Opinion: 3 Main Drivers Will Define the Next Phase of the AI Race — and Only a Few Companies Actually Have Them

The AI trade is back in full force after taking a well-earned pause through much of early 2026. Not every AI play has reclaimed all-time highs, but semiconductor names are firmly back in the driver’s seat as demand stays hot and the promise of agentic and physical AI shifts from concept to reality.

The S&P 500 has been relentlessly pushing higher, logging 23 all-time highs by early June 2026 alone and crossing the 7,600 threshold for the first time, yet questions persist about whether the same winners that powered markets over the past two years can keep doing the heavy lifting. With enthusiasm concentrating around the energy and AI chip layers of the stack, specifically what NVIDIA (NASDAQ:NVDA | NVDA Price Prediction | NVDA Price Prediction) CEO Jensen Huang famously described as a “five-layer cake” at CES 2026, real questions linger about what the next chapters of the AI revolution look like and which companies are positioned to benefit.

That five-layer framework, which Huang elaborated at Davos and again at GTC 2026, stacks energy, chips, cloud infrastructure, AI models, and applications into a tightly coupled system. The early innings of the AI boom rewarded layers one and two most directly. The next phase, where the application layer matures and monetization catches up to capital expenditure, is where the real competitive sorting will happen.

AI benefits will eventually spread more broadly, but they will concentrate in certain areas first. It is easy right now to discount software companies relative to model makers and chip plays. But as AI monetization becomes a tangible reality rather than a projection, it is worth identifying the structural drivers that will separate durable winners from transient ones.

It’s time to think about a new set of moats when evaluating the AI plays

Picking winners at the application layer is genuinely hard, in part because transitioning from a SaaS-first model to an agentic-first model is not a smooth or guaranteed process. The more useful question is: which firms have the structural advantages to win at that layer, the layer that Huang himself considers critical to the entire stack’s value?

The answer comes down to a new type of competitive moat. Pre-AI era moats, things like brand loyalty or distribution networks, may prove less durable in a world where agents can autonomously source, switch, and transact. Some AI-era moats, by contrast, could function more like launch pads than defensive walls, compounding advantage as the technology scales. Three pillars stand out above all others: the data moat, the physical asset moat, and the automation opportunity.

The data moat

Proprietary datasets remain the most cited structural edge in AI, and for good reason. Data is the fuel that powers model performance, and without a supply of it, even a well-funded competitor struggles to catch up. But the more nuanced version of this argument is becoming more important as the industry approaches a “data wall,” the point at which the stock of publicly available training data is largely exhausted. What matters increasingly is a firm’s capacity to generate new, high-quality synthetic data rather than simply possessing a large existing dataset.

The deeper layer still is ontology: how data is structured, contextualized, and made actionable for an AI agent. Even the richest dataset in the world is worth little if it cannot be harnessed effectively. What is the ground truth? What context does an agent need to be genuinely useful to an enterprise customer? These are the questions that determine whether a data moat translates into a real competitive advantage.

Palantir (NASDAQ:PLTR) has built what may be the widest ontology moat in enterprise software, and its recent results suggest the market is beginning to price that in. In Q1 2026, Palantir reported total revenue of $1.63 billion, up 85% year over year, with U.S. commercial revenue surging 133% to $595 million. Management subsequently raised full-year 2026 guidance to $7.65 billion to $7.66 billion. That growth is driven by its Artificial Intelligence Platform, which grounds large language models in client data and context, effectively acting as the enterprise operating system that makes AI agents useful in production environments rather than just in pilots.

The physical moat and automation prospects

Physical asset moats are the second major differentiator. The category covers power infrastructure, manufacturing capability, rockets for orbital compute and connectivity, and the full stack of assets that go into what Huang calls an “AI factory.” These are hard to replicate quickly, which makes them among the most durable moats in the AI era. Energy and physical bottlenecks are still the single biggest constraint on how fast the broader AI buildout can proceed.

SpaceX and Tesla (NASDAQ:TSLA) hold top-tier positions on this dimension. Tesla’s Optimus humanoid robot program has moved from demonstration to production: the company ended Model S and Model X manufacturing in early May 2026 and is converting its Fremont factory into what it intends to be the world’s first large-scale humanoid robot production line, with mass production targeted to begin in late July or August 2026. A second facility at Gigafactory Texas is targeting a long-term annual capacity of 10 million robots per year. Physical AI, as 2026 is increasingly illustrating, is moving from lab demos into production-grade deployment across manufacturing and logistics. Tesla’s combination of Optimus, its proprietary FSD compute, and its vehicle manufacturing scale gives it a physical moat that few competitors can match.

Finally, automation represents perhaps the most near-term AI monetization opportunity. Whether the form is warehouse robotics or AI tools that allow smaller engineering teams to accomplish the output of much larger ones, the fundamental value proposition is doing more with less. As enterprise focus shifts toward revenue per employee as a key productivity metric, investors are likely to become increasingly willing to accept higher capital expenditure when it is credibly paired with lower ongoing operating costs. Palantir’s annualized revenue per employee reached $1.5 million in Q1 2026, a figure that illustrates what AI-leveraged operating models can look like at scale. That combination of structural data moats, physical asset advantages, and automation upside is the framework that will define the real winners of the next AI phase.

Editor’s note: This article was updated to reflect NVIDIA’s five-layer AI stack framework as described by Jensen Huang at CES 2026 and GTC 2026, Palantir’s Q1 2026 revenue of $1.63 billion (up 85% year over year) and raised full-year guidance of $7.65 to $7.66 billion, Tesla’s Optimus production timeline at the converted Fremont factory, and the S&P 500’s 23 all-time highs recorded through early June 2026.