We Are In The Sandbox Era For GenAI. So Invest Accordingly.

_{Photo creds: Sand Timer in Morning Sunlight on Wooden Table, by Fernando Capetillo}

Over the past year of reporting on AI, I’ve watched astronomic valuations get slapped on startups building foundational infrastructure—chipsets, core models, and research labs. But here’s the thing: we’re in a sandbox era, where everyone’s experimenting, and the real winners won’t be the ones selling the newest iteration of the shovel. They’ll be the ones digging up value that people actually use.

The Bubble Is Real (But It’s a Filter, Not a Crash)

Getting back to those valuations, I’m opening by saying that an AI bubble is a thing that is happening. Though it’s not an apocalyptic one, I’d argue that it’s a market-level filter bringing us back closer to our fundamentals. Some have even gone on the record to say it - re: the CEO of an AI startup has the confidence to admit it [1], so hats off to you!

Let’s start with the elephant in the room: There are loud whispers that we are in an AI bubble. I think Ali Ghodsi is only partially right. Yes, he’s right in the way that pouring large amounts of investment capital into research labs (Perplexity, OpenAI, Mistral) that are monetizing on a subscription-basis is sending your fund’s money to work on a treadmill of unsustainable costs. I’d argue that it’s not the wisest best to place when considering them through the lens of the traditional Venture Capital lifecycle. 

Harking back to some time ago, generic e-commerce platforms died, but niche players like eBay and Amazon thrived by solving real problems. And I’d position that the argument stands firmly into today’s “Industry 4.0 rise”/”AI-dustrial revolution”. Considering that over $12B of 2024’s AI funding went to “fundamental infra” [2], I’d position that simple tools that non-developers actually pay for will come out on top. These large-resource heavy AI models sure do have their applications, but for many of the uses we are currently seeing in AI Products (the application-level stuff that businesses and users are tapping into regularly), these types of “rent-a-model”  subscriptions are bound to be abandoned by a large portion of your startup/business users. 

Vertical Integration Is Eating the World

Here’s the secret: no one wants to outsource their brain. Startups and enterprises alike are racing to in-house their AI stacks. Why? Control. Reliability. Cost. OpenAI’s outages and eye-watering bills are a wake-up call: if your product depends entirely on a third-party LLM/GenAI Model, you’re playing Jenga with your valuation.

But let’s be real—most companies won’t build their own chips. What they will do is fine-tune open-source models (Llama, Mistral, DeepSeek) on proprietary data to solve niche problems. Think healthcare diagnostics, legal contract review, or logistics optimization. These aren’t “AI companies”—they’re vertical SaaS tools with AI under the hood. And they’re thriving because they own their workflows.

Much of the secret sauce for AI stands on a foundation of various Open-Source technologies  that are free to use thanks to licensing that’s essentially attribution-free and truly meant to be customized and commercialized. Historically, this idea of privatizing models is a very “new” thing, and steps outside of common practice. Furthermore, the entrance of players like DeepSeek in the space (minus some of the model’s political opinions) marks a direct threat to the privatized research lab’s perceived supremacy. Big money + Broad vision =/= Best Results. Read more on Open Source benefits & tradeoffs [3]

But from what I’ve lived, seen and heard, almost everyone does end up doing some amount of internal R&D on their custom models. As the technology behind individual AI products matures, its customization needs become increasingly niche, in so many cases, that these larger public models are likely to get dropped in exchange for vertical integration - in-housing the core tech becomes a priority for serious founders. Startups paying $20K/month for API access to platforms like OpenAI will eventually ask: “Why not fine-tune an open-source model for major cost reductions?“

My point here isn’t to sway entirely away from core tech, it’s rather to flag that investors might enjoy the longer-term benefits of a skew in their positions closer to product companies rather than core technology (which people eventually vertically integrate/in-house).

Here’s a bit of data to substantiate - there’s real tangible reason to prove that over 12 billion out of 97 billion of AI investments in 2024 were on “fundamental infra startups” rather than on product companies selling to non-developers. [4]

Humble Sidenote: I can’t speak this much on LLMs & the past waves of GenAI without expressing a level of gratitude to the contributions that the collective past research efforts have yielded us. Sincerely, thank you.

What’s the Investment takeaway from Vertical inevitability? The Caterpillars Will Grow Wings (If They Evolve)

Let’s address the elephant’s cousin: yes, many AI startups today look like “LLM + SaaS UI” toys. Critics call them amateurish. But guess what? These “MicroSaaS” ventures are the caterpillar stage of AI implementations. They’re cheap to build, fast to market, and perfect for testing demand. Take Tome (AI presentations) or Character.AI (custom chatbots). Their stacks are simple, but they’re gaining market share today while figuring out their moats. Will they get copied? Absolutely. But survival depends on what they do next with their initial burst of momentum: reinvesting revenue into proprietary data, custom models, or workflow integrations that competitors can’t replicate.

REMINDER: These startups, and MicroSaaS are the caterpillar state of AI Products as we know them. They’ve got a short chrysalis stage before becoming full grown creatures of their own breed

Our current sandbox era rewards speed. Startups that cling to outsourced LLMs will commoditize themselves into oblivion. Those that evolve—by verticalizing, owning their tech, and solving messy real-world problems—will graduate to the next phase.

Are these Caterpillar SaaS a way to gain market share while securing a runway to better understand the problem space? Absolutely. It's the founders’ responsibilities from that point onward to look at reinvesting into building differentiation. Remember, the law of the jungle still applies - After more growth, the product (if it survives) will mature to evolve to a later phase.

My impression is currently to stay bullish on product companies. The sandbox era is chaotic, but it’s where the future gets built. Time to invest like it.

Here’s some markers that I’m watching in this market’s startups:

• Bets on vertical workflows, not horizontal infrastructure. Startups with industry-specific datasets (e.g., AI for biotech labs, not another chatbot API).

• Looking at founders building with bricks, not with concrete. I’m looking at tools and products that are designed to help clients operationalize open-source models (fine-tuning platforms, evaluation tools) and bring them into production environments.

• Founders with a relentless focus on capital efficiency. The best AI startups I’ve seen lately are respecting their fundamentals:

  1. Bootstrapping niche tools

  2. Scaling with revenue;

  3. THEN scaling with VC cash.

The future continues to be weird, yes, but it's also brimming with possibilities.

Sam from The AI Product Report

Want to talk about this longer? Need more customized help on the matter? Email me [email protected] don’t be shy, let’s talk.

I know there are a LOT of other AI-Product topics to cover like feedback loops and ethics, so let me know if that’s something you want to see discussed!

Here’s an anonymous channel for you to send me your thoughts if the comments section isn’t for you! 😉