flam AI Series B funding round, led by QED Investors, with participation from Shah Rukh Khan and others.
What does Flam’s Falcon model actually do?
Falcon is a 26-billion-parameter MoE language model that powers Flam’s Visual Agents, claiming a 30-millisecond time-to-first-token response.
Is Flam an Indian startup?
Flam is headquartered in San Francisco with core engineering teams in Bengaluru, founded in 2021 by three BITS Pilani alumni.