NEWYORK
Goldman Sachs macro traders Cosimo Codacci-Pisanelli and Rikin Shah have recast the artificial intelligence investment narrative as a shift from a "software cycle" to a "capital expenditure cycle" — a structural change that will suppress bond market rebounds across the curve's long end.
The distinction cuts to the heart of rate market mechanics. Software cycles required minimal capital investment and lowered equilibrium rates. The current AI infrastructure surge operates in reverse: massive capex demand intensifies competition for a finite pool of duration buyers, pushing yields higher. AI infrastructure construction now mirrors the physical scale of railway development, not the capital-light software expansion of the internet era.
Goldman Sachs projects $1 trillion in hyperscaler capital expenditure is coming, noting that Wall Street's 2027 consensus forecasts severely underestimate the true scale. The firm advises bond investors to short curve rallies on the long end rather than chase rebounds, betting that fiscal supply pressure — now bolstered by a "second leg" of private sector capex financing demand — will keep longer-dated yields compressed.
Both public and private entities are contending for scarce duration. Credit issuance is surging to fund these capital demands, and equity financing channels have opened as well. Current forecasts for data center capex are under continuous upward revision.
Goldman Sachs acknowledges that AI's "destination" — lower inflation, reduced computing costs, higher corporate margins, a lower neutral rate — may ultimately prove correct. But the firm argues the "journey" costs have been seriously underestimated. Physical bottlenecks in electricity, power grids, skilled construction workers and cooling systems could obstruct inflation normalization. The firm places the burden on market bulls to demonstrate "whether productivity gains can be realized and how long it takes."
Softer U.S. nonfarm payroll data in June tempered July rate hike expectations, but Goldman Sachs views a hike as "optional, not mandatory." Inflation data will remain the decisive factor for the Federal Reserve.
The equity market has already begun pricing this dynamic. Goldman Sachs observes a rotation from the "spending side" of AI capex to the "beneficiary side" — cloud providers, chipmakers and specialized software vendors positioned to capture long-duration cash flows from hyperscaler buildout.
Anshul Sehgal, Global Co-Head of FICC at Scotiabank, noted that AI hyperscalers now account for over 25 percent of U.S. market investment demand, concentrating purchasing power among a narrow set of vendors. Government-backed initiatives in the United States and European Union are channeling capital toward alternative refining capacity for critical metals, creating investable pathways for non-Chinese producers. Sovereign net-zero commitments and corporate power-purchase agreements generate long-duration contracted cash flows for utilities and grid specialists.
