Over the past few months, one question has reverberated through boardrooms, investor forums, and hallway conversations between developers: Is software as we knew it dead? The short answer is no, but the road to winning with it has become noticeably steeper. At the core of this debate is not a simple technological fad, but a profound reconfiguration of how digital value is built, financed, and traded. Behind the noise is a concrete reality: Big Tech is investing unprecedented amounts in AI infrastructure, while the revenues generated by those same systems are still far behind. This gap, far from signifying a bubble, is creating a window of opportunity of several years for those who know how to navigate it.
The current picture shows a sector that spends around six hundred billion dollars annually on AI-related infrastructure, but barely earns a fraction of that figure. The two main actors in foundational models account for most of the revenues, and it is estimated that the break-even point between cumulative expenditure and revenues will not arrive until the beginning of the next decade. This implies that we are living in a phase of pure investment: capital flows because there is a firm conviction that generative artificial intelligence will capture a significant part of the world's spending on knowledge work. But that same conviction generates tensions: at some point, someone will ask why you spend so much more than you earn, and the market could take a breather. Therefore, any business plan must contemplate this possibility.
To understand what this means for a software development company, it's helpful to separate two big worlds: that of making AI and that of using AI. The first ranges from chip manufacturing to building data centers to training massive models. It's a titanic endeavor that consumes most of the capital, but isn't accessible to most startups: the barriers to entry are immense and the margins, while attractive, are concentrated in very few hands. The second world is where the game is really played out for software creators. The key question here is how companies are going to consume that trillion dollars of value generated by AI in the coming years. Will they do so directly from the foundational models, leaving little room for intermediaries? The experience of previous technological waves suggests not: in the same way that the client-server gave way to hundreds of enterprise applications, or the AWS cloud enabled a full generation of SaaS, AI models will be the foundation on which vertical solutions with real added value will be built.
Now, what protects these solutions against the fact that the foundational model itself can replicate them? The answer lies in competitive pits. The ones that really work are those that pure artificial intelligence, on its own, cannot emulate. For example, combining software with physical sensors—vision, touch, real-world data—creates a solid barrier because the big model isn't going to start making cameras or installing sensors. Marketplaces with network effects are also difficult to replicate: if a platform already connects buyers and sellers, a developer with a chat interface can't boot that network overnight. Proprietary, not public, data is another glaring moat: If your business relies on information that the model has never seen, it simply can't do what you do. And finally, there's the full-stack approach, where instead of selling software to a company, you become the company itself. A clear example is wealth managers that use language models to advise clients; OpenAI is not going to open a tax advisory office.
But there are two more subtle, and probably more relevant moats for software startups: the data flywheel and field-deployed engineering. The first is to start with a simple application and, as you learn how users work, build an increasingly specialized and differentiated product. The second involves having engineering teams on the ground that capture the real context of the customer, something that models cannot do on their own. Both require careful execution and continued investment. This is where the role of a technology partner like Q2BSTUDIO makes sense: by developing bespoke applications that integrate these advantages, you can build a value proposition that large model companies can't easily match.
Another critical dimension is the cost structure. Before the emergence of generative AI, most SaaS companies had similar spend profiles. Now the dispersion is enormous: there are companies that dedicate 70% of their revenues to computing (proprietary models), while others barely spend 10% on the consumption of third-party models. The key is not to fall into the trap of simultaneously funding a massive sales team and high token consumption. Either the product sells itself, thanks to an unstoppable user experience, or you need a sales team to drive it. What doesn't work is trying to pay for both with tight margins. In this sense, artificial intelligence can act as the new marketing: if the product is so good that users adopt it organically, it can allow for higher spending on computing, but only if the traditional acquisition cost has been eliminated.
The ecosystem is also seeing old growth models break down. Public valuations of software companies fell because average growth went from 30% per year to about 10%. That doesn't mean that the software is dead, but that investors have adjusted their expectations. For a founder, this implies that they need to have a very strong category conviction. If your product is confused with the undifferentiated mass of low-growth software, the operation will be meaningless. On the other hand, a company that leads a category with real growth, even if it is slower, can continue to generate attractive returns in the long term. What no longer works is assuming that the multiples of five years ago will return.
And here another question arises: what happens to software created before 2022? The answer is that about 10% became obsolete overnight, because it solved an AI problem that suddenly cost $5 per million tokens instead of $30 million in development. The rest are divided into three groups: one-third are safe because their proposition is different enough (e.g., proprietary data-driven forecasting), another third benefit from adding AI as additional functionality on top of a solid product, and the last third is directly threatened and needs to reinvent or disappear. Honesty when diagnosing which group you are in is crucial to make investment and product decisions. From a technology company's perspective, offering AI for enterprises with a focus on contextual integration and real business value is what makes the difference between being a commodity or being a must-have.
Another phenomenon that deserves attention is that of the so-called AI agents. Faced with the vision that everything will be controlled by a single overarching model, a more modular model is emerging: autonomous systems that perform specific tasks, often combining different models and data sources. This opens up a huge space to build vertical solutions that orchestrate these agents intelligently. But it also poses cybersecurity and governance challenges: how do you make sure an agent doesn't act unexpectedly or expose sensitive information? Cybersecurity then becomes a key enabler, not a later add-on. Similarly, managing the infrastructure where these agents run is critical: AWS and Azure cloud services provide the necessary scalability and elasticity, but require a well-designed architecture so as not to skyrocket costs.
On the analytical level, business intelligence is also undergoing a transformation. Traditional Power BI reports are no longer sufficient when you have language models that can answer questions in natural language. But the real advantage is not in replacing dashboards, but in combining them with systems that learn from usage patterns. Here, business intelligence services evolve into platforms that integrate data, models, and automation. A company that manages to bring all these pieces together – custom application development, artificial intelligence, cloud, cybersecurity and BI – will be in an unbeatable position to ride the current wave.
In closing, it is worth remembering that the initial question about the death of software is only a symptom of a much deeper transformation. Software is not dead; What has died is the naivety that building an application with a database and a couple of forms was enough to build a lasting business. Today, software must be intelligent, it must integrate with the physical world, it must protect against threats, and it must provide actionable insights. At Q2BSTUDIO, we understand that the key is to combine all those layers with strategic vision and impeccable technical execution. The challenge is greater, but the rewards are also greater for those who know how to adapt.
.jpg)
.jpg)
.jpg)
.jpg)