Investment Views from Our CIOs
Key Takeaways
What’s on the Minds of Our CIOs?
The central unresolved question of the AI cycle has not changed: when will monetization and measurable productivity arrive to justify the capex? We cannot answer that definitively in one quarter, but we can break it into tractable pieces. This quarter we took two: the competition between AI models, and the debt funding the build-out. We emerged with maintained conviction in the AI infrastructure opportunity across both equities and the select debt that finances it.
Top Investment Questions for the Road Ahead
Does AI model competition derail the AI infrastructure opportunity?
"AI model competition does not change our thesis that owning AI infrastructure is a durable investment opportunity. But it does shift which companies win and lose. Ultimately, we expect a hybrid world where both models exist for different use cases."
Sung Cho, Portfolio Manager, Fundamental Equity
In Focus: AI Model Competition and a Hybrid World
At its core, this is a contest between two regions and two approaches. Our technology analysts do not expect the model debate to resolve soon, but see a case for coexistence.
- Open-weight models are freely available, adaptable, lower-cost models that handle high-volume workloads such as customer service and content creation. China is currently leading the effort in this race.
- Closed-weight models are proprietary, frontier-capability models that retain high-value tasks such as drug discovery and scientific research. The US leads in this contest.
What We See: Three Nuances
- Cost comparisons are rarely apples-to-apples. Chinese models may be cheaper to run on a per-prompt basis, but if they must run several times over to match the output of a single US model, that per-task cost advantage narrows considerably.
- The binding constraint differs by region. For China, we see the limiting factor as compute—capping its access to the AI frontier and motivating efficiency in algorithms and inference. For the US, we believe the ultimate constraint is power availability, driving investment in on-site generation to bypass grid-connection queues.
- China is building out its own semiconductor supply chain. Already advanced in chips for non-AI applications (EVs, smartphones), it is pushing further into the AI chip supply chain. We see competitive pressure most clearly in memory, where South Korean firms have led. AI accelerators in the US and leading-edge foundry in Taiwan appear better insulated, in our view.
Investment Implications
- AI infrastructure—stay convicted. We believe the opportunity is durable regardless of which model regime prevails. Cheaper models expand the universe of viable use cases, and total token demand grows with them (Jevons paradox). A quant lens may be especially valuable for dynamically tilting across winners and losers as the theme evolves across the entire AI supply chain.
- Software—be selective. We believe the sector's outlook depends more heavily on which model regime prevails. Favor companies across both large and small caps that route work to the best-fit model, over those tied to a single architecture.
- Asian equities—expect possible evolution. China's AI semiconductor ambitions create competitive pressure concentrated in memory, where South Korea leads; do not assume incumbent Asian index leadership is static.
What’s next for AI debt?
"AI-related debt is becoming a larger and more complex part of public credit—set to reach $2 trillion by 2028.1 For investors, the focus shifts from the scale of issuance to the quality of the structure: what type of debt it is, what supports repayment, and how liquid and resilient it will be in stress."
Ben Johnson, Global Head of Corporate Credit, Fixed Income and Liquidity Solutions
In Focus: AI Debt Is Not a Single Asset Class
In 2026, AI became a defining fixed income story, because the sheer scale of capital expenditure requires every corner of the capital markets to fund it. In our view, it comprises at least three distinct risk profiles that happen to share a theme:
- Investment grade (IG) bonds: Hyperscalers, key providers of AI infrastructure with diversified business models and well-established non-AI revenue streams; earnings support repayment
- Leveraged credit: Neoclouds, which rent out compute capacity; contracted compute revenue support repayment
- Off-balance-sheet financings: Special purpose vehicles funding data centers or chips; the specific collateral, contracts, and structure support repayment
What We See: Wider Hyperscaler IG Bond Spreads Are a Technical Story, not a Credit Story
Hyperscaler credit metrics are deteriorating under the weight of capex spending—but from a position of considerable strength. We read recent spread widening as driven mostly by technical factors:
- A relentless pipeline of new issuance—deals coming more often and in larger size
- Unpredictable issuance timing, tied to capex plans that keep increasing
- Mixed evidence of AI capex converting into monetization
The Off-Balance Sheet Market Is Scaling Fast
We estimate hyperscaler IG bonds will account for over 70% of AI debt this year but going forward we expect data center and chip financing—which sits off-balance sheet and tends to be either IG or high-yield2—to claim a larger share. To date, just 25 data center deals and one chip deal carrying off-balance-sheet IG or high-yield debt have priced. Yet this segment could grow from less than $40 billion in 2025 to over $850 billion by 2028, when we expect it to account for around 44% of total AI debt outstanding.3
Three Things We’re Watching
- Market evolution: Do index-inclusion criteria evolve to capture off-balance-sheet debt, broadening the investor base?4 Do rating agencies adjust how they assess balance sheets for future lease commitments?
- The first stress event. When a financed data center runs into trouble, the test is whether losses stay contained with the project or travel through the financing chain.
- Competition for capital: Does appetite for public AI credit fade if investors prefer to take AI exposure through equities and private markets?
Investment Implications
We are cautious on certain hyperscaler IG bonds due to the technical headwinds. Debt issued to finance data centers or chips can offer attractive income, at a premium to corporate bonds, compensating for structural complexity and illiquidity. Capturing these premiums demands disciplined underwriting across three dimensions:
- Collateral: the specific assets and contracts that source repayment
- Structure: provisions that mitigate credit and refinancing risk, such as debt amortization and residual value guarantees5
- Liquidity: resilience under stress
Will higher inflation and rates challenge the cycle?
"Today's economy resists any single historical template. It combines shocks that defined past decades—energy, tariffs, and geopolitics—with new structural forces such as AI, energy transition, and supply chain reorientation. That complicates the path for inflation, central banks, and rates, requiring a dynamic approach to fixed income exposures."
Simon Dangoor, Deputy CIO of Fixed Income and Head of Macro Strategies
In Focus: What Will Drive Rates?
Our forum took place against a backdrop of rising bond yields, in a month when the Fed, ECB, and BoJ all raised rates—the first such monthly occurrence.6 Central banks have turned hawkish despite limited evidence that the energy shock has broadened into persistent underlying price pressures. The motivation is risk management—an unwillingness to repeat 2022 by falling behind the curve—enabled by growth resilient enough to absorb tightening without triggering recession. The question foremost on our investors' minds: what could reignite upward pressure on rates from here? Two potential candidates stand out—inflation and fiscal.
What We See: Varied Paths for Inflation
- The risk-supportive path: If the trifecta of tariffs, energy, and AI-driven price pressure eases, a "goldilocks" regime of robust growth and stable prices could follow.
- The risk-negative path: If inflation instead falls because growth is decelerating—as resilience to higher rates and energy prices fades—we view the setup as far less attractive for equities and credit.
The Long-End and Fiscal Risk
Fiscal deficits are the recurring elephant in the room at our quarterly forum. They come into focus when conditions around them change—higher rates, higher inflation, or policy that worsens the trajectory. Each can generate rate volatility and unsettle markets. This puts France's spring 2027 election in focus.
Is the Market Repricing the Neutral Rate?
Consensus anticipates a measured hiking cycle. The live question is whether markets come to price in more tightening to tame inflation amid resilient growth. One clue: bond yields have risen this year while inflation expectations have stayed anchored—consistent with the market marking up its estimate of the neutral rate rather than its inflation forecast.
Investment Implications
- Rates: We are cautious on rates, focusing on relative value opportunities.
- Income: Focused on sourcing income, including in high-yield and securitized credit.
How Have Macro and Market Convictions Changed from Last Quarter?
- Rates: Central banks have shifted to rate hikes to risk-manage the energy shock. We expect the Fed to engage in a shallow hiking cycle.
- Growth: Resilience is now global, not just a US story. This has moved the economy and market from late-cycle back into expansion.
- AI debt: The latest round of earnings confirmed that AI capex plans are rising. In tandem, we revised our projections for AI debt supply in the pipeline higher.
What’s Shaping Our Outlook for 4Q?
Macro Trends Impacting Our Investment Views
We believe the AI investment cycle and steady consumer spending—supported by a healthy labor market and positive “AI wealth effects” from equity markets—should keep growth resilient. But we are watchful of headwinds from persistent energy inflation, higher rates, consumption divergence across income quartiles, and potential labor market weakness.
Tariffs, energy, and AI—alongside other volatile prices like airfares—pose continued upward pressure on inflation. We see a path for these price pressures to ease but remain mindful that resilient growth and fresh supply shocks could keep inflation above central bank targets.
Central banks have shifted to rate hikes to risk-manage the energy shock. We expect measured hiking cycles from the Fed and ECB, while the BoJ accelerates its tightening as the energy shock compounds existing domestic inflation. But an uncertain backdrop makes policy a moving target.
Market Views Considered in Our Portfolio Construction
Diversified by region (emerging markets, Europe, Japan) and by style (small caps), anchored by the AI infrastructure theme, using both fundamental and quant approaches.
Cautious with a focus on relative value opportunities driven by diverging central bank policies. For instance, we anticipate a more gradual tightening path in the UK than is currently discounted by market pricing. Conversely, we project a swifter hiking cycle in Japan relative to what is implied by forward rate curves.
Focused on sourcing attractive income across sectors, including high-yield, securitized credit, and select AI debt financings.
The Bottom Line
1 Goldman Sachs Asset Management. Figures represent internal estimates for off-balance sheet data center and semiconductor financing within IG and HY public credit markets. Projections are based on forecasted AI capex, expected public credit market share, and the allocation of funding across various credit channels. Data as of September 16, 2026
2 Why AI Debt Sits Off-Balance Sheet: Hyperscalers utilize special purpose vehicles (SPVs) and joint ventures to build and lease back data centers, keeping project-level debt within separate legal entities. This structure ring-fences operational risk, aligns debt maturity with long-lived asset lives, and shields parent-level credit ratings by avoiding direct leverage on the corporate balance sheet.
3 Goldman Sachs Asset Management. Figures represent internal estimates for off-balance sheet data center and semiconductor financing within IG and HY public credit markets. Projections are based on forecasted AI capex, expected public credit market share, and the allocation of funding across various credit channels. Data as of September 16, 2026.
4 Off-balance-sheet AI debt is frequently structured as "144A-for-life" securities. While these instruments are eligible for the Bloomberg US Corporate High Yield and Bloomberg Global Aggregate indices, they are excluded from the Bloomberg US Corporate Index (the primary US IG credit benchmark).
5 These credit-enhancement features seek to address downside risk from two angles: amortization progressively amortizes principal to reduce overall capital exposure, while Residual Value Guarantees (RVGs) establish a contractual floor to protect against asset depreciation. While these mechanisms may improve repayment certainty and recovery prospects, the inclusion of an RVG shifts a portion of the risk profile from asset-level performance to counterparty credit risk and do not entirely eliminate risks.
6 Central Bank Websites, Macrobond, Goldman Sachs Asset Management, Bloomberg. As of October 1, 2026.
