Centralizing: The Three Driving Forces Shaping the Future of AI
Investors are suffering from an AI Hangover now, but Capital, Data, and Distribution are shaping AI progress and impact.
Hello Cyber Builders 🖖,
Artificial Intelligence (AI) will transform industries and economies, but it will not happen overnight. This week, I want to focus on the three fundamental forces driving this change: capital, data, and distribution, to understand where AI is heading and how to stay ahead.
First, we are in a very particular context: many investors, journalists, and entrepreneurs are talking and posting about AI “bubble.” A bubble that would result in a massive investment since early 2023 without seeing effective results.
Investors are having an AI Hangover.
As with any hangover, it will pass. It’s a short-term sentiment. It won’t say. That’s why I think it is an excellent topic to discuss: taking a step back and looking at what is shaping the AI-driven industry and realizing—yes—it will take a little time to impact the bottom line of business.
AI is about centralization. Without doing politics or promoting the virtues of decentralization (e.g., the Web3 advocates' position), it’s clear that AI is concentrating power in the hands of a few. Only a select number of organizations globally possess the resources to develop, deploy, and offer AI systems at scale.
This concentration is not accidental; it’s driven by three fundamental forces—capital, data, and distribution—structuring AI applications and determining who succeeds.
The driving forces—capital, data, and distribution—determine who will thrive in this AI-driven future. These forces are centralizing forces. The large incumbents have a clear advantage because they can secure capital faster, capture more data, and leverage existing distribution channels.
In this post, I’ll break down each force, exploring what they are, why they matter, and how they shape the AI landscape. I'll also discuss what these forces mean for Cyber Builders—those of us creating security products, designing security programs, and leading cybersecurity efforts in organizations. Finally, I’ll identify the winners of each force, highlighting the companies and entities best positioned to capitalize on the AI revolution.
Do you have an AI Hangover?
I recently covered the investor market's state and how numbers cover an overall cautious or even negative sentiment.
Many are questioning the real economic value of AI apps and their return over investment horizons. The market has been spotting AI deals for almost two years and has invested billions in early-stage startups based on the promise of existing businesses to be built. Now, VCs and board members ask founders, “Guys, when will you charge anything to customers?”
Stuart Allsopp recently emphasized the potential risks in the AI-driven market (link), stating, "Mega cap tech stocks beginning to lead on the downside, which may suggest the AI bubble is beginning to burst." He pointed out that economic imbalances, including wide fiscal deficits and low savings rates, could threaten the current market valuations. Allsopp also highlighted the inflation of valuations due to AI-driven market enthusiasm with few real-world solutions, indicating that the market is bumping up against physical and human resource constraints.
The hangover will pass. There are many different cases of dissolution. The startups that presented themselves as AI luminaries were only thin GPT4 wrappers. The founders asked for a large pool of GPUs but needed accurate data on how to train valuable models. The companies who thought they would revolutionize a vertical market without any prior experience, knowledge, or personal network.
All have missed the driving forces of AI. It is time to highlight them.
Capital: The Fuel Behind AI’s Unstoppable Growth
Sequoia Capital’s article, AI’s $600B Question, reveals the monumental scale of investment required to stay competitive in the AI space. What started as a $200 billion problem has ballooned into a $600 billion challenge, driven by the need for vast capital expenditures (CapEx) on GPUs and data centers.
These resources are crucial for training large language models (LLMs), with companies like Microsoft, Google, Facebook, OpenAI, and Anthropic making multi-billion-dollar commitments to secure their place in the AI arms race.
“It’s easy to calculate this metric directly. All you have to do is take Nvidia’s run-rate revenue forecast and multiply it by 2x to reflect the total cost of AI data centers (GPUs are half of the total cost of ownership—the other half includes energy, buildings, backup generators, etc.). Then you multiply by 2x again to reflect a 50% gross margin for the end-user of the GPU, (e.g., the startup or business buying AI compute from Azure or AWS or GCP, who needs to make money as well).”
Large companies' quarterly results confirm that back-of-the-envelope computation. On average, Microsoft, Meta, and Google spend 12 billion dollars on Generative AI. Employees and hardware are the two main components of these investments.
Who Are the Winners?
Tech giants like Google, Microsoft, and Amazon are the winners in such a capital-intensive environment. They have the financial muscle to acquire and sustain the infrastructure to develop and train foundational models. A few well-funded startups like OpenAI and Anthropic have managed to enter the game, but realistically, only about ten companies worldwide have the resources to compete at this level.
What It Means for Cyber Builders
This centralization of capital presents a unique challenge for Cyber Builders. The resources to develop advanced AI are increasingly concentrated, making it harder for smaller players to compete.
However, AI models are now essential to any software, meaning you must strategically incorporate them into your products. Any product manager in the industry has an AI roadmap.
So, will all software be hosted in a large public cloud, or will data be sent to AI champions API? Will it become a mandatory architecture for software vendors?
I don’t think so. Alternatives exist and are developing:
Small models are getting better and better, often outperforming one-year-old models. For example, the good (already) old ChatGPT 3.5 that blasted everyone's minds in late 2022 is deprecated by models you can run on your laptop.
Moreover, these tiny models are coming to the edge. By the edge, I mean the device in your pocket, such as your iPhone. In 2025, we will probably see many new applications running directly on phones without needing the cloud.
So fortunately for our autonomy and freedom of choice, the emergence of smaller, more efficient models offers a lifeline. These models, which can be run on more modest infrastructure, reduce costs while maintaining control over the AI.
Data: The Lifeblood of AI
We’ve all heard “data is the new oil,” but that analogy doesn’t cut it. Unlike oil, where value is in volume and gone once used, most data’s worth in AI isn’t about quantity—it’s about private, specific, and real-time data.
You may think an enormous amount of data is needed to train foundational models. You are right, and I am not talking about that here.
I look at the use cases for the rest of us—not the large, heavily capitalized companies able to train the foundational model.
Cyber Builders looking to build a companion tool for their daily operations or start a cybersecurity startup need to look at specific data. If not, you can expect that over time (I experienced it myself), the random Claude or ChatGPT answer will be better.
Private Data
Private data is gold because it’s exclusive. This isn’t just publicly available stuff; it’s your customer databases, experimental results, or proprietary insights. It is not a set of data you scraped over the web.
Look at the rise of RAG applications like DUST and Cisco Motific, which create platforms to integrate private data with AI models securely. These applications are not just playing with public datasets—they’re adding value by using data that no one else has (e.g., your data).
Specific Data
The future of AI isn’t broad and general—it’s specific.
Google’s work in healthcare with Med Palm2 and cybersecurity with Sec Palm2 shows where we’re headed: AI models tailored to niche fields, using specialized data to deliver precision. Like going to a specialist doctor for expert care, you’ll use specialized AI to get deeper insights into specific domains.
Once again, if you do not provide specific data for your applications, you will face ChatGPT competition.
Real-Time Data
Data’s value skyrockets when it’s fresh. Foundational models will always be limited because they are trained on a specific date and are unaware of the latest news or facts. That’s why it is so essential to secure access to real-time data to answer basic questions such as “What’s happened today?” or “Who is the prime minister of Belgium or France?” (the answer changes quite often 😄).
OpenAI’s deals with The New York Times, El País, and Le Monde show why real-time data matters. Sure, it’s excellent if AI can tell you about ancient Greece, but it’s even better when mixing that with what’s happening today in politics, sports, or the economy. Real-time, curated data sets are essential for AI to provide relevant and actionable insights.
What It Means for Cyber Builders
For Cyber Builders, the data security field is still in its infancy. We’ve seen Data Loss Prevention (DLP) tools from platforms like Google Drive and Microsoft, but let’s be real—data security is still complex. Why? Because protecting data isn’t just about technology; it’s about governance and understanding what data you have. Most companies aren’t data-centric—they focus on customers, products, or worse, their organizational structure, which changes every couple of years anyway.
It is a recurring motto of the data segment in the IT industry to have “data governance.” This old idea asks organizations to identify, structure, and prune (e.g., remove duplicates and incomplete entries) to get the most out of their data.
To get it right, data must be at the core of your security strategy. This means making data as much of a priority as endpoint security has become. Every aspect of your security policy and tooling should revolve around protecting data.
Cyber Builders also care about data security during training and inference of AI models. Something that is often overlooked.
New solutions must be easier to deploy and manage because data will ultimately fuel AI’s success. If you’re a Cyber Builder, this is your call to action: make data protection your top priority because, in the AI-driven future, it’s your most valuable asset.
Distribution: Connecting AI with Real-World Impact
Distribution is the driving force that connects AI capabilities with real-world applications and end users. As AI evolves, the focus will shift from broad, general models to specialized solutions tailored to specific industries and use cases. Data remains crucial, but access to users and use cases is equally important.
Take, for instance, the rumor of Harvey, a legal tech AI startup acquiring an existing large law firm (link), gaining private data and a direct line to potential customers and valuable use cases.
This isn’t just about training models or building software—it’s about understanding where people spend their time, what tasks need automation, and how to deliver meaningful services.
Since the early 2023 boom around Generative AI, many companies have automated customer support, cold emailing, and similar tasks—straightforward examples that don’t require deep domain expertise.
But moving forward, should we expect to see tech startups acquiring non-tech companies that have yet to digitalize fully? Or should we expect the reverse - large traditional companies transforming into digital and AI-driven businesses? The older readers I have remembered, the same questions were asked at the peak of the Internet bubble in the late 90s.
In both cases, M&A and acquisitions aren’t just about technology; they’re about tapping into established customer networks and industry knowledge essential for distribution.
To understand these trends, look at the article “The Future of AI is Vertical” from Bessemer Venture Partners. The next significant wave of AI innovation will be driven by Vertical AI, which refers to AI solutions designed for specific industries rather than generalized, horizontal applications. Vertical AI is unlocking new market opportunities by addressing niche areas previously considered too small or complex for traditional SaaS solutions. These AI-driven solutions, especially in industries like healthcare and legal services, are helping companies streamline tasks, improve decision-making, and access new levels of automation that were not feasible before. Bessemer predicts rapid growth in this space, expecting multiple Vertical AI startups to reach over $100 million in annual revenue within the next few years.
For AI to succeed, it’s not just about building a great model; it’s about getting it into the hands of users who need it. That’s where distribution comes in—bridging the gap between AI’s potential and its practical impact.
What It Means for Cyber Builders
For Cyber Builders, distributing AI tools presents both a challenge and an opportunity. As AI becomes more specialized, it’s crucial to understand your users' specific needs and how to integrate AI into their workflows. This means building tools and ensuring they reach users through the proper channels.
In cybersecurity, this could involve developing AI solutions specifically designed for particular industries or use cases and effectively distributing those tools. It also means considering how your security tools can be integrated into larger AI ecosystems, ensuring they are accessible to the organizations and users who need them most.
But don’t miss another impact. It also means that Cyber Builders within the organization, professionals such as SOC Analysts, Penetration Testers, Compliance Consultants, and Detection Engineers, will develop specific AI tools. AI will be used by integrators, service providers, consultants, and vendors. These additional entities will integrate AI into their particular uses and make its benefits available to a wider range of users.
I would bet that this automation will be driven by security practitioners themselves, not their vendors.
AI is a systemic technology. It takes a lot of time to see productivity impacts.
Even though these three forces are shaping a future fueled by AI, AI’s impact on productivity and employment may not be as rapid or dramatic as some fear. Research, especially from prominent economists like Daron Acemoglu, suggests that the initial gains in productivity could be marginal, around 0.5% over the next decade.
AI is a systemic technology requiring more than just the technology itself to thrive—it needs capital, human resources, regulatory frameworks, and cultural shifts, much like the widespread adoption of electricity and automobiles in the past.
For AI’s promise to be realized, society needs to progressively adapt its systems, including training programs, antitrust laws, and measures to ensure inclusivity, such as addressing biases and promoting diversity within the AI space. This gradual shift allows regulators and industries time to adjust, but it doesn’t mean there’s room for complacency.
European Cyber Builders must lead in shaping AI inclusively and fairly, provided they actively build the necessary frameworks today. AI can ultimately benefit everyone rather than widening societal inequalities.
Conclusion
In summary, AI today is driven by the powerful forces of capital, data, and distribution. These factors are concentrating power in the hands of those with access to specific capacities, large pools of capital, and established channels. Incumbents hold a massive advantage because of their access to these resources, and while technology is democratizing, leveraging these forces remains key to success.
For Cyber Builders, whether you’re founding a startup or working within an organization, your path to success lies in the data you have, the use cases you serve, and the automation AI can provide. Even as centralization favors large players, the real advantage lies in how you apply AI to solve real problems. If you’re inside an organization with access to private, real-time data, you’re already positioned to leverage AI and come out ahead.
I am curious about what you are building. Let’s connect, and feel free to comment.
Laurent 💚
Go Deeper with
who had a nice post series about LLM in Enterprise: