Construyendo la infraestructura de IA: Una guía para los estrategas en tecnologías avanzadas sobre la curva exponencial de desarrollo tecnológico

The build-out of AI infrastructure is no longer a trend; it is a fundamental paradigm shift, and we are in the steep, exponential phase of its adoption curve. This is a multi-year supercycle, and the numbers show it is just getting started. Global spending on the foundational hardware for AI is projected to reach $758 billion by 2029. That is the scale of the infrastructure rail being laid for the next computing era.

The trajectory is explosive. In the second quarter of 2025 alone, organizations increased spending on AI compute and storage hardware by 166% year-over-year, surging to $82.0 billion. This wasn't a blip. It was a clear signal that the adoption curve has entered its steep, accelerating phase. The dominance of servers in this spending is absolute. They accounted for 98% of the total AI Centric spending in that quarter, growing at a staggering 173.2% year-over-year.

This is the textbook pattern of exponential growth on an S-curve. The early, slow phase of experimentation is over. We are now in the steep climb where investment accelerates rapidly as the technology's value becomes undeniable. The focus is overwhelmingly on servers with embedded accelerators, which are the preferred platform for AI workloads and grew by over 200% in Q2. The forecast suggests this ramp will continue through 2026 and beyond, extending the mass deployment phase.

The bottom line is that this is not a cyclical boom. It is a structural build-out of the infrastructure layer for a new paradigm. The sheer scale of the projected spending, the explosive quarterly growth, and the near-total focus on servers all point to a multi-year supercycle that is firmly in its steep, exponential adoption phase. For investors, the question is not if this infrastructure will be built, but which companies are positioned to supply the rails.

The Infrastructure Layer: Key Rails for the AI Paradigm Shift

The AI paradigm shift is being built on a set of durable infrastructure layers, or "rails." While Nvidia's GPUs are the headline, the structural demand for AI inference workloads is spreading far beyond just chips. This is a multi-layered build-out, and the winners are the companies supplying the essential components that make massive clusters function.

Nvidia's dominance in GPUs is the foundation, but the growth in AI inference workloads is structurally dependent on a broader set of technologies. As clusters scale, the focus is shifting from raw compute to the systems that connect and support it. This is where Broadcom emerges as a key beneficiary in the AI networking market. The company's backlog tells the story: it has around $20 billion of non-accelerator AI backlog as hyperscalers deploy clusters that exceed 100,000 compute nodes. This demand is for high-speed, high-bandwidth, low-latency interconnects, moving data efficiently between thousands of GPUs. Broadcom's strength in switches and optical networking components positions it as a critical rail for the supercycle.

Another fundamental rail is high-bandwidth memory (HBM). The demand for this specialized memory, which provides the massive bandwidth needed for AI training and inference, is creating a structural shortage. This scarcity is a powerful tailwind for suppliers. Micron's stock, for instance, has rallied 228% as it captures this surge in demand. The company is not just a beneficiary; it is a key enabler of the compute power that drives the entire stack.

The bottom line is that the AI infrastructure supercycle is a multi-layered build-out. Nvidia provides the compute engines, but Broadcom supplies the nervous system for the cluster, and HBM manufacturers like Micron provide the high-speed memory that fuels it. These are the durable rails of the paradigm shift, moving beyond the hype of individual chips to analyze the core beneficiaries of the structural demand for AI infrastructure.

Financial Impact and Market Selectivity: Separating the Rails from the Noise

The infrastructure boom is translating to corporate financials with explosive force, but the market is maturing from broad bets to selective picks. The divergence in stock performance is the clearest signal that investors are no longer rewarding all AI big spenders equally. They are rotating away from companies where growth in operating earnings is under pressure and capex is being funded via debt, while doubling down on those demonstrating a clear link between spending and revenue.

This selectivity is underscored by a persistent gap between analyst forecasts and reality. Consensus estimates for AI hyperscaler capital expenditure have consistently underestimated actual spending. Real growth in this spending has exceeded 50% in 2024 and 2025, far outpacing the projected ~20% that analysts had penciled in. This pattern of underestimation has now led to a major upward revision in forward-looking capex projections. The consensus estimate for the group's 2026 capital spending is now $527 billion, up from $465 billion at the start of the third-quarter earnings season. The sheer scale of this committed spending confirms the multi-year supercycle is in full swing.

Yet, the market is starting to price in the quality of that spending. The average stock in a Goldman Sachs basket of AI infrastructure companies returned 44% year-to-date, a powerful rally. But this return vastly outpaced the consensus two-year forward earnings-per-share estimate for the group, which grew just 9%. This disconnect highlights the risk: the timing of an eventual slowdown in capex growth poses a direct threat to these companies' valuations. The performance of stocks across the large public AI hyperscalers has shown a dramatic loss of correlation, falling from 80% to just 20% since June. This dispersion is driven by investor confidence in whether AI investments are generating tangible revenue benefits.

The bottom line is a market in transition. The initial phase rewarded infrastructure builders like Nvidia and Broadcom for their role in laying the rails. The next phase, as Goldman Sachs Research notes, will involve AI platform stocks and productivity beneficiaries. For now, the focus is shifting to companies where the capital expenditure is not just a cost, but a proven driver of future earnings. The global semiconductor market, a key beneficiary of data center infrastructure demand, could grow by more than 25% in 2025, reaching an estimated $975 billion. But within that massive growth, the winners will be the select few who can convert spending into sustainable profits.

Catalysts, Risks, and What to Watch for the Exponential Trajectory

The infrastructure supercycle is entering a decisive phase. The initial build-out of the rails is well underway, but the next leg will be defined by a shift from pure hardware to software and scale. The catalysts and risks now center on monetization, competitive advantage, and the sustainability of the global supply chain.

The most immediate catalyst is a clear pivot in the investment thesis. As Goldman Sachs Research notes, the next phases of the AI trade are expected to involve AI platform stocks and productivity beneficiaries. This is a fundamental shift. The market is moving beyond rewarding companies for spending on infrastructure and is now focusing on those that can successfully integrate AI into their core business models to drive revenue and profit. The year 2026 will separate long-term winners from those who have not successfully integrated AI-native business models, as the focus shifts toward scale, monetization and competitive advantage.

A major risk to the exponential trajectory is the rapid expansion of semiconductor production capacity, particularly in China. As the global semiconductor equipment market sets new highs, China's rapid expansion of production capacity is emerging as a potential disruptor of the medium- to long-term supply structure. This aggressive build-out, driven by national policy and massive investment, could eventually lead to oversupply and pricing pressure, challenging the current supercycle's durability. The tension between sustained AI demand and a flood of new capacity is the key variable to watch for the next two years.

For investors, the critical variables are now clear. The catalyst is the monetization of AI investments, which will favor platform companies and enterprises that embed AI deeply into their products. The primary risk is a supply glut from aggressive capacity expansion, which could compress margins and disrupt the pricing power that has fueled the sector's rally. The exponential growth curve remains intact, but its slope will be determined by how well companies convert massive capex into sustainable earnings and how the global supply chain evolves.