Goldman Sachs warns that investment linked to artificial intelligence could continue growing well beyond current market expectations. The bank’s analysts say the largest cloud providers and technology companies — the hyperscalers — may significantly raise capital spending tied to AI by 2027.
Concrete figures and forecasts
- Goldman Sachs’s base case estimates hyperscaler investments at about $1.1 trillion (1100 billion) by 2027, above the market consensus of $920 billion.
- In an optimistic scenario, the bank says spending could reach $1.4 trillion (1400 billion).
- The bank also estimates that the number of tokens processed by AI models could increase 24-fold by 2030.
Goldman attributes the rapid rise in compute demand largely to the spread of enterprise AI agents and growing model usage.
Why infrastructure demand is rising
Higher token consumption drives simultaneous needs for more data-center capacity, semiconductors, networking equipment and power infrastructure. Rising hardware and construction costs also push up investment requirements.
Goldman points to rapidly expanding order backlogs at major cloud providers: the combined contracted backlog for Google Cloud and Amazon Web Services rose to $832 billion at the end of the first quarter of this year, up from $358 billion six months earlier — a jump the bank views as supporting its optimistic outlook.
Market balance and timing
Analysts expect supply-demand balance for AI infrastructure to form only around the second half of 2027. That implies the investment wave could sustain elevated demand across the AI ecosystem for an extended period.
Goldman draws historical parallels: it estimates AI-related investment could equal about 1.5% of U.S. GDP in 2026. By comparison, past major technology-era investment peaks — such as railway construction, electrification or the rise of the auto industry — reached 2–3% of GDP, suggesting the AI cycle may still have room to grow.
Risks and constraints
The bank identifies several bottlenecks that may hamper capacity expansion even if funding is available:
- delays in data-center projects;
- shortages in memory chips;
- constraints in power supply;
- lack of skilled labor.
These issues could slow the rollout of necessary infrastructure.
Winners and valuation risks
Goldman’s analysts see beneficiaries in the semiconductor, networking, cooling and energy sectors as investment rises. At the same time, they warn that many stocks tied to AI infrastructure already trade at elevated valuations, increasing volatility and the risk of correction.
Limited current evidence of productivity gains
The bank also notes limited proof so far that AI has broadly produced sizeable productivity improvements. According to Goldman’s review of Q1 earnings: 54% of companies mentioned AI, but only 11% quantified productivity benefits and just 2% showed concrete earnings impacts. That suggests firms are currently spending on AI faster than the financial benefits are appearing in results.
This article does not constitute investment advice or a recommendation.
