Why Most AI Startups Are Doomed & How to Beat the Odds
Insights from a provocative SXSW 2025 panel on the future of AI ventures
Key Takeaways
- Value Creation Gap: Most AI startups are struggling to demonstrate sustainable value creation, with consumers unwilling to pay what it costs developers to build AI products
- Healthcare AI Challenges: The healthcare sector is particularly vulnerable to “AI tourists” who create quick API wrappers without solving fundamental problems
- Valuation Bubble: Current AI company valuations may require the AI revolution to be five times more productive than the internet to justify their worth
- Domain Expertise Critical: Successful AI startups combine deep domain expertise with technological innovation to solve specific pain points
- Regulatory Realities: Many sectors have entrenched regulatory frameworks and professional organizations that present substantial barriers to AI disruption
- Human-AI Collaboration: The most promising path forward is AI augmenting human capabilities rather than replacing humans entirely
At SXSW 2025, a provocatively titled session “Most AI Startups Are Doomed” brought together a diverse panel of experts from healthcare, robotics, venture capital, and deep tech to debate the sustainability of the current AI boom.
- James Wang — General Partner at Creative Ventures
- Chidi Nwankpa — Principal at Gore Range Capital and Co-Founder of GlobalHealth AI
- Sergei Polevikov — Co-founder of WellAI and an AI Advisor
- Kartik Tiwari — Co-founder and CTO of Andromeda Surgical
While the title suggested pessimism, the discussion revealed nuanced perspectives on the challenges and opportunities in the AI landscape:
- The State of AI Entrepreneurship
- The Healthcare AI Wasteland
- The Valuation Bubble
- The Commoditization Problem
- The Path to AI Startup Success
- Conclusion: Doomed or Determined?
The State of AI Entrepreneurship
From the outset, the panel acknowledged the statistical reality that most startups fail, but focused on whether AI startups face unique challenges that make them particularly vulnerable in the current environment.
Kartik Tiwari framed the fundamental problem succinctly: “There hasn’t been a really good sign, or even a really good objective proof of a good, sustainable product that a consumer is willing to spend the amount of money that developers are actually spending to build that product.”
The Healthcare AI Wasteland
Healthcare emerged as a particularly problematic sector for AI implementation. Sergei Polevikov, who writes a newsletter investigating healthcare AI companies, expressed concerns about what he calls “AI tourists” — companies that create quick wrappers around existing large language models without addressing fundamental healthcare problems.
“For digital health… we’ve had a lot of problems, especially with AI ChatGPT healthcare experience, a lot of what I call AI tourists,” Polevikov explained. “Companies who make a quick wrapper API connection to ChatGPT, for example, do it very beautifully, sell to certain group of people, but in my view, that’s not sustainable.”
He went further, claiming that of approximately 126 AI scribe applications in healthcare, “about 122 of them are rip-offs, basically just putting some wrapper on somebody else, and only four, in my opinion, are true innovators.”
The Valuation Bubble
The panelists expressed concern about inflated valuations reminiscent of the dot-com bubble. James Wang noted that for current AI company valuations to make sense, “the current AI boom needs to be five times as productive as the internet was by the end of it.”
Chidi Nwankpa described the current market correction: “Yeah, it’s been brutal. We’ve had multiple companies reaching out to us. They’re all going for a down round. There’s beens a bloated valuation, and now just people are just coming down.’”
Sergei Polevikov drew explicit parallels to the dot-com era: “To me, it’s exactly like end of the 90s. Back then it was www everything — everybody was buying. You have huge premiums. Right now, you get 40% premium on valuation just for mentioning AI.”
The Commoditization Problem
The panel debated Sam Altman’s recent comment that the “middle layer” of AI — the large language models themselves — will likely be commoditized, with value moving either to applications or hardware.
James Wang outlined why LLMs are becoming commoditized: “They have the same models, roughly speaking… They have the same data. Everyone scraped the same data from the internet… and they have the same compute, meaning that they all can buy NVIDIA GPUs.”
This commoditization was demonstrated by xAI’s rapid progress: “xAI just went out, spent some money and basically ended up with a frontier model,” Wang noted, suggesting that proprietary data may be the only reliable moat for AI companies.
The Path to AI Startup Success
Despite the pessimistic title, the panel offered concrete advice for AI startups looking to beat the odds:
1. Solve Specific Pain Points
Chidi Nwankpa emphasized focusing on underserved needs: “Can you find a pain point that a lot of people are complaining about, but most people don’t want to go in? Is there a barrier to entry? Is it something that people are not just excited about?”
2. Combine Domain and Technical Expertise
“The way to make your startup relevant is to make sure that you’re combining deep domain expertise with the technology,” Nwankpa explained. “Once you can combine both of them, you have someone that understands the space, knows where the main pain points are… you also have someone that understands the technology and how to merge the two.”
3. Focus on Profitability and Cost Discipline
Sergei Polevikov’s research found that successful companies take a more disciplined approach: “If you as a nimble, hard working startup that’s focused on profitability and costs, if you can survive that first short term, long term, you’re becoming more successful than all these big names, noisy companies that all these famous VCs are promoting.”
4. Take an Evolutionary, Not Revolutionary Approach
Kartik Tiwari emphasized that transformation happens incrementally: “Any revolution is evolutionary. It is applicable to societies. It is also applicable to products. No product just came in and just revolutionized everything… It already had a lot of baselines that brought up the customers, brought up their supply chain ecosystem.”
5. Keep Humans in the Loop
A consistent theme was the importance of human-AI collaboration rather than full automation. As one audience member pointed out and the panelists agreed, attempting to automate an entire process at once is often counterproductive.
“Shooting for 100% automation actually takes you a few steps back instead of making you move forward,” noted Tiwari.
Sergei Polevikov concurred: “AI plus humans are the way to go, especially in healthcare.”
Conclusion: Doomed or Determined?
While acknowledging the challenges facing AI startups, the panel concluded on an optimistic note. “AI startups are not doomed,” Chidi Nwankpa stated. “I think there’s ways to create a startup that will be successful. You just have to be more creative about it and not do what every other person is doing.”
For additional insights, listen to the full panel discussion:
StartingUpGood supports fresh entrepreneurial approaches to social impact. FOLLOW US on social media:
Medium | LinkedIn | X | Facebook | Instagram | Threads | Bluesky Social
Check out SDGCounting for the latest news on tracking the progress of the Sustainable Development Goals. #SDGs #GlobalGoals.
Note: Generative AI tools were used in the creation of this article to assist with translation, summarization, and editing.
