Two Bets for 2026: Riding the AI S-Curve from Compute to Application

The current phase of artificial intelligence is not just another tech cycle. It is a foundational infrastructure boom, drawing parallels to the monumental buildouts of the past. Just as the transcontinental railways or the interstate highway system rewired the nation's economic geography, the AI infrastructure arms race is laying the physical and digital rails for the next paradigm. This is a story of massive scale, where capital expenditure has exploded from roughly $100 billion in 2023 to more than $300 billion in 2025 among the largest tech players, a figure that could soon approach half a trillion dollars. The consensus estimates for this spending have consistently lagged reality, with real growth exceeding 50% in both 2024 and 2025. This persistent underestimation is a hallmark of a true S-curve inflection point, where the initial, cautious projections fail to capture the exponential ramp-up of investment.

The strategic importance of this buildout is now undeniable. According to Fidelity, the AI boom has accounted for roughly 60% of recent US economic growth, overshadowing all other sectors. For investors, the central question for 2026 is whether this massive capital outlay will ultimately generate returns that justify its cost. The divergence in stock performance among AI hyperscalers already signals a market sorting through the winners. As the Goldman Sachs report notes, investor rotation is away from infrastructure companies where earnings growth is pressured and capex is debt-funded, toward those demonstrating a clear link between spending and revenue.

This shift is driven by a critical transition in AI's own adoption curve. The technology is moving decisively from isolated proof-of-concept projects to full-scale production deployment. This move forces a fundamental redesign of enterprise infrastructure. The old model, built for periodic data processing, is ill-suited for AI's new demands. Recurring workloads mean near-constant inference-the act of using a model in real-time processes-which can lead to escalating costs and data sovereignty challenges. The economics of consumption are waking up organizations: while inference costs have plummeted, usage has exploded, sending monthly AI bills into the tens of millions for some. The solution is no longer a simple cloud-or-on-prem choice, but a complex orchestration of compute platforms tailored to each workload. This is the core of the infrastructure layer being built today.

Strategy 1: Bet on the Compute Layer (Semiconductor Enablers)

The initial phase of the AI buildout was about laying the physical rails. Now, the market is shifting to the next leg of the S-curve: who provides the fundamental compute power to run on those rails. For investors, this means a strategic pivot from pure infrastructure builders to the semiconductor enablers that are best positioned for exponential growth as the deployment accelerates.

Goldman Sachs Research frames this transition clearly. The next phases of the AI trade will involve AI platform stocks and productivity beneficiaries, not just the initial infrastructure builders. This is a market sorting through winners. The divergence in stock performance among hyperscalers already shows investors are being selective. They are rotating away from infrastructure companies where earnings growth is pressured and capex is debt-funded, and toward those demonstrating a clear link between spending and revenue. The semiconductor layer, by contrast, sits at the very beginning of that value chain, with demand that is both well-established and outstripping supply.

This sets the stage for a market broadening beyond the mega-cap tech giants. Asit Sharma, a senior analyst at The Motley Fool, predicts that smaller semiconductor stocks and software stocks will outperform in the coming years. The logic is straightforward: the risk profile here is fundamentally different. While data center builders face massive, upfront capital expenditure and uncertain monetization timelines, semiconductor companies are selling a product with a proven, high-margin demand. As one report notes, chip stocks have virtually none of the risks faced by infrastructure players, including the burden of depreciating GPU inventory and unclear business models. This makes them a more resilient bet as the buildout matures.

A prime example of this thesis in action is Broadcom. The company is carving out a significant niche in custom AI chips, and its growth is set to explode. Its AI revenue is forecast to soar from $20 billion in fiscal 2025 to over $50 billion in fiscal 2026 and potentially reach $100 billion in fiscal 2027. This trajectory represents a classic exponential ramp, moving from a meaningful contributor to a core pillar of the business. It illustrates the power of being an enabler: as the AI infrastructure complex expands, the demand for the foundational chips that power it grows even faster.

The bottom line is that the compute layer is where the first principles of the AI paradigm are being realized. These are the companies building the fundamental rails, and their growth is tied directly to the adoption rate of the technology itself. For a forward-looking investor, that connection to the core S-curve makes them a compelling bet as the AI trade evolves.

Strategy 2: Bet on the Application Layer (Productivity Platforms)

The AI S-curve is now transitioning from the foundational buildout to the operational phase. The market's focus is shifting from who builds the rails to who runs the trains. This is the strategic pivot for 2026: investing in the software platforms that will orchestrate AI at scale within enterprises, capturing value as adoption broadens from pilot projects to core business functions.

ServiceNow exemplifies this thesis. The company is not selling AI; it is building the backbone for automated digital workflows that now incorporate AI as a native layer. Its platform, which manages everything from IT service requests to HR processes and customer service, is being rearchitected into an AI-driven enterprise engine. By embedding thousands of pre-configured AI agents across these workflows, ServiceNow enables the automation of complex, multi-step tasks with minimal human oversight. This isn't a side feature; it's the company's central growth pillar, aimed at scaling AI from isolated demos into tangible business value. The strategic importance is clear: as enterprises move beyond proof-of-concept, they need a unified platform to govern AI agents and models, a role ServiceNow is actively shaping with tools like its AI Control Tower.

This shift is also reflected in the market's setup. While the mega-cap infrastructure builders have seen their valuations stretch, the software layer has lagged. ServiceNow's stock is down around 30% over the past year, significantly underperforming the broader market. This pullback has made valuations more attractive for long-term AI winners. The company trades at a premium, but that price reflects its position as a platform, not just a vendor. The key point is that software stocks, in general, have been a laggard, creating a potential entry point for investors betting on the next phase of AI consolidation.

The technical transition supports this bet. 2026 is the year of agentic workflows moving from prototype to production. As evidence shows, models like OpenAI's o3 and Claude 4 are achieving expert-level reasoning, but the real challenge is scaling them efficiently. The gap between a capable AI and a productive enterprise tool is closing. Platforms like ServiceNow, which can manage and orchestrate these agents, are positioned to capture the value of this transition. Gartner's projection that 40% of enterprise applications will embed AI agents by mid-2026 underscores the massive, near-term opportunity. The market is rotating away from pure infrastructure and toward these productivity beneficiaries, where the link between AI spending and operational efficiency is becoming direct and measurable.

Catalysts and What to Watch

The investment thesis for 2026 hinges on a bifurcated AI landscape. The market is in the early stages of sorting winners, and several near-term developments will validate or challenge this setup. Investors should watch for three key catalysts that will separate the enablers from the infrastructure builders.

First, the widening gap between chip stocks and AI infrastructure stocks will be the most visible signal of market rotation. The consensus is clear: smaller semiconductor and data center ecosystem stocks are set to outperform in the coming years. This divergence is already evident in stock performance, where software and chip names have lagged the mega-cap infrastructure giants. The logic is straightforward. As the buildout matures, the market is rotating away from companies where earnings growth is pressured and capex is debt-funded, toward those demonstrating a clearer link between spending and revenue. Nvidia remains the dominant infrastructure play, but the real growth story is in the enablers like Broadcom, whose AI revenue is forecast to $20 billion in fiscal 2025 and potentially reach $100 billion by fiscal 2027. The pace of this rotation will be a primary indicator of whether the market is truly pricing in the next phase of the S-curve.

Second, the pace of AI platform adoption will be validated by companies like ServiceNow. The company's strategic pivot to an AI-driven enterprise engine is a direct play on the transition from pilot to production. The key metric to watch is not just revenue growth, but the embeddedness of AI agents within core enterprise workflows. Gartner's projection that 40% of enterprise applications will embed AI agents by mid-2026 sets a concrete benchmark. For ServiceNow, success will be measured by its ability to scale these pre-configured agents across IT, HR, and customer service, moving from a software vendor to a platform that governs AI operations. The stock's underperformance-down around 30% over the past year-has created a valuation opportunity, but the real validation will come from tangible evidence of this platform becoming the default for enterprise AI orchestration.

Finally, the resolution of power constraints will determine the physical limits of the AI buildout. As AI workloads scale, electricity demand is rising faster than the US power grid-much of it built decades ago-was designed to handle. This is no longer a background issue but a central operational and strategic constraint. The catalyst here is the shift in data center strategy from passive energy consumers to active grid stakeholders. Operators are co-investing with utilities in infrastructure upgrades and deploying on-site generation and storage. This transition will be validated by the number of data centers achieving power density targets and the speed at which they can secure and deploy diverse power solutions, from renewables to natural gas with carbon capture. The companies that solve this problem will be the ones able to scale the compute layer, making power innovation a critical, non-negotiable factor for the entire infrastructure stack.