Sep 3, 2026 · 51 min
AI borrowing tests credit investors’ discipline
Columbia Threadneedle Scans AI Debt Rush for Alpha Opportunity
The episode examines whether unprecedented AI financing will create durable credit opportunities or expose investors to concentrated risk as spreads remain tight.
- 1Tight spreads and strong fundamentals leave investors with limited compensation for credit risk.
- 2AI borrowing spans public bonds, private financing, and hyperscaler debt, creating both capacity concerns and potential alpha.
- 3Volatility, bottom-up research, and selective positioning may matter more than relying on uncertain macro forecasts.
Don't miss
Tom Murphy frames AI credit as a choice between financing a transformative industry and funding a historic malinvestment, with volatility potentially creating alpha.
The brief
Tom Murphy of Columbia Threadneedle Investments joins Bloomberg Intelligence’s David Havens to assess tight spreads, heavy issuance, and the financing demands of AI infrastructure.
Murphy’s framework combines fundamental research, technical conditions, and valuation discipline because strong corporate balance sheets do not fully offset historically thin compensation for credit risk.
AI borrowing is spreading across hyperscaler bonds, private data-center financing, high-yield, structured credit, and non-dollar markets, raising questions about capacity, concentration, and liquidity.
The private investment-grade market is active but less standardized and less deep than public bonds, forcing investors to manage duration, mark-to-market volatility, and issuer concentration.
Murphy sees temporary spread widening as a chance to add exposure rather than sell, while highlighting bank credit as a sector strengthened by capital, funding, transparency, and regulation.
The closing tension is stark: AI credit may finance a transformative industry or a historic malinvestment, making volatility a potential source of alpha for nimble investors.
What was said on this episode
20 statements · 11 positive · 3 negative · 4 mixed · 2 neutral
Avoiding declining hyperscaler debt improves portfolio performance.
“there are definitely things that you can sidestep. You know, not owning it if it goes down, that's additive to portfolio performance.”
Listen at 15:04
Hyperscalers currently have relatively immature balance sheets.
“the companies have had relatively immature balance sheets, these hyperscalers.”
Listen at 15:12
Hyperscaler 2027 capital expenditure expectations range from $1 trillion to $1.4 trillion.
“it's anywhere from a trillion to $1.4 trillion for these CapEx, for these hyperscalers.”
Listen at 17:09
AI-related companies will produce both winners and losers.
“There's definitely going to be winners and losers”
Listen at 19:04
Diversified AI companies could resume strong cash flow and repay debt if AI disappears.
“if AI goes away tomorrow, if some of these companies just pivot back to their, I'm not going to say original businesses, but their businesses that were kind of the drivers before AI came along, they would just go back to generating tremendous amounts of free cash flow and they could pay down the debt that they're taking on.”
Listen at 19:22
Continued AI growth will eventually convert hyperscaler capital expenditure into cash flow.
“if AI continues to grow the way that these companies seem to have conviction that it does, Eventually, that permission structure is going to be there where this capex is going to turn into cash flow.”
Listen at 19:38
Investors will become more comfortable once AI spending reaches cash-flow conversion.
“I think that people will be more comfortable.”
Listen at 20:15
The hyperscaler debt market is experiencing both widening and tightening spreads.
“I do feel like we're in a marketplace, and I have to give one of my colleagues credit for this, of wider wides and wider tights.”
Listen at 22:20
Spread volatility will persist until AI investments’ cash-flow timing becomes clearer.
“I think we're there until we get to that hump of the S-curve where we can kind of have better clarity as to when our investments turn into cash flow.”
Listen at 22:26
Hyperscaler data-center exposure may not equal equivalent underlying-company debt exposure.
“if I'm buying a hyperscaler data center that has company X as the lessee, is that worth just as much as having a dollar's worth of the debt of that underlying company?”
Listen at 25:59
The AI financing market has not yet reached its capacity limits.
“I don't think we're there yet.”
Listen at 26:20
Yields of 6%–7% will attract additional investors to AI-related debt.
“if we're talking 6%, 7% yields on these things, we will find other investors.”
Listen at 26:56
Columbia Threadneedle prefers AI companies with diversified business mixes.
“we like the ones that have the broader business mixes”
Listen at 27:46
AI pure plays and related companies will have divergent outcomes.
“there's going to be winners and losers.”
Listen at 28:06
Strong OpenAI and Anthropic credit metrics could reduce other hyperscalers’ issuance needs.
“if OpenAI and Anthropic actually come out with pretty strong credit metrics, their need to lean on some of these other companies would be less, and therefore issuance needs from some of the other hyperscalers could be less.”
Listen at 28:52
Strong pure-play AI credit quality could benefit other AI companies’ financing conditions.
“it could actually end up being a positive.”
Listen at 29:04
Columbia Threadneedle has avoided chip-backed financing deals because collateral valuation is difficult.
“we haven't invested in any of the chip deals, James, for exactly the reason you talked about.”
Listen at 29:49
Some AI chips remain functional after five or six years.
“these management teams have said, look, we've got chips that are five and six years old that are still functioning.”
Listen at 30:01
A chip removed from a data center does not necessarily become obsolete.
“it doesn't mean that that chip becomes obsolete.”
Listen at 30:15
AI chips have longer useful lives than the market previously expected.
“the lives of those are longer than they, you know, than the market probably previously thought.”
Listen at 30:37
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.