Why did Anthropic's models gain popularity over OpenAI's?

Anthropic's models have shown superior performance in coding tasks and agent development, which are significant problem spaces for startups. This practical success, combined with a familiar 'personality' for users, has driven their adoption.

How does Gemini compare to ChatGPT and Claude for everyday use?

Gemini is praised for its reasoning capabilities and effective use of Google's index for real-time information, making it a strong alternative for replacing traditional search queries. ChatGPT is noted for its sticky memory, making it feel more personalized for ongoing conversations and tasks.

Are startups building their own AI models or using existing ones?

Startups are increasingly abstracting away model choices, building orchestration layers to swap models dynamically. Some are also focusing on smaller, domain-specific models, fine-tuning open-source alternatives to compete with larger players like OpenAI in specific niches.

Is the current AI boom a bubble like the telecom bubble of the '90s?

While there's heavy investment, the AI boom is seen by some as more analogous to the telecom era's infrastructure build-out. This period of 'installation' creates opportunities (like YouTube from telecom glut) for application-layer startups during the subsequent 'deployment' phase.

What are the biggest challenges facing AI infrastructure development?

Current challenges include a lack of sufficient power generation for massive data centers, limited land availability due to regulations, and supply chain constraints for essential components like jet engines for power units. This is driving interest in unconventional solutions like space-based data centers.

Will AI lead to fewer jobs or just change the nature of work?

While AI increases efficiency, it also raises customer expectations, leading companies to continue hiring to meet demand. The trend suggests AI is augmenting rather than simply replacing human roles, with execution and people remaining key bottlenecks.

Is the era of the one-person, trillion-dollar company here?

Not yet, but companies are achieving significant revenue with remarkably lean teams. For example, Gamma reached $100 million ARR with only 50 employees, indicating a trend towards high revenue generation with fewer people, though fully autonomous trillion-dollar companies are still some way off.

The Truth About The AI Bubble

Y Combinator

Science & Technology3 min read31 min video

Dec 22, 2025|100,342 views|1,696|109

YC Y Combinator

Save to Pod

Key Moments

TL;DR

AI economy stabilizes; Anthropic leads YC, application layer shines, and infrastructure build-out fuels future opportunities.

Key Insights

AI economy has stabilized with clear layers for models, applications, and infrastructure.

Anthropic has overtaken OpenAI as the preferred LLM at Y Combinator.

The real opportunity in AI is shifting back to the application layer.

Infrastructure build-out, like data centers and power generation, is creating opportunities.

The AI 'bubble' is seen as a period of necessary infrastructure investment, not a sign of market collapse.

Idea generation for AI startups is returning to normal levels of difficulty.

STABILIZATION OF THE AI ECONOMY

The AI economy has notably stabilized in 2025, moving beyond its chaotic early stages. A clearer structure has emerged, with distinct layers for model providers, application developers, and infrastructure builders. This stability suggests a more predictable path for AI-native companies, with established playbooks for developing products on top of advanced models. The era of constant, disruptive announcements that completely redefine startup opportunities appears to be waning, allowing for a return to more normalized levels of idea generation difficulty.

SHIFTING MODEL PREFERENCES AND COMPETITION

A significant development in 2025 is the changing landscape of preferred AI models among startups applying to Y Combinator. While OpenAI dominated previously, Anthropic has emerged as the top choice, with models like Claude preferred by a majority. Gemini is also climbing, showing increased adoption and positive user experiences. These shifts highlight a dynamic market where model performance, particularly in areas like coding and reasoning, directly influences developer choices and application development.

THE RESURGENCE OF THE APPLICATION LAYER

While foundational models and infrastructure development have been crucial, the true long-term opportunity in AI is increasingly seen as residing in the application layer. As models become more commoditized and interchangeable, startups are focusing on building innovative applications that leverage AI's capabilities. This strategic shift means that while model companies continue to advance, the next wave of significant value creation and startup success is expected to come from companies that effectively translate AI potential into user-facing products and services.

INFRASTRUCTURE BUILD-OUT AND INNOVATION

The demand for computing power has driven massive investment in AI infrastructure, including data centers and energy solutions. This 'installation phase' of technological revolution, characterized by heavy capital expenditure, can appear like a bubble but is essential for future deployment. Innovations are pushing boundaries, from space-based data centers to advanced fusion energy projects, addressing critical constraints like power generation and land availability, thus fueling further AI development and application.

THE 'AI BUBBLE' AS ACCELERATED INVESTMENT

Concerns about an 'AI bubble' are reframed by historical parallels, such as the telecom boom. This period of intense investment, particularly in infrastructure like GPUs and data centers, is viewed as a necessary accelerator, similar to how the telecom glut enabled services like YouTube. The abundance of compute resources, driven by competition among infrastructure providers and model labs, ultimately lowers costs and creates fertile ground for application-layer startups. This environment benefits new ventures by providing ample resources and reducing barriers to entry.

EMERGING TRENDS IN STARTUP DEVELOPMENT

The ecosystem is seeing a rise in specialized AI models, often fine-tuned from open-source foundations for specific tasks or devices. This trend democratizes AI development, allowing smaller, more focused models to compete even with larger, general-purpose ones in niche domains. Furthermore, while companies are achieving significant revenue with smaller teams, the expectation for AI-driven efficiency is balanced by rising customer expectations, which still necessitate hiring individuals with strong execution capabilities, suggesting that lean, hyper-efficient teams are the future.

THE FUTURE OF SMALLER, HIGH-REVENUE COMPANIES

A notable trend is the increasing ability of startups to reach substantial revenue milestones, such as $100 million in ARR, with significantly smaller employee counts. This suggests a shift towards hyper-efficient operations, where AI enables greater output per employee. While the era of a single person running a trillion-dollar company may not be imminent, the trajectory points towards incredibly lean, high-revenue businesses that leverage AI to maximize productivity and minimize overhead, representing a fundamental change in startup scaling.

Mentioned in This Episode

●Products

●Software & Apps

●Companies

●Organizations

●Studies Cited

●Concepts

●People Referenced

Common Questions

In the winter 2026 selection cycle for Y Combinator, Anthropic surpassed OpenAI as the most frequently chosen LLM API. Anthropic's models are performing particularly well in coding-related tasks, leading many founders to select them for their applications.

Topics

AI Economy LLM Adoption ChatGPT Comparison AI Startups Tech Revolutions Domain-specific AI AI Funding

Mentioned in this video

companyZephr Fusion

A space fusion company that found a viable path to gigawatt energy production by operating in space.

personCarlo Perez

An economist who wrote about tech trends and technology revolutions, categorizing them into 'installation' and 'deployment' phases.

organizationEmergence

A company mentioned as a winner in the 'vibe coding' category.

companyLora

A company representing the second wave of AI-native companies, competing with established players like Harvey, and also mentioned in relation to investor benefits from fine-tuning costs.

companyStarCloud

A company that proposed building data centers in space, initially met with skepticism but later followed by major tech companies.

companyBoom

A company working on energy solutions, mentioned in the context of solving the power generation problem for AI data centers.

studyMIT report

An MIT report that claimed a high percentage of enterprise AI projects fail, which the speakers discuss in the context of slow AI adoption due to organizational resistance.

companyGiga

A company mentioned as part of the second wave of AI-native companies, competing in the market.

companyGamma

Mentioned for achieving $100 million in ARR with only 50 employees, showcasing a trend of high revenue with lean teams.

companyHelion

A company focused on fusion energy, mentioned as addressing the power needs for AI data centers.

conceptARC AGI prize

A prize that will be offered in the winter 2026 batch as a nonprofit, potentially accelerating AI development.