Energy-supply issues broaden the discussion beyond valuations. According to the International Energy Agency, global electricity demand from data centers, crypto mining and AI could double between 2022 and 2026, raising concerns about grid reliability in North America and Europe. Executives at GE Vernova have echoed those concerns, noting that expansion plans for large cloud operators increasingly hinge on local utilities’ ability to deliver incremental megawatts.
Alphabet viewed as the most adaptable hyperscaler
Among the major cloud providers, Alphabet is considered the best positioned to navigate slower power deployments. Its Gemini 3 large-language model has demonstrated energy-efficient training characteristics, and press reports indicate Alphabet is in advanced talks to supply on-device generative AI services for roughly 1.5 billion Apple devices. Analysts argue that such an arrangement would diversify Alphabet’s revenue mix and provide a buffer against data-center build slowdowns.
Amazon and Microsoft continue to invest in custom silicon to manage energy use, yet their forthcoming chip generations face the same grid limitations affecting the broader sector. For now, all hyperscalers are concentrating new capacity in regions with favorable power-purchase agreements and expedited permitting, especially parts of the U.S. Midwest and the Nordic countries.
OpenAI readies an outsized initial public offering
Privately held OpenAI, backed by Microsoft, SoftBank and Nvidia, is expected to pursue an initial public offering large enough to secure automatic inclusion in the S&P 500. Portfolio managers note that passive index funds would have to raise cash to purchase the new constituent, potentially shifting capital away from other technology names. Nevertheless, OpenAI’s future data-center footprint will be subject to the same power-supply restrictions facing its publicly listed peers.
Nvidia remains central to both AI and accelerated computing
The centrality of Nvidia in advanced computing stems from its combined hardware-and-software stack rather than graphics-processing units (GPUs) alone. The company’s CUDA software layer is widely viewed as a competitive moat that competing chip vendors have yet to replicate. Nvidia’s forthcoming “Vera Rubin” processor is designed to reduce inference errors—commonly described as AI “hallucinations”—and to enable reasoning tasks that current architectures handle inefficiently. A follow-on product, code-named “Richard Feynman,” is in development as part of a multiyear roadmap aimed at replacing conventional CPU-centric systems with GPU-centric clusters.
Imagem: Internet
Analysts calculate that the accelerated-computing market could reach multi-trillion-dollar levels once legacy enterprise servers are refreshed, independent of generative-AI workloads. That outlook underpins long-term “own it, don’t trade it” recommendations on Nvidia despite sector-wide multiple compression during 2025.
Geopolitical and technology risks persist
The primary geopolitical risk highlighted by strategists is Taiwan’s exposure to potential Chinese military action. Taiwan Semiconductor Manufacturing Co. remains the dominant contract manufacturer for leading-edge chips, including Nvidia’s high-performance GPU line. A disruption to Taiwanese fabrication facilities would affect supply chains across the semiconductor industry.
On the technology front, nuclear power, quantum computing and speculative data-center extensions that rallied in early 2025 have pulled back as investors reassessed commercialization timelines. While small-modular reactors receive periodic policy support, energy analysts view nuclear as too slow to alleviate the near-term power bottleneck confronting data-center operators.
Portfolio considerations for 2026
Institutional desks advising clients on 2026 positioning generally recommend maintaining exposure to four segments: hyperscale cloud operators, electrical-equipment suppliers, network-infrastructure providers and foundational-model developers. Names most frequently cited include Alphabet, Amazon, Microsoft, Nvidia, Broadcom, Eaton and GE Vernova. Analysts argue that these firms exhibit the scale, capital resources and product pipelines necessary to function in a power-constrained environment.
Outside the core AI complex, strategists point to opportunities in pharmaceuticals, consumer brands and financial services, but emphasize that long-duration technology assets continue to represent the market’s most dynamic growth segment. For investors weighing valuation risk against structural demand, the prevailing view is that limited electricity supply sets a cap on capacity growth but does not negate the need for advanced AI hardware and software. Consequently, disciplined exposure to AI-linked equities is expected to remain a portfolio imperative through at least 2026.
Crédito da imagem: CNBC
