Digital Infrastructure: Momentum in 2025, Validation in 2026

By | December 15, 2025

With 2025 drawing to a close, we (Xona Partners) recap key highlights in digital infrastructure across data centers and telecom assets, including wireless, fiber, satellite, and submarine systems, while looking ahead to developments in 2026 (download below). We titled the preface Digital Infrastructure: Momentum in 2025, Validation in 2026 to reflect our view that 2026 will be a pivotal year for AI infrastructure buildouts. In recent months, concerns have grown that AI represents a bubble—primarily in the valuation of AI‑related assets—contributing to heightened market volatility. This does not diminish the transformative nature of AI, which makes it an inflection‑type bubble: regardless of valuation, it is a technology that will clearly mark the distinction between what came before and what follows. Investors are increasingly questioning monetization prospects and the prevailing approach to AI infrastructure, which relies on scale, with more compute power pursued as the path to better performance. As a result, 2026 will test and validate many of the assumptions currently guiding the market.

Download the full version here:

K-Shaped outlook

We highlight the Kshaped nature of digital infrastructure, where the hot AI data center market contrasts with a stagnating telecom sector, particularly in access services such as wireless and fiber. Consolidation pressures remain intense, with scale increasingly necessary for survival. This is not unexpected, as it reflects the continuation of a trend that began more than a year ago. In July 2024, we cautioned against a protracted downturn in incumbent access services (see Mobile Infrastructure Capex: Permanent Weakening or Short-Term Decline?). As a result, 2026 may resemble 2025, with service providers, vendors, and ecosystem players focused on operational and financial performance, pursuing divestments, and trimming product lines. Against this backdrop, several key takeaways emerge that frame the outlook for 2026.

Key Takeaways

  • AI-driven infrastructure spend is reshaping the market with a wave of data center buildouts, placing data center interconnect at both inter and intra facility levels in sharp focus as a premium asset class, with U.S. demand forecast to double backbone capacity by 2029 through 92,000 new route miles supported by hyperscaler contracts and 30-40% annual traffic growth, while subsea cable investment surges with $13 billion committed between 2025 – 2027 through projects such as Meta’s 50,000 kilometer Project Waterworth and new builds by Google, Amazon, and Microsoft to diversify global routes.
  • Telecom consolidation is accelerating as capex spending, which weakened in 2024, remains muted and is projected to extend further into 2026, compounded by weak 5G monetization. Landmark deals on the service provider side such as Charter and Cox’s $34.5 billion merger, AT&T’s $5.75 billion acquisition of Lumen fiber assets, and EchoStar’s $43 billion spectrum divestitures highlight survival through scale as the dominant strategy. At the same time vendors are scaling down, exiting non-core businesses, and reorganizing to align with AI infrastructure demand.
  • Capex intensity in AI data centers is unprecedented, with gigawatt-scale campuses announced worldwide, in some cases backed by sovereign AI initiatives that treat compute as strategic infrastructure. Yet the sector faces mounting risks of overbuild, fragile neo-cloud business models, and above all energy constraints that have become the defining bottleneck, driving demand for powered land strategies, alternative energy solutions, and heightened regulatory scrutiny across major regions.
  • Valuation risks remain elevated, with record fundraising concentrated in mega funds and annual AI infrastructure spending exceeding $300 billion in 2025. Investor caution around AI economics is rising, driven by short GPU depreciation cycles, uncertain workload maturation, and limited RoI from generative AI deployments. Given the multilayered nature of the AI stack, asset value disinflation is uneven across segments, reinforcing the importance of disciplined evaluation of both technical fundamentals and commercial viability.
  • LEO satellites are emerging as an early convergence point between terrestrial and space infrastructure, while traditional mobile and fixed access providers struggle with monetization and GEO operators face shrinking revenues and consolidation pressures, exemplified by SES and Intelsat. This dynamic is heightening investor interest in Ground Station as a Service (GSaaS) models that resemble tower and data center investments. At the same time, the satellite sector is undergoing rapid transformation as Amazon’s Kuiper prepares for commercial rollout to challenge Starlink’s five year lead, and sovereign programs increasingly treat satellites as strategic assets, positioning national constellations as critical to digital infrastructure resilience.

Concluding Thoughts

The cloud computing infrastructure ecosystem, including hyperscalers and social media giants like Google, AWS, Microsoft and Meta, benefited enormously from the overbuilt fiber capacity installed during the dotcom boom. These firms and others like them did not have to invest in connectivity infrastructure during their formative years, which boosted earnings and return on investment. In the dotcom era, the price of connectivity collapsed because supply was abundant and demand lagged.

Today, these same companies must spend heavily to deploy AI data centers, committing tens of billions to infrastructure that includes GPUs and compute engines with depreciation cycles of about 2 – 3 years. Many have extended amortization to more than 5 years, which lifts earnings and prolongs the cycle. The concern is that if large compute capacity becomes widely available while demand remains below supply, the market could experience a repeat of the early 2000s. This is not an unrealistic scenario. Scale may not always produce better AI, as argued by Gary Marcus. Adoption continues to lag, as shown in the widely cited MIT study, even as more compute is deployed. Innovation in compute technology may also unlock more power and further shift the balance between supply and demand.