Yesterday, we examined how the gravitational pull of the tech giants’ platforms is reshaping the AI playing field, creating both competitive pressures and new growth channels for companies that bring unique, complementary value.

Today, we take a wider view. Signs are emerging of a more selective, rational market. Early opportunities in lightweight, API-based applications are evolving as platforms move up the stack, adding more application-layer capabilities directly into their core products. At the same time, investors are focusing larger bets on companies that can demonstrate strong defensibility and staying power.

What we may be seeing is the market’s filtration process at work, raising the bar for success while also creating clearer paths for disciplined builders to stand out and thrive.

Part III of our series: How the market is shifting from hype to substance, and why that could make this a compelling moment to be building in AI.

The initial, frenetic phase of the AI gold rush is coming to an end. The Cambrian explosion of apps built on third-party APIs was a thrilling moment of innovation, but that era of fragmented, easy wins is closing. A new, and ultimately more rewarding chapter has begun. We are now in the midst of “A Filtration”—a necessary and healthy market rationalization. This process is separating features without a clear path to defensibility from durable companies, clearing out the noise, and creating a clearer path for those building with discipline and foresight.

For the strategic founder, this filtration isn't a threat. It's the ultimate advantage.

A simple playbook has dominated: wrap a thin user interface around a third-party API and ship it. This led to an oversaturated market where thousands of startups were built on the same commoditized foundation, all competing for the same niches.

That playbook is becoming increasingly difficult to sustain. As major platforms bundle similar AI capabilities into their core software for free, the strategic vulnerability of a simple "wrapper" is being exposed. The market is no longer rewarding hype; it's rewarding defensibility. The core issue lies in the lack of sustainable business models. Many AI tools’ function as features rather than standalone companies, making them challenging to monetize at meaningful scale. Without a deep, defensible moat, these products can be replicated and bundled for free by tech giants’ platforms, making it harder to establish a recurring revenue model. While some wrappers can evolve into defensible businesses, often face headwinds without proprietary data, workflow integration, or unique moats.

A quiet but important shift is taking shape in the AI race. While much of the discussion still centers on larger, more capable models, leading labs are increasingly investing in systems that improve by learning from live, real-world usage, not just static, historical datasets.

Over the past year, companies like OpenAI, Anthropic, and Google DeepMind have been building high-fidelity simulations of widely used digital environments, from major websites to enterprise systems of record. These virtual testbeds allow AI agents to practice complex, multi-step workflows and learn from delayed success signals. This could mean closing an enterprise sale, completing a lengthy debugging session, or running a multi-day campaign, scenarios where the outcome only becomes clear well after the first step.

This approach, combining reinforcement learning with human and AI feedback (RLHF and RLAIF), is beginning to influence where competitive advantage sits:

Real-Time Data Is Becoming a Key Differentiator

Defensible training signals are increasingly coming from live interaction loops rather than static archives. Companies that capture a steady flow of user decisions, workflow completions, and edge-case resolutions can create feedback systems that refine performance in ways general-purpose models may struggle to match. OpenAI’s integrations with Microsoft Copilot, ChatGPT, and its API network give it broad visibility into usage patterns, insight it can apply to product improvements and vertical-specific features.

The Line Between Infrastructure and Applications Is Blurring

Recent commercial results point to this shift: Reportedly OpenAI doubled ARR from $6B to $12B in six months, while Anthropic grew from $1B to $5B over a similar period. Notably, $1.4B of Anthropic’s API revenue comes from just two customers — GitHub Copilot and Cursor, both application-layer products that could face competitive pressure if the model provider moves further up the stack.

Rapid Model Preference Changes

In developer tools, for example, Claude 3.5 Sonnet has quickly become the default in certain contexts, overtaking GPT-4 for specific use cases. Anthropic’s expansion into end-user products, such as Claude Code, reportedly generated $400M in ARR within weeks. This shows how quickly a platform provider can identify high-usage patterns in its API and integrate similar functionality into its own offerings.

For companies building on top of these platforms, the strategic challenge is how to maintain an advantage when the infrastructure provider has both visibility into product traction and the ability to act on it.

The defensibility playbook is evolving. It now includes creating proprietary reward functions, developing domain-specific evaluators, and designing reinforcement learning systems tuned to the nuanced workflows in your target market. In the current market environment, the companies that endure will be those whose moats remain strong even if the platform itself steps into the same territory.

This technological shift is mirrored in how capital is now being allocated. Investors are concentrating their resources on companies that not only have defensible technology, but also the ability to capture and sustain these real-time data advantages.

The venture landscape is reflecting this shift. While headline investment figures remain high, capital is concentrating in a handful of mega-deals. In the second quarter of 2025, the top 10 venture deals accounted for over 45% of all capital invested. This isn't a funding drought. It's capital precision. The story is in the widening gap between the value of AI deals and the count of AI deals. In the first half of 2025, AI's share of total deal value soared to 53%, while its share of deal count was a much lower 29.2%. This concentration of capital is part of the filtration at work, investors are doubling down on companies that can clear the market’s defensibility bar.

This chart doesn't show a lack of funding; it shows a flight to quality. Investors are no longer spraying capital across a wide portfolio of undifferentiated apps. Instead, they are making disciplined, concentrated bets on companies that demonstrate true, defensible moats. For founders who can prove they are building a durable platform rather than a temporary feature, the capital is more focused and more committed than ever.

But for companies that are still working toward this defensibility bar, alternative outcomes are becoming more common, including strategic acquisitions and “acqui-hires” that allow teams to continue their work inside larger platforms.

With the IPO market largely frozen, mergers and acquisitions (M&A) have become the primary exit path for startups. Analysis of this trend reveals how the market is evolving. In 2024, nearly 90% of exits involved companies that had raised no further than a Series B round. These should not be seen as failures; they are strategic acquisitions and talent-first exits.

In a consolidating market, being acquired by a larger platform while not the first outcome most teams envision, “acqui-hires” have become a pragmatic reality. This has given rise to the "acqui-hire", a talent-first acquisition where a major firm hires a startup's core team, prizing their expertise above a product. Recent headlines are filled with high-profile examples of this. Perhaps the most notable are:

1. Google – Windsurf, Google reportedly entered a $2.4 billion licensing and hiring deal with AI coding startup Windsurf. This involved hiring the CEO Varun Mohan, co-founder Douglas Chen, and parts of the R&D team into DeepMind to work on agentic coding projects, while licensing Windsurf’s technology reportedly under non‑exclusive terms.

2.  Meta – Scale AI, Meta agreed to invest $14–15 billion investment in Scale AI, taking a ~49% stake. This deal lured Scale’s founder, Alexandr Wang, and key team members into Meta’s newly formed AI group.

3.  Microsoft's – Inflection AI, where it hired the startup's co-founders and the majority of its technical team in an unconventional transaction valued at around $650 million.

4.  Shopify – Shopify executed several strategic “tuck-in” acqui-hires/team hires, onboarding entire startup teams to accelerate its AI and product innovation. The acquisitions included Peel Insights (e-commerce analytics), Ritual (mobile ordering), Threads (workplace communications), ChannelApe (inventory operations), Stellate (developer tooling), and Checkout Blocks (checkout customization).

While not the celebrated unicorn exit, it's a testament to the quality of the team, often a startup's most valuable asset. In some cases, these can provide a productive path forward where elite talent is scarce. Our mandate, however, is to help founders build beyond this. While strategic acquisitions can be valuable outcomes, the goal is to create a company so highly valued, so deeply embedded in a customer's workflow, that an acquirer desires the entire business model, the team, the technology, and the irreplaceable market position.

The Filtration is creating a landscape where substance trumps hype. It is clearing the field for serious founders, focusing capital on truly defensible ideas, and creating pragmatic exit pathways. This rational environment is not something to be feared, it's something to be prepared for. The critical question for founders is no longer just "What can you build?" but "What can you build that lasts?"

Tomorrow: Now that we understand this new, rational landscape, how do you build a company that thrives in it? In our next piece, we unveil the playbook for building the durable, defensible moats that this new market demands. As we saw in Part II, the gravitational pull of the tech giants’ platforms is strong — Part IV, we show how to build the moat to withstand it.

Click here to download the article from our website 🔗