The State of AI Report - Recap For Founders

The Gold Rush is On, and You Don't Need to Be a Geologist to Strike It Rich

The biggest winners in this new era won't necessarily be the ones who build the most complex "pickaxes"—the massive, foundational AI models. The truly transformative fortunes will be made by clever entrepreneurs who discover the most valuable ways to use those pickaxes to find gold—solving real, tangible problems for specific groups of people.

To illustrate this journey, let's imagine an entrepreneur named Sarah. She's passionate and knowledgeable about the challenges facing local retail businesses, but the world of AI feels alien to her. She sees the headlines and feels a mix of excitement and anxiety, wondering how this powerful new technology could possibly fit into her vision. Her journey from uncertainty to confident action will serve as a guide through the key findings of the 2025 State of AI reports.

The conversation around AI has fundamentally changed. If 2023 was the technology's "Big Bang," a moment of sudden, explosive potential, then 2025 is the "First Light," where the initial chaos is clearing to reveal the shape of a new universe filled with real companies, proven business models, and clear patterns of success. The question is no longer if a business should adopt AI, but rather the practical how-to of AI execution. This report is that "how-to"—a builder's playbook designed not for the AI researcher, but for the visionary founder. Your deep expertise in your chosen field is not a disadvantage; in the age of AI, it is your greatest competitive advantage.  

Chapter 1: The New Lay of the Land: AI Isn't Coming, It's Here

The first step for any pioneer is to understand the terrain. A decade ago, AI was a niche concept, confined to research labs and the tech industry's farthest corners. Today, it is a fundamental and non-negotiable part of the business and social environment, as foundational as the internet or the smartphone. Ignoring it is not an option, because your customers, your competitors, and the very expectations of the market are being reshaped by it in real time.

The Great Adoption Wave (Your Customers Are Already Here)

Imagine Sarah visiting a local coffee shop, a potential future customer for her business idea. She observes the owner, a busy multi-tasker, using ChatGPT to draft a week's worth of social media posts and then switching to a simple AI-powered app to optimize staff schedules. In that moment, Sarah has a revelation: she isn't trying to sell a futuristic concept to a skeptical audience. Her future customers are already AI users. Their expectations for every piece of software they touch are being fundamentally altered by these daily interactions.

This observation is not an anecdote; it's a reflection of a massive global shift. In the business world, adoption has exploded. In 2024, 78% of organizations reported using AI in at least one business function, a dramatic leap from 55% just the year before. This means the market is not only receptive but is actively seeking AI-powered solutions.  

On the consumer side, the scale is even more staggering. An estimated 1.7 to 1.8 billion people globally have used AI tools, with a core group of 500 to 600 million engaging with them daily. This is no longer a phase of casual experimentation; it represents "habit formation at an unprecedented scale". And this adoption is broad. While Gen Z leads in overall use, it's Millennials who are emerging as the true power users, integrating AI into their daily routines more frequently. Perhaps most surprisingly, parents have become an "unexpected power user" group, with 79% having used AI—compared to 54% of non-parents—to manage the complexities of family life. This widespread, cross-demographic adoption confirms that AI is not a niche product for tech enthusiasts; it's a mainstream utility for everyone.  

From Hype to the Bottom Line

Like many entrepreneurs, Sarah initially viewed AI as a fascinating but potentially expensive gimmick—a "nice to have" rather than a core business driver. What changed her perspective was the hard data showing that AI is no longer a science project but a significant factor on corporate profit and loss (P&L) statements.

AI has become a real, and rapidly growing, budget item. Companies are now allocating a substantial 10-20% of their research and development budgets specifically to AI initiatives, and this figure is projected to grow across businesses of all sizes in 2025. This isn't speculative spending; it's an investment that's delivering tangible returns. A growing body of research confirms that AI demonstrably boosts productivity and, in many cases, helps to narrow skill gaps within the workforce by augmenting employee capabilities. The true value emerges when companies use AI not just as an add-on, but to fundamentally "rewire how companies run".  

The market has responded to this value creation. The leading AI companies are now generating close to $20 billion in annual revenue. More importantly for a startup founder, businesses are proving their willingness to pay. In a stunning display of market maturation, 44% of U.S. businesses now pay for AI tools, an almost nine-fold increase from just 5% in 2023. The average contract size for these tools has reached an impressive $530,000, underscoring the significant value they provide. This data is a clear signal: if you can build an AI solution that solves a real business problem, the market is ready and willing to pay for it.  

The Global Race (Why It Matters to Your Startup)

The complex geopolitics of AI can seem distant from the day-to-day concerns of a startup. However, understanding the global landscape is crucial because it determines the tools, talent, and markets that will be available. For Sarah, a simple analogy helps: imagine two giant, competing hardware superstores opening up. One is the "America-first AI" ecosystem, and the other is a rapidly expanding "Chinese-made" ecosystem. The tools they sell (the AI models) are becoming increasingly similar in quality, but they are built on different supply chains and philosophies. As a founder, the ecosystem you choose to build within will influence your access to partners, investors, and customers.  

The United States currently remains the leader in producing the highest number of top-tier AI models, with U.S.-based institutions releasing 40 notable models in 2024, compared to China's 15. However, this lead in quantity is no longer a guarantee of superiority. Chinese models have closed the performance gap with astonishing speed. On key industry benchmarks, the performance difference shrank from double-digits in 2023 to near-parity in 2024. Furthermore, China has strategically leaned into its domestic "open-weights" ecosystem, overtaking Meta as the global leader in this space and establishing itself as a credible number two in the AI race. This isn't just a two-horse race, either. Ambitious "sovereign AI" initiatives are emerging in regions like the United Arab Emirates, which are building their own national compute backbones, creating new global hubs of AI investment and innovation.  

For a founder, this global competition is a double-edged sword. It creates complexity, but it also fosters intense innovation and provides more options than ever before. The rise of a powerful open-source ecosystem, driven significantly by China, offers a viable alternative to relying solely on a few dominant U.S. companies.

The most critical takeaway from this new landscape, however, is not the high adoption rate itself, but the profound gap between usage and value. A new "AI divide" is emerging, mirroring the digital divide of the early internet era. While a remarkable 78% of companies report using AI, a staggering 95% of them are getting zero measurable return on their investment. They are stuck in endless pilot projects or are using AI only for minor individual productivity boosts, like drafting emails. This chasm between simple adoption and true business transformation is the single greatest opportunity for a new startup.  

Large, incumbent companies are struggling with this transition. The real value of AI comes from fundamentally redesigning business workflows, a process that requires significant change management and organizational agility—things that large corporations often lack. They are trying to bolt AI onto old, inefficient processes. A startup, by contrast, is nimble. It has no legacy systems or entrenched bureaucracy. It can design its entire business around a new, AI-native workflow from day one. This is why AI-native companies are nearly four times more likely to have achieved product-market fit than incumbents who are merely adding AI features to existing products. This "GenAI Divide" is not a market failure; it is a market opening. It is the space on the map where a founder like Sarah can stake a claim and build a transformative company.  

Chapter 2: Your Expedition Gear: Understanding the Tools of the Trade

To navigate the AI frontier successfully, a founder doesn't need to be able to build an engine from scratch, but they do need to know the difference between a motor, a wheel, and a chassis. This chapter demystifies the core AI concepts an entrepreneur must grasp, using simple, non-technical analogies to explain the tools that will power their venture.

Beyond the Chatbot: Meet Your New AI "Interns"

Sarah's "Aha!" moment came when she stopped thinking about AI as a tool for writing and started thinking of it as an autonomous intern. She realized she didn't just want an AI that could draft a marketing email; she wanted an AI that could also access her customer list, identify the right segment to send it to, check her calendar for the optimal send time, schedule the email, monitor for replies, and automatically update her customer relationship management (CRM) system with the outcome. This is the essence of an "agentic workflow."

This is the cutting edge of AI product development, where the real value lies. Nearly 80% of successful AI-native companies are investing heavily in building these agentic workflows—autonomous systems designed to execute multi-step actions on behalf of a user. It's the difference between a calculator and an accountant. A calculator is a tool that requires constant human input for every step. An accountant is an agent who understands a higher-level goal ("file my taxes") and can execute the multiple steps required to achieve it.  

The underlying technology is now mature enough to make this possible. The latest AI models have made significant leaps in "reasoning" capabilities. They can now plan a sequence of actions, verify their work against a set of rules, and reflect on the outcome to self-correct, much like a human would. This ability to reason through complex tasks is the key that unlocks agent-like behavior and allows developers to build the "AI interns" that will power the next wave of software.  

The Carpenter's Toolbox: Why One AI Doesn't Rule Them All

As Sarah began exploring AI, she was bewildered by the sheer number of different models: OpenAI's GPT-4, Anthropic's Claude, Google's Gemini, open-source options like Llama and Mixtral. Her mentor offered a simple analogy: it's like a carpenter's toolbox. A carpenter would never use a sledgehammer to drive a delicate finishing nail. The tool must match the task.

Similarly, a savvy AI builder doesn't rely on a single, all-powerful model for every job. They might use a large, creatively brilliant (and expensive) model to generate a new marketing campaign concept. For summarizing customer feedback emails, they might use a smaller, faster (and cheaper) model. And for analyzing structured sales data, they might use a highly specialized open-source model that has been fine-tuned for that specific task.

This is now the industry standard. Companies are converging on "multi-model architectures," using an average of 2.8 different models within a single customer-facing product. This strategy allows them to precisely optimize for three critical variables: performance, cost, and use case specificity. For a startup, this approach is a game-changer for capital efficiency. Instead of paying a premium to use a top-tier model for every minor task, they can blend the best of the proprietary world with powerful and efficient open-source alternatives, dramatically lowering their operational costs while delivering a superior product experience.  

Renting vs. Owning: Choosing Your AI Engine

The mentor extended the toolbox analogy to explain the strategic choice between different types of models. Using a proprietary model from a major lab like OpenAI or Google is like renting a high-performance engine for your car. It's incredibly powerful, reliable, and easy to get started with—you simply connect to their service via an API. This is the fastest path to market for a startup that needs state-of-the-art capabilities out of the box. However, you are dependent on the landlord (the tech giant). They set the price, they can change the rules, and your business is ultimately built on their platform.

Using an open-source model is like owning your own engine. It requires more upfront expertise to acquire, modify, and maintain. You need the technical talent to fine-tune it and run it efficiently. But the payoff is immense: you have total control. You can customize it perfectly for your specific niche, you aren't beholden to another company's pricing whims, and your core intellectual property is your own.

Both paths are viable. Proprietary models from labs like OpenAI still hold a narrow lead at the absolute frontier of capability. But the open-source ecosystem is no longer a second-rate alternative. Models like those from China's DeepSeek or France's Mistral can "punch above their weight," often matching or even exceeding the performance of proprietary models on specialized tasks at a fraction of the cost.  

The strategic decision for a founder is not simply "which model is the best?" but rather, "what is our company's long-term model strategy?" This shift to a multi-model mindset is a fundamental change in how modern software is built. A non-technical founder's most important contribution is not selecting the specific models, but designing the workflow the product will enable. By meticulously mapping out the steps of the "agentic" process that solves a customer's problem, the founder creates the blueprint. A technical partner can then select the right "tools" from the carpenter's box for each step in that blueprint. The enduring value is created in the design of the workflow itself, not in the selection of any single model. In fact, the two trends are deeply intertwined: an agentic workflow is a multi-step process, and a multi-model architecture is the most efficient way to execute that process. A brilliant workflow design naturally leads to a defensible and efficient technical architecture.

Chapter 3: The Startup Playbook: How to Stake Your Claim on the Frontier

With a clear understanding of the landscape and the tools, it's time to chart a course. This chapter provides a strategic playbook for building a successful AI-native business, focusing on the specific advantages that a small, focused startup has over large, slow-moving incumbents.

Don't Boil the Ocean, Find a Creek

Sarah's initial instinct, like that of many aspiring founders, was to think big. She was tempted by the idea of building a "better ChatGPT" for small businesses. Her mentor quickly steered her away from this path with a crucial piece of advice: "You can't build a better Swiss Army knife than the companies that invented it. But you can build the world's best corkscrew." This is the power of verticalization.

The data is unequivocal: the dominant strategy for successful AI-native builders is a focus on "agentic workflows and vertical applications". While the giant labs compete to build ever-larger, more general-purpose models (the Swiss Army knives), the greatest opportunity for startups lies in creating highly specialized tools for specific industries and functions (the corkscrews). After her conversation, Sarah abandoned her vague plan for a general business assistant and decided to focus on a single, high-pain problem: inventory management for independent coffee shops. This sharp focus is a startup's superpower.  

To help founders identify these opportunities, the following matrix connects common business headaches with the potential for an AI-powered solution.

Table 1: The Entrepreneur's AI Opportunity Matrix

Your Secret Weapon is Memory (The New Moat)

When Sarah first launches her coffee shop inventory AI, it's helpful but generic. However, as her first customers use it, the system begins to learn. It learns the unique patterns of each shop: the unexpected Tuesday morning rush at the cafe near the university, the fact that rainy days lead to a spike in latte sales, the specific delivery schedule of the local organic milk supplier. This accumulated knowledge—this "context and memory"—becomes the AI's superpower. A generic AI from a tech giant could never know these hyperlocal, business-specific details. This becomes Sarah's competitive moat.

For founders of AI applications, "context and memory may be the new moats". This is how a startup can build a defensible business. The key is to design a system that turns this context into a "compounding advantage". The more a customer uses the product, the smarter and more personalized it becomes for them. This creates immense switching costs; the thought of starting over with a "dumb" generic competitor becomes unthinkable. This virtuous cycle of usage, data, and personalization is a powerful retention mechanism that locks in customers and protects the business from larger rivals.  

The Price is Right: Charging for True Value

With a valuable product, the next challenge is pricing. How should Sarah charge for her tool? A simple, flat subscription of $50 per month feels inadequate. Her AI is saving a large, high-volume cafe over $2,000 a month in reduced food waste, but it's only saving a tiny sidewalk kiosk around $100. A flat fee would undercharge the large cafe and overcharge the small one.

The unique economics of AI are forcing a revolution in software pricing. Many companies are abandoning one-size-fits-all subscriptions in favor of "hybrid pricing models" that combine a base subscription fee with charges based on usage or intensity. Taking this a step further, some pioneering companies are experimenting with pricing based purely on the results or value delivered. Imagine Sarah's pitch: a small monthly fee to cover base costs, plus a charge equal to 10% of the documented savings her AI generates for the coffee shop each month. This perfectly aligns her company's success with her customer's success, making it an incredibly powerful sales proposition. While many companies initially offered AI features for free to drive adoption, the market is maturing. More than a third (37%) plan to adjust their pricing in the next year to better reflect the value customers receive, signaling a clear shift towards value-based monetization.  

This move towards value-based pricing also illuminates an alternative path to growth. The headlines are often dominated by "AI Supernovas"—startups that sprint to $100 million in revenue in their first couple of years, often by burning through vast amounts of venture capital and operating with razor-thin 25% gross margins. This is a high-risk strategy available to only a select few. The ability to price based on value, however, opens up a more sustainable path. If a product can deliver a clear, measurable financial return, it can command premium prices. That revenue can then be used to fund growth, creating a capital-efficient, profitable business without total reliance on VC funding. This is a viable and attractive path for founders who want to build an enduring company.  

Building Your A-Team (When You're Not the Techie)

Sarah knows she has the vision and the industry expertise, but she can't build the product alone. As a non-technical founder, her most critical job is to assemble the right team.

This is one of the biggest challenges in the current market. Talent is a key differentiator, but it's also a severe bottleneck. The average time to hire a specialized AI/ML engineer now exceeds 70 days, and more than half of all companies (54%) report falling behind their AI hiring goals due to a limited pool of qualified candidates. The most successful AI teams are not siloed groups of coders; they are cross-functional pods that bring together AI/ML engineers, data scientists, and—most importantly—AI product managers. As the non-technical founder, this is your role: you are the ultimate product manager, the keeper of the vision, the person who deeply understands the customer's problem that the technology is meant to solve. Your job is to translate that vision so clearly that a technical team can bring it to life.  

Chapter 4: Navigating the Hazards: A Realist's Guide to Budgets, Risks, and Competition

Every frontier has its dangers. A successful expedition requires not just a destination but also a keen awareness of the potential pitfalls along the way. This chapter provides a sober, realistic look at the challenges of building an AI company, from managing unique financial pressures to fending off giant competitors and building a foundation of trust.

The True Cost of Your Expedition

Sarah's coffee shop AI business is growing. She's onboarding new customers, and the feedback is fantastic. Then, she gets her first big cloud computing bill, and it's a shock. She realizes that while her AI "intern" doesn't draw a salary, it does need to "eat" a tremendous amount of expensive electricity in the form of compute power.

This highlights the unique cost structure of an AI business, which evolves dramatically over its lifecycle. In the very early stages of product development, "talent is generally the biggest expense". Hiring the specialized engineers needed to build the initial product is the primary cash burn. However, a critical shift occurs as the product finds market fit and begins to scale. As more customers use the product more intensively, "cloud costs, model inference, and governance start to make up the majority of spend". This is a crucial financial planning insight. Unlike traditional software-as-a-service (SaaS) businesses where costs scale more linearly with customers, an AI business's operational costs can grow exponentially with usage, requiring careful management of compute resources.  

This trend is driven by an insatiable demand for processing power at the macro level. The industry has entered an industrial era of AI, marked by the construction of multi-gigawatt data centers. In this new reality, access to "power and land are now as important as GPUs". For a startup founder, this means the cost of compute is a strategic variable that must be constantly monitored and optimized.  

Giants in the Valley

One day, Sarah gets a scare. A headline announces that a major tech giant is launching a new suite of AI tools for small businesses. Her first reaction is panic. How can her tiny startup possibly compete? Her mentor calms her down by reminding her of the startup's inherent advantages. The giant's tool is generic, a Swiss Army knife designed to be mediocre for everyone. Her product is a specialized corkscrew, perfectly designed for the specific needs of her niche customers. She is faster, more focused, and has a deep, personal connection to her user base.

The threat from incumbents is very real. The reports predict that "the incumbents strike back as AI M&A heats up," meaning large companies will increasingly use their cash reserves to acquire innovative startups rather than trying to build competing products from scratch. However, the data also provides a powerful counter-narrative. AI-first startups are growing 1.5 times faster than their peers, demonstrating their ability to outmaneuver larger rivals. Most tellingly, AI-native companies are nearly four times more likely to have reached critical scale and proven product-market fit compared to established companies that are simply adding AI features to their legacy offerings (a 47% success rate versus just 13%). This data is definitive proof of the startup advantage. Agility, focus, and a culture built around AI from day one are winning advantages in this new market.  

Building on Solid Ground: Trust, Safety, and Doing It Right

A frantic call comes in from one of Sarah's best customers. The AI made a significant error in a milk order, costing the coffee shop hundreds of dollars. In that moment, Sarah realizes that trust is the single most important asset her company has. An AI that isn't reliable is worse than no AI at all. In response, she immediately works with her team to implement a "human-in-the-loop" system. The AI still does the heavy lifting—analyzing data and predicting needs—but it now drafts a purchase order that the shop owner must review and approve with a single click. This simple change perfectly balances the efficiency of automation with the control and oversight of a human expert.

This is a microcosm of a broader industry trend. As AI becomes more powerful, AI-related incidents are rising sharply. Businesses are increasingly concerned about mitigating risks related to inaccuracy, cybersecurity, and intellectual property infringement. The conversation around AI safety is maturing, moving away from abstract, long-term fears about "existential risk" and toward tangible, immediate problems like reliability, deception, and the monitorability of AI systems.  

There is still a gap between companies recognizing these risks and taking meaningful action. However, the smartest organizations are making AI governance a C-suite priority. For a startup, this means the founder must own the responsibility for safety and trust from day one. Implementing practices like having employees review a portion of AI-generated outputs is not a sign of failure; it's a mark of a mature and responsible organization. In fact, building a product that prioritizes transparency and reliability can become a powerful competitive advantage. Some researchers are even exploring what they call a "monitorability tax"—the idea of deliberately accepting a slightly less capable but more transparent and understandable AI system. In a market that is becoming crowded with powerful but opaque tools, a startup that wins on trust can build the most loyal customer base of all.  

Conclusion: Your First Step into the Frontier

Sarah's journey began with a feeling of being overwhelmed and left behind by a technological wave she didn't understand. It ends with her as a confident founder, leading a growing business. She didn't need to become a machine learning engineer. She needed to understand the new landscape, learn the function of the new tools, and apply them to a problem she knew intimately. She succeeded by focusing on a vertical niche, building a defensible moat through the compounding advantage of context and memory, and creating a business model that aligned her success with that of her customers. She was prepared for the unique financial realities of an AI company and built her brand on a foundation of trust and reliability.

Her story holds the core lesson for every aspiring entrepreneur in this new age. The AI revolution is not about becoming a technical expert. It is about being an expert in a problem. Your deep, nuanced understanding of your customers, your industry, and the frictions they face every day is the most valuable asset you possess. AI is simply the most powerful tool that has ever been created to solve those problems.

The path forward does not begin with code; it begins with empathy. Don't try to boil the ocean. Go back to the Opportunity Matrix in Chapter 3. Pick one "Founder Headache" from that list—or one of your own—that you know inside and out. Spend the next week talking to ten people who live with that headache every day. Don't ask them about AI. Ask them this: "If you could hire a smart, fast, but very literal-minded intern to help you with this, what exact, step-by-step instructions would you give them?"

The answer to that question is not a technical specification. It is the beginning of your company.