It’s difficult to talk about the macroeconomic landscape these days without the conversation eventually turning toward how artificial intelligence (AI) could influence—perhaps even reshape—the current paradigm. While there’s clearly a wide range of possible outcomes, our outlook explores AI’s likely impact on the global economy over the next few years, longer-term macro considerations, and some potential obstacles faced by the industry. 

Short term: capex will be a major driver of growth

Fueled by expectations of significant productivity gains across industries and rapid user adoption, the boom in large language models (LLMs) since 2022 has sparked a robust capital expenditures (capex) cycle in the United States. This phenomenon has accelerated to the point where we estimate that investment in processing equipment and software—a proxy for overall AI spending—accounted for nearly a quarter of the nation’s annualized GDP growth in the first three quarters of 2025 (after adjusting for the large share of semiconductors that are imported). We think AI spending mostly explains the surprising resilience of the U.S. economy this year, despite heightened policy uncertainty and tariffs, and we expect it to continue for the next few years.

Over the next two years, AI-related investment looks set to proceed at warp speed. As of this writing, analysts forecast that the five largest U.S. cloud computing hyperscalers will spend over US$1 trillion on AI capex in 2026 and 2027. These investments are likely to include datacenters, graphics processing units (GPUs), dynamic random access memory (DRAM), and storage, in addition to microprocessor production capacity and actual AI software development (both in the U.S. and abroad). The $1 trillion figure could actually prove to be a conservative estimate, with new deals being announced almost daily and global players increasingly joining the AI race, especially in China.

With this explosive growth comes rising concerns around overbuilding and creating excess capacity. However, we think four important factors mitigate this risk and support the AI infrastructure buildout, which should, in turn, continue to support global economic growth over the shorter term:

1. The rate of AI adoption is quite rapid. Both consumers and businesses are rapidly adopting AI technologies, with demand for new computing models increasing by a factor of about 4.4x per year according to EpochAI. The faster AI becomes entrenched as a business tool, the sooner the technologies will begin to generate meaningful revenue for corporations. As a path to enhanced profitability emerges, the higher the likelihood that the AI infrastructure buildout will be sustainable.  
2. Spending is coming from profitable companies. Unlike debt-fueled bubbles of the past, today’s capex has so far been largely funded through companies’ retained earnings and free cash flows, helping to limit the risk of a deleveraging spiral spilling over to the broader economy. This trend could be starting to change, which is of little surprise given the sheer scale of the capital needed to sustain AI-related expansion in the coming years. Key industry players have recently announced debt financing for their upcoming infrastructure builds.
3. Fiscal policy tailwinds may further boost AI spending. Fiscal policy in many countries could also serve as a strong tailwind over the next 12 months. Case in point: the expensing of capex in the United States under the One Big Beautiful Bill Act. Moreover, tariff exemptions for AI-related imports are saving firms billions of dollars, improving the economics of investing in technology. 
4. Monetary policy may provide another form of support. For example, we believe the U.S. Federal Reserve is likely to continue cutting interest rates into 2026, which should bolster U.S. capital investments in AI as financing costs drop off. Historically, several market corrections have coincided with rising rates, which tend to amplify stress on capital-heavy businesses.

Long term: there’s clear upside, but questions remain

Over the longer term, the economic benefits of these massive investments hinge on future productivity gains made possible by AI adoption. On that front, estimates vary greatly. Depending on assumptions about adoption rates, cost savings, and potential revenue generation, a Cognizant impact study conducted in partnership with Oxford Economics projects that activity in the AI space could lift U.S. GDP by between as little as 0.7% and as much as 6.0% over the next decade. Even just a median baseline impact of 2.5% would be materially additive to the long-term economic outlook.

AI’s potential impact on inflation is interesting. The primary near-term pressure point is energy. Indeed, data centers’ seemingly insatiable demand for energy strains power grids and drives electricity prices higher. In that context, data center energy efficiency (e.g., a colder climate to reduce cooling costs) and access to inexpensive power sources could become more important cost considerations. However, broad enhancements to productivity could prove deflationary.

Taken together, the net long-term effects are likely to be slightly disinflationary, as the far-reaching benefits of innovation should outweigh issues surrounding energy scarcity. At the margin, lower inflation would favor slightly easier monetary policy, which would flow through to the rest of the yield curve. That said, the recent shift toward financing AI-related capex with debt could add to overall bond issuance and global supply, potentially increasing the cost of capital in certain areas.  

The impact of AI on the labor market is also uncertain. There are currently plenty of anecdotes in circulation around how AI is rewriting job requirements in the technology space, but at the moment, evidence of broad-based AI-driven job losses is inconclusive. Over a longer period, however, we would expect a three-tiered effect on the labor market: 1) certain types of jobs will ultimately be replaced by AI technologies; 2) other jobs are likely to use AI to enhance productivity; 3) professions that don’t exist now could emerge as a result of evolving technology.

There are obstacles to overcome

Finally, looking ahead, the AI industry is not without potential risks and headwinds. To fully realize its long-term potential, we think the industry will have to overcome certain obstacles, such as:

• Energy bottlenecks. Power requirements for new AI models have doubled every year, and data centers globally already consume as much power as France or the United Kingdom. By 2030, usage could reach what Japan is using today. Energy is becoming a barrier to completing projects in a timely manner, as connection delays and transformer shortages are mounting. In response, many tech companies have started stockpiling their own energy supplies. Should this trend persist, energy infrastructure developments could guarantee ongoing power to the AI space without hurting other segments of the economy.
• Monetization. At the moment, LLMs’ subscription fees cover only a fraction of AI’s operating and capital expenditures. Over time, we think major players will require a positive return on their investments, which could rein in AI spending. Also, LLMs’ market structure doesn’t benefit from network effects the same way social media did, making it theoretically harder to extract subscriptions and add revenue streams. We think monetization implies expanding AI applications beyond LLMs to sectors like robotics and healthcare.
• Swift depreciation: GPUs and network equipment have relatively short lifecycles. Tech companies generally anticipate a hardware life of only around five years due to rapid technical innovation and high utilization rates. The challenge will be to implement a sustainably profitable business model, given the very high costs of maintaining cutting-edge datacenter capabilities.