- The Philadelphia Semiconductor Index has surged 78% through Tuesday while the iShares Expanded Tech-Software Sector ETF is down 0.4% over the same period, a gap of more than 78 percentage points within what is commonly described as a single AI trade.
- Sitara Sundar, head of alternative investment strategies at JPMorgan private bank, argues two cycles are running at once. The finance and infrastructure cycle is in its mid-innings as hyperscalers shift from funding capital spending out of cash flow to issuing in capital markets, while AI integration into the wider economy remains early with productivity gains only beginning to reach businesses.
- She describes the volatility in technology as healthy and normal, and something that was anticipated coming into this year. Her positioning favours select hyperscalers, selectivity within semiconductors, and venture capital and private equity-owned businesses positioned to benefit from AI-driven productivity over a three-year horizon.
- On enterprise AI, where Meta launched a new platform this week, she expects hyperscalers to prevail in broad consumer-facing uses while specialist operators succeed in niches such as finance and law. On government bond yields at multiyear highs, she argues the neutral rate for the AI buildout differs fundamentally from that of the broader economy because demand for compute is unaffected.
What Happened?
Sundar was speaking in a Bloomberg Television interview. The story was produced with the assistance of Bloomberg Automation.
Why It Matters?
The 78-point spread is the most useful fact here and it complicates the consensus rotation. Semiconductors have captured essentially all of this year AI performance while software, the sector supposedly positioned to benefit from AI productivity, has delivered nothing. The market has been voting that value accrues to the compute layer rather than the application layer. That sits awkwardly beside the repositioning now underway at large allocators, with BlackRock reshuffling models to spread AI exposure beyond the obvious pioneers and Wells Fargo Investment Institute cutting technology while upgrading industrials. Those managers are moving toward beneficiaries precisely because beneficiaries have not performed, which is either early or wrong, and the evidence so far has not distinguished between the two. Sundar own argument contains a tension worth naming. She notes that hyperscalers are now issuing in capital markets rather than relying on internal cash flow, then argues the neutral rate for the AI buildout is fundamentally different from the broader economy because compute demand is unaffected by yields. Both cannot comfortably hold. A company funding capital expenditure through debt is rate-sensitive by construction, and with the 30-year above 5.5% and the 10-year above 5.2% the cost of that issuance has risen materially. Demand for compute being robust does not make the financing of it cheap. Her dual-track framing is nonetheless a reasonable way for wealth clients to hold the position, and her three-year horizon for productivity beneficiaries is honest about how long that leg takes to pay.
What Next?
Watch whether the software and semiconductor gap begins to close, since that is the only evidence that would validate the rotation toward AI beneficiaries now underway across several large allocators. Hyperscaler debt issuance is the specific item to track given Sundar observation about the funding shift, and the spreads at which that paper prices will show whether the AI neutral rate argument holds. Third quarter results are the near-term test of whether productivity gains are reaching business income statements or remain prospective. For enterprise AI, Meta new platform and the specialist operators in finance and law will produce the first real evidence on whether her expectation of coexistence is right. Private market exposure over a three-year horizon is the hardest leg to evaluate, and investors should ask what marks those positions carry today.
Affected Tickers and Coins: META, IGV, NVDA, MSFT
Source: Bloomberg
