The “AI trade” has long been discussed as an equity story: investors buying into the companies building models, supplying chips, powering data centres and capturing the next wave of productivity. But the market’s attention is shifting. Increasingly, the same underlying thesis is showing up in fixed income—quietly at first, then with enough scale that it starts to behave like a cross-asset bet rather than a collection of independent exposures.
That matters because bonds are often treated as the stabiliser in a portfolio. They’re supposed to diversify equities, dampen volatility and provide carry. Yet when bond markets become dominated by the same narrative driving stocks, the diversification benefit can erode. The risk isn’t simply that investors own “AI” assets; it’s that they may be owning the same economic outcome through different instruments, with similar timing, similar sensitivities and similar behavioural triggers.
In other words: concentration risk is not confined to equities. It can migrate into credit spreads, duration positioning, sovereign issuance demand and even the way investors interpret macro data. When that happens, the portfolio can look diversified on paper while becoming tightly linked in practice.
To understand how this plays out, it helps to separate two ideas that are often conflated. One is thematic investing—allocating capital to sectors or companies associated with AI. The other is thesis-driven correlation—where multiple asset classes respond to the same set of expectations about growth, inflation, productivity, fiscal policy and risk appetite.
Equities tend to reflect thesis changes quickly because prices adjust continuously and sentiment can move faster than fundamentals. Bonds adjust differently. They price expected cash flows and discount rates, and they incorporate risk premia that can lag. But once the market begins to treat a theme as a macro regime—rather than a single sector story—fixed income can start to move in step with equities. That’s when the “AI trade” stops being a narrow allocation and becomes a broader market factor.
One reason this migration is happening is that AI has become entangled with the biggest drivers of bond returns: real yields, inflation expectations, term premia and credit risk. AI-related investment cycles are capital intensive. They require power generation, grid upgrades, fibre networks, semiconductor fabrication, data centre construction and logistics. Those investments can influence inflation dynamics through demand for construction materials, labour and industrial inputs. They can also influence growth expectations by raising productivity potential—at least in theory—and by changing the trajectory of corporate earnings.
Even if the direct beneficiaries are concentrated in a handful of companies, the macro effects can be diffuse. If investors believe AI will accelerate productivity, they may expect stronger long-run growth and potentially different inflation outcomes. If they believe AI will increase energy demand and supply constraints, they may expect inflation persistence. Either way, the bond market is forced to price a new set of assumptions. And when those assumptions become widely shared, the bond market can start to reflect a consensus view that is tightly correlated with equity sentiment.
There’s also a more mechanical channel: the way capital flows into fixed income products. Over the past decade, the bond market has become increasingly shaped by systematic strategies, index-linked allocations and benchmark-aware trading. When a theme becomes popular, it doesn’t just attract discretionary stock pickers. It attracts factor investors, quant overlays, risk parity frameworks and liability-driven investors who rebalance based on yield targets, duration bands and spread levels.
If the same investors are buying the same types of credit—say, high-quality issuers with strong balance sheets, or specific segments of tech-adjacent corporate debt—the result can be a form of “credit crowding.” The bonds may not be issued by the same companies as the stocks, but they can still be exposed to the same underlying economic thesis: demand for AI infrastructure, the durability of cash flows from technology-led capex cycles, and the market’s willingness to underwrite growth.
This is where concentration risk becomes subtle. In equities, concentration is obvious: a handful of names can dominate index performance. In bonds, concentration can hide in plain sight. A portfolio might hold a diversified basket of issuers across sectors, but if those issuers share similar leverage profiles, similar refinancing needs, similar sensitivity to interest rates and similar exposure to the same end-market demand, the portfolio’s effective risk can be far less diversified than it appears.
Consider credit spreads. Spreads are often treated as a measure of idiosyncratic default risk plus a general risk premium. But in practice, spreads can also reflect the market’s confidence in a particular growth narrative. If investors believe AI-driven capex will translate into stable revenue streams and resilient margins, they may compress spreads across a broad swath of corporate credit. That compression can occur even in parts of the credit market that aren’t directly tied to AI, because the market is pricing a lower probability of recession and a higher probability of continued earnings resilience.
Now imagine the opposite scenario. If the AI narrative stumbles—whether due to regulatory friction, slower-than-expected monetisation, a capex slowdown, or a shock to financing conditions—spreads can widen quickly. The widening may not be limited to the most “AI-exposed” issuers. It can spill into the broader credit complex because the market’s risk premium reprices the same macro assumptions that were previously used to justify spread compression.
This is the core correlation risk: when multiple asset classes are driven by the same thesis, they can sell off together even if their instruments differ. Equities may fall because earnings expectations are revised. Bonds may fall because discount rates rise, liquidity tightens, and credit risk premia widen. The portfolio can experience a double hit: lower equity valuations and weaker bond prices, with both moves reflecting the same change in narrative.
Duration adds another layer. AI optimism can influence the shape of the yield curve. If investors expect stronger growth, they may push up longer-term yields. If they expect productivity gains that reduce inflation pressure, they may pull down certain segments of the curve. The net effect depends on the market’s interpretation of the AI regime. But once the market converges on a view, duration positioning can become crowded. Investors who are effectively betting on the same macro path can end up with similar duration exposures, even if they think they are diversifying across sectors.
There’s also the question of liquidity and market structure. Bond markets are not immune to the same dynamics that have made equity markets prone to fast repricing. In periods of stress, liquidity can evaporate, bid-ask spreads widen and correlations rise. If many investors are positioned for the same outcome—whether that outcome is “soft landing with AI productivity” or “inflationary capex cycle”—then the unwind can be abrupt. The market doesn’t need everyone to be wrong; it only needs enough participants to trade in the same direction at the same time.
This is why the “AI trade” in fixed income can be more dangerous than it looks. It’s not just about holding bonds that are somehow related to AI. It’s about holding bonds that are priced using the same assumptions that underpin equity valuations. When those assumptions shift, the correlation can jump.
A unique feature of AI as a theme is that it sits at the intersection of technology, energy, regulation and geopolitics. That intersection can create non-linear risks. For example, AI compute requires electricity and cooling. If energy supply becomes constrained or if policy changes increase the cost of power, the economics of AI infrastructure could deteriorate. That would affect corporate cash flows and could also influence inflation expectations. Both channels feed into bond yields and credit spreads.
Similarly, AI is subject to regulatory scrutiny around data, privacy, safety and competition. Regulatory outcomes can change the expected profitability of AI-related business models. Equity markets may react immediately, but bond markets can also reprice quickly once investors revise the probability distribution of future cash flows. Credit markets are particularly sensitive to changes in perceived survivability and refinancing risk. If investors begin to doubt the durability of certain revenue streams, spreads can widen even for issuers that appear fundamentally sound.
Geopolitics adds another dimension. Supply chain constraints for semiconductors and equipment can affect capex timelines. Sanctions and export controls can alter the competitive landscape. These factors can influence both growth expectations and risk premia. Again, the bond market may not be “buying AI” directly, but it is pricing the macro and corporate consequences of the AI ecosystem.
So what does “concentration risk” look like in practice for investors? It often shows up in three ways.
First, risk models can underestimate shared exposures. Many portfolio risk frameworks rely on factor decomposition—duration, credit quality, equity beta, volatility regimes. But thematic concentration can create hidden common factors. Two portfolios might both be “investment grade diversified,” yet both could be heavily exposed to the same macro narrative that drives spreads and yields. When the narrative shifts, both portfolios can move together, even if their reported factor exposures differ.
Second, the market’s consensus can become self-reinforcing. When investors pile into the same thesis, they can push valuations to levels that assume favourable outcomes. In equities, this is visible in multiples. In bonds, it can be visible in spread compression and in the willingness to accept lower compensation for risk. If the market becomes comfortable with low spreads because the thesis is popular, the system becomes fragile. A small negative surprise can trigger a larger repricing because there is less “room” in valuations for uncertainty.
Third, the unwind can be correlated. In stress, investors often reduce risk simultaneously. They sell what is liquid first, cut exposures that are easiest to exit and rebalance toward benchmarks or liquidity targets. If many investors hold similar bond exposures—similar sectors, similar maturities, similar credit qualities—then the selling pressure can be concentrated. That can amplify drawdowns and make recovery slower.
This is why the issue is not merely academic. Concentration risk in fixed income can translate into real portfolio losses during regime shifts. It can also distort performance attribution. An investor might attribute underperformance to “rates” or “credit” without recognising that the underlying driver was a thematic repricing that affected both. The