After diving into the merged findings from Bessemer’s “State of AI 2025” and Janelle Tang’s Next Big Teng’s analysis, I’m struck by how the AI startup landscape is crystallizing into a fascinating dichotomy. Three years post-ChatGPT, we’re seeing the market mature in ways that tell us a lot about where things are headed.

What’s immediately apparent is that we have two very different species evolving in the AI ecosystem:

Supernovas are the headline-grabbers – these companies hit $40M ARR in their first year and rocket to $125M by year two. They’re the OpenAIs and Anthropics of the world, prioritizing hypergrowth over everything else. But here’s the catch: they’re running on thin 25% gross margins (sometimes negative), which makes me seriously question their long-term sustainability. They’re sprinting at Olympic pace, but can they maintain it without collapsing?

Shooting Stars, on the other hand, follow a more traditional SaaS trajectory – they take about 4 years to hit $100M ARR, but do so with healthy 60%+ gross margins. These companies (think Jasper or Notion AI) are building more deliberately, focusing on sustainable economics rather than blitzscaling at all costs.

Taking a step back, I think this bifurcation reveals something important about the AI market: there isn’t just one path to success. The “grow at all costs” mentality that dominated the early days is giving way to a more nuanced understanding of how AI businesses can thrive.

What I find particularly compelling is Bessemer’s prediction that browsers will become the AI operating system. We’re already seeing early signals with tools like Comet and Dia, and it’s only a matter of time before OpenAI and Google jump in, potentially triggering a new “browser war.” This makes intuitive sense – browsers already serve as our gateway to most digital experiences, so having AI agents that can work across tabs and applications feels like a natural evolution.

The timeline for generative media is also fascinating – images in 2024, voice in 2025, and video breaking out in 2026. Companies like Veo 3, Sora, Kling, and Marey are positioning themselves at the forefront of this wave. I’m particularly interested in how this will transform content creation across marketing, education, and entertainment.

What I don’t believe is that all these predictions will unfold exactly as forecasted. The suggestion that a new AI-native social giant could emerge (drawing parallels to how PHP enabled Facebook and mobile video enabled TikTok) feels speculative at best. Yes, voice, memory, and generative media could fuel something new, but predicting the next social platform has historically been a fool’s errand.

Let’s just think about the implications for founders. The report suggests several strategic directions:

1. Building persistent memory and context into AI systems creates defensible moats
2. Creating systems that execute actions (not just record data) delivers more value
3. Starting narrow and expanding works better than trying to boil the ocean
4. Speed of implementation is becoming critical as adoption cycles collapse
5. Vertical AI solutions are penetrating even traditionally resistant industries

The M&A angle is particularly noteworthy. As incumbents look to close their AI capability gaps, startups with strong traction become prime acquisition targets. This creates an interesting dynamic where founders need to balance building for independence while remaining “M&A-ready.”

The report’s dichotomy between Supernovas and Shooting Stars is useful but potentially limiting. I suspect many successful AI companies will chart hybrid paths that borrow elements from both archetypes. The real world rarely fits neatly into binary categories.

And the thin gross margins of Supernovas raise serious red flags. While venture capital can subsidize growth for a while, eventually the economics need to work. I’m skeptical that companies can sustain hypergrowth without figuring out how to improve those 25% margins substantially.

What’s missing from this analysis is deeper exploration of regulatory implications. As private evaluations and data lineage become critical for enterprise adoption, how will this intersect with emerging AI regulations? The report touches on this but doesn’t fully connect the dots between compliance requirements and business strategy.

Overall, this merged report provides valuable benchmarks that help separate hype from reality in the AI market. The recognition of multiple paths to success feels particularly grounded, and the directional insights about browsers becoming the dominant AI interface align with observable trends.

For anyone building in this space, the key takeaway is that taste and intuition still matter enormously. In a world racing toward automation, human judgment remains the differentiator. That’s something worth remembering as we navigate the evolving AI landscape.¹

¹ Though the report suggests tracking specific companies like Comet, Dia, Veo 3, Sora, Kling, and Marey, I’d recommend expanding this watchlist to include emerging players in the private evaluation and data lineage space, which could become increasingly important as regulatory scrutiny intensifies.