Why Founder-Led Companies Outperform
There is a compelling investment thesis hiding in plain sight: companies still run by their founders tend to outperform those led by externally hired, board-appointed CEOs. The evidence is not anecdotal. An analysis spanning roughly 11,000 stocks over a 30-year horizon suggests that founder-led businesses can deliver on the order of three times the stock price performance of their non-founder peers over very long periods. The intuition behind the numbers is straightforward. Founders carry a long-term vision and a stubborn willingness to overcome any obstacle in pursuit of it. They are not merely stewards of a business; they are builders who treat impediments as problems to be engineered around rather than reasons to retreat.
That lens is especially useful right now, because the artificial intelligence buildout is being shaped by exactly this kind of founder mindset. Three companies in particular — Dell, Oracle, and Palantir — are riding the same agentic AI wave, each from a slightly different takeoff point.
The Two Phases of AI
To understand where value is accruing, it helps to separate AI into two distinct activities. The first is the foundational model layer: the work of training models to "think." This is the compute-hungry phase that built the first great wave of demand for graphics processing units. The second is the agentic layer: asking those trained models a question, running inference, and getting back a useful result that can drive action.
The first wave was defined by GPUs powering the training of foundational models. The next phase pushes inference outward and downward — even into personal devices. A notable strategic signal recently came when the dominant chipmaker announced its move into the CPU business, with high-memory CPU-GPU combination chips designed to be embedded directly into laptops. The implication is significant: a laptop itself can begin to perform the agentic, inference-side work locally rather than relying entirely on distant data centers. The same company that supplied the picks and shovels for training is now positioning to power inference at the edge.
Dell: Winning on Two Fronts
Dell is benefiting from both phases at once. On the infrastructure side, its AI server business has been blowing past expectations — server growth well ahead of street estimates, a higher guide, and a growing backlog. This is shaping up to be a genuinely large business rather than a temporary spike.
The differentiator is data sovereignty. A casual question — say, the best subway route to the New York Stock Exchange — can happily be answered in the public cloud. But a hospital, an airport, or the defense department cannot push sensitive data into a public cloud. These organizations need on-premises servers to run their AI workloads against foundational models inside their own walls. That is precisely the niche Dell's AI server business fills.
This capability also explains Dell's traction with the most demanding customer of all: the government. A major Pentagon deal underscores the point. The government is not an easy customer — it demands extensive compliance, safety, and security — but the infrastructure required to satisfy it is exactly what large enterprises want too. Companies must protect their intellectual property, the data they use to generate profit, and they must reassure their own customers that proprietary information will not escape. A vendor that can meet the government's bar is well positioned to meet the enterprise's bar. In that sense, Dell's government relationships function as a template for how to serve enterprise at scale.
On the second front, Dell stands to gain from the new generation of Nvidia CPU-GPU chips going into its laptops. The prospect of a portable machine capable of doing real agentic work locally is genuinely exciting — enough to tempt even committed users of competing laptops to add a Dell device to their toolkit.
Palantir: The Trusted Connector
Palantir sits at a different point on the same wave. Its role is that of the trusted endpoint connector. An enterprise's data lives inside existing software systems, and to make that data useful through AI, something has to reach in and touch it via the foundational models. Palantir provides that bridge.
There is a deeper problem it solves as well. Without a human overseeing the process, automated AI workflows degrade into what is colloquially called "AI slop" — output that looks plausible but cannot be trusted. Palantir addresses this through its forward-deployed engineers, who wire together disparate systems and keep humans in the loop. The result is that an enterprise can finally extract useful information from data that has been buried and fragmented across its ERP and CRM systems, pulling it all together in a way that remains fully secure. That combination of connectivity, human oversight, and security is what makes it valuable across both government and commercial contracts.
Oracle: Boring, Profitable, and Misunderstood
Oracle rounds out the trio. When excitement built around the AI data centers it was constructing, its stock spiked dramatically — enough that its founder, who owns roughly 40% of the company, briefly leapfrogged his peers to become the wealthiest person in the world. Markets, however, are fickle. Skeptics soon raised two objections: the debt being taken on to fund the buildout, and the risk of customer concentration.
Both objections deserve scrutiny, and both look weaker on inspection. First, the assets being financed are medium- to long-term in nature. GPUs are likely to last longer than people assume — perhaps a decade, with roughly two years spent on training workloads and eight on inference. When you are building a long-lived asset, putting debt against it is not reckless; it is sound financing. Second, the fear of customer concentration is tempered by how quickly demand can be reallocated. In one telling instance, a data center that OpenAI passed on was scooped up by Meta within a remarkably short window. Capacity in this market does not sit idle for long.
What Oracle is really doing is building the boring but profitable backbone of AI infrastructure, and the market is beginning to believe in that story again after a sharp retracement and an even sharper recovery — the stock has climbed nearly 30% year to date. The number to watch in the next earnings report is the private cloud business, which is expected to be strong and arguably matters more than any other line item. The underlying question is whether a founder with a long history of vision and execution got ahead of the central bottleneck in AI. The signs suggest he did. One vivid example of that founder's resourcefulness: when one of the company's data centers could not get electricity delivered to it, the team lit it up using Bloom fuel cells — a "bring your own energy" solution to a problem that would have stalled a less determined operator.
The Through-Line
What unites these three companies is not merely exposure to AI; it is the founder's instinct to identify a bottleneck and engineer around it. Dell is solving the problem of secure, on-premises and edge compute. Palantir is solving the problem of trustworthy connection between AI and real enterprise data. Oracle is solving the problem of raw infrastructure and even the energy to run it. Each addresses a different constraint on the same agentic AI wave, and each is led by someone with the vision and the stubbornness to see the buildout through. For investors trying to position around the growth of agentic AI, that shared founder DNA may matter as much as the technology itself.