
Korea's Soombit AI Raises About ₩12B as Its Generative-AI Radiology Reports Head to RSNA
South Korean medical-imaging AI startup Soombit AI (숨빗AI) has raised a Series A round of approximately ₩12 billion led by SV Investment, according to Rapportian's report of the company's announcement on 17 November 2026. Altos Ventures, Kiwoom Investment and iM Investment Partners participated. The timing is deliberate: within two weeks the company will be at RSNA 2026 in Chicago — the world's largest radiology meeting — preparing to show a global audience what comes next after chest X-rays. For health-AI founders in Singapore, China and Southeast Asia, the round deserves a closer read.
Pitching a technical product to investors or buyers? Explore Jason Lim's Future of Selling presentation skills program to sharpen how you tell your story — because in deep tech, the evidence is only half the pitch.
What Happened
Soombit AI was founded in July 2024 by the chest X-ray AI team from KakaoBrain. Co-founder and CEO Woong Bae (배웅) previously served as KakaoBrain's Chief Healthcare Officer and, before that, as managing R&D director at VUNO; he holds an M.S. in Bio and Brain Engineering from KAIST. The company and its leadership are listed on the Soombit AI team page.
The new Series A takes cumulative funding to roughly ₩20 billion, including a ₩5 billion seed round from Altos Ventures in 2024, according to Rapportian. The company has also been selected for Korea's Scale-up TIPS special program, which provides up to ₩3 billion in government-matched R&D funding over three years for what is described as "multi-sequence CT·MRI generative AI solution development and global commercialization."
The flagship product, AIRead-CXR, received Class 3 digital medical device approval from Korea's Ministry of Food and Drug Safety (MFDS) on 1 April 2026 — described in Edaily's coverage as Korea's first approved generative-AI medical device. Per Rapportian, the system has been trained on approximately 14 million chest X-rays and reports since 2021, detects more than 57 findings, and drafts reports in one to two seconds while working within existing PACS workflows without changes. The company cites six peer-reviewed papers: three in Radiology, two in the American Journal of Roentgenology and one in European Radiology.
Next up is AIRead-CT, covering brain, chest and abdomen-pelvis CT draft reports, targeted for release in 2027 and set to be unveiled at RSNA 2026 in Chicago, 29 November to 2 December (the company's newsroom lists Booth 4060, South Hall Level 3).
"Based on the draft-report generation technology and know-how proven in chest X-ray, we will quickly expand to CT and MRI and become indispensable infrastructure for radiologists worldwide," CEO Woong Bae said, according to Rapportian (translated from Korean).
Why This Matters for Asia's Health-AI Startups
This section is analysis, not reporting on Soombit. The most instructive part of this story is not the funding amount — roughly ₩12 billion is a modest Series A by global standards — but the sequencing. Soombit secured a national regulatory approval and peer-reviewed clinical evidence before its major international push. That approval functions as a credibility wedge: when a Korean generative-AI medical device is cleared by the country's drug regulator, conversations with hospital buyers, distributors and investors in other markets start from proof rather than promise.
That is a playbook founders in Singapore, China and Southeast Asia can borrow. Comparable regulatory pathways for AI-enabled medical devices exist in other jurisdictions, and a first-market approval — wherever it comes from — can anchor a global narrative. The lesson generalizes beyond healthcare: in deep tech, the approval, the clinical evidence or the benchmark result is the story the founder must tell. Capital follows demonstrated, verifiable progress, not the reverse.
What Founders Can Take Into the Room
Three practical lessons for technical founders preparing investor or conference presentations:
- Lead with the evidence, not the architecture. Soombit's story is built on an approval date, a paper count and a dataset size — specifics an investor can verify. Swap in your equivalents.
- Translate speed into workflow value. "Draft reports in one to two seconds" matters because it is framed against a radiologist's existing workflow, not as a model benchmark.
- Time announcements to the room. Unveiling AIRead-CT at RSNA puts the product in front of the exact audience that can adopt it. Match your launch moment to where your buyers already gather.
Related News
- He Built a Startup at BAJA SAEINDIA. Twenty Years Later, He's the Keynote Speaker.
- 15 Sales Presentation Mistakes That Are Costing You Deals
- AI Can Write Your Speech. It Cannot Hold the Room for You.
How Can Asia Startups Compete on the Global Stage?
A Preply survey of 2,007 UK adults found that 21% say they hate or avoid public speaking (the survey covers the UK, not Asia, but the gap it points to is universal). For founders from Singapore, China and the wider region competing against well-funded Western peers, communication is one of the few advantages that costs nothing to build. Three ways to work on it:
- Rehearse the first ninety seconds. Investors and conference audiences decide quickly whether to lean in; your opening should carry the proof point, not the pleasantries.
- Build a proof deck. Approvals, pilot results, publication counts and customer workflow gains are the currency of deep-tech pitches. Put them early and cite them precisely.
- Practice out loud, in the room. Reading slides silently hides every rough transition. Speaking the pitch aloud — ideally to someone outside your field — surfaces them.
If you want a structured way to sharpen these skills, explore Jason Lim's Future of Selling presentation skills program — a practical way to strengthen how you pitch before your next round or conference stage.
Ad: DreamakerAI SalesOS is an AI CRM and business operating system built for founders.
Sources: Rapportian (reporter 정희석, 17 November 2026); Soombit AI newsroom; Edaily.
