AI Bubble Risks and Gold’s Role, per Rosenberg & Bernstein on Excess Returns

The Excess Returns episode brings former Merrill Lynch colleagues David Rosenberg (Rosenberg Research) and Richard Bernstein (RBA / Janus Henderson) back together to dissect the 2020s AI-driven market, inflation, and portfolio positioning.

Host Matt argues that both men have been unfairly labeled as a "perma bear" (Rosenberg) and "perma bull" (Bernstein), and instead highlights their shared focus on discipline and sector rotation. The conversation centers on one big argument: according to Rosenberg and Bernstein, the AI trade has become a dominant, concentrated driver of market returns and capital spending, and that very dominance may ultimately be its vulnerability.

Rosenberg contends that the stock market’s strength in August 2026 is being mistaken for economic strength, when in his view it largely reflects a "generational" AI theme rather than underlying macro resilience. Bernstein, drawing on lessons from past bubbles, argues that massive capital flows into AI-related assets and data centers are starving other parts of the economy—especially housing and infrastructure—of necessary investment, planting the seeds for future economic and market imbalances.

Bernstein characterizes the last few decades as an "era of bubbles" fueled by repeatedly loose monetary policy, and he places the AI trade squarely in that lineage. He says roughly half of US business capital expenditure is now AI or AI-related, growing strongly in real terms, while what he calls "ex‑AI" capex is negative year over year. In his framework, bubbles are inherently inflationary because they grossly misallocate capital across the economy.

Using the late‑1990s tech boom as a template, Bernstein recalls how money then flooded into dot‑coms, fiber optics, and technology while US refineries and energy infrastructure were literally running on 1970s technology. His argument now is similar: he believes capital is being thrown at data centers, chips, and AI platforms while housing and transport infrastructure are being starved, contributing to US housing unaffordability and logistical bottlenecks.

Rosenberg’s thesis is more macro-focused. He argues that, excluding the AI boom, many core parts of the US economy are soft: non‑residential construction is contracting, the housing sector is shrinking on a year‑over‑year basis, and auto- and rate‑sensitive sectors are far from boom conditions. He estimates that the four‑quarter average of real GDP growth after upcoming third‑quarter data will be around 1.4%, below the Federal Reserve’s own estimate of 2.0% potential growth. In his view, this points to building slack and a fundamentally disinflationary backdrop even as AI spending props up headline growth.

Together, they converge on the idea that AI is both the main support for current equity valuations and the main driver of a dangerous concentration of risk, while a broad set of non‑AI assets may offer better risk‑adjusted opportunities over time.

On monetary policy, Bernstein highlights the Taylor rule debate. He notes that in prior years many conservative economists treated the Taylor rule as the "Rosetta Stone" of sound policy and criticized low rates whenever the rule signaled hikes. He points out that the Atlanta Fed now maintains around 30 versions of the Taylor rule, and, according to his reading, all currently suggest the Federal Reserve should be hiking rates. Bernstein says he personally thinks rates should rise based on broader economic data, but his main point is to call out the silence of prior critics now that the same framework is pointing toward tighter policy.

Rosenberg counters that Taylor‑rule prescriptions are highly assumption‑driven, especially regarding the "right" real rate and expectations, and that actual policy has long diverged from those model outputs. He stresses incoming growth data: he expects a four‑quarter real GDP run rate of roughly 1.4% against 2.0% estimated potential, which he interprets as demand falling below supply and putting downward pressure on underlying inflation.

To support his disinflation view, Rosenberg cites:

• The Dallas Fed trimmed mean inflation running around 2.2%, down from 2.7% a year earlier and at its lowest five‑year trend.
• Unit labor costs growing about 0.5% year over year, which he calls evidence that labor is not an inflation driver now.
• Real disposable personal income running fractionally negative, with most income growth accruing to corporate profits.

On markets, Rosenberg argues that the Federal Reserve need not "pop" the AI trade because the credit markets are already tightening around it. He points to widening credit default swap (CDS) spreads and rising financing costs for many tech names as indicators that credit is starting to question the sustainability of AI valuations. He recalls from the housing bubble that credit markets sniffed out trouble in mortgage and asset‑backed securities well before the equity peak in October 2007, and he expects a similar sequence where credit leads any eventual AI reversal.

Gold is another focal point. Rosenberg attributes recent gold weakness to rising real interest rates and a strong US dollar during the US‑Iran conflict, arguing that gold’s price historically has an almost perfectly inverse correlation with real rates. He describes the recent decline as one of roughly ten mini bear markets or corrections since gold bottomed around 1999.

For gold’s long‑term drivers, he lays out a supply‑demand narrative:

• From 1980 to 1999, gold fell from about $850 to roughly $300 as central banks sold reserves and increased Treasury holdings; bullion’s share of central bank reserves dropped from over 70% to about 10%.
• The Washington Agreement in 1999 effectively placed a moratorium on further large‑scale central bank gold sales, coinciding with gold’s secular bottom.
• Since around 2010, central banks have been consistent net buyers of gold. Rosenberg estimates that bullion now accounts for roughly 25–30% of reserves and notes that gold supply only grows 1–1.5% annually, while demand—driven largely by central banks—has been running closer to 2–3%.

He suggests that if reserve allocations continue to mean‑revert toward historical peaks, this central‑bank demand could remain a powerful tailwind for gold, despite cyclical setbacks tied to real rates and the dollar.

Both guests explicitly acknowledge that their frameworks could be wrong or early, and they highlight several offsetting forces. Rosenberg stresses the importance of avoiding stubbornness, recalling how difficult it was to maintain a bearish housing view pre‑crisis while market participants and even successful short sellers like John Paulson faced intense pressure before being vindicated. He frames conviction as a balance between backbone and flexibility.

On rates, Rosenberg concedes that there are arguments for further hikes and that recurring price shocks—especially from energy and geopolitics—could keep headline inflation elevated. He notes that earlier wage‑price dynamics in 2021–2022 were not truly "transitory" and did resemble a wage‑price spiral, albeit over 18 months rather than a 1970s‑style decade. He warns that further exogenous shocks could again disrupt his disinflation outlook.

Bernstein, for his part, is careful not to claim omniscience on the Federal Reserve’s correct policy; he says he does not fully subscribe to treating the Taylor rule as infallible and emphasizes that his recent commentary has been partly meant to "poke" the community that previously treated it as such. He allows that the Fed historically has not fully internalized the economic damage of financial bubbles, implying that policy could stay looser or tighter than his models would suggest.

For gold, Rosenberg admits that short‑term moves can be dominated by liquidity events and forced selling rather than fundamentals. He cites 2008 as an example, when gold sold off after Lehman Brothers as investors met margin calls, only to resume its longer‑term bull trend later. In his August 2026 view, he argues that the recent correction looks more like another cyclical blip within a secular uptrend, but he does not give a timeline for his own long‑run price target and emphasizes that timing is inherently uncertain.

Finally, both guests underscore that, despite their skepticism about the AI frenzy, they are not outright shorting the equity market. Rosenberg notes that around 50% of his model portfolio is in equities, with low beta and a focus on high Sharpe ratios rather than aggressive directional bets. This implies that, in their frameworks, valuation and sentiment may be stretched, but the path and timing of any unwind are highly uncertain and represent a key risk to any contrarian positioning.

Toward the end of the discussion, Rosenberg and Bernstein focus on sentiment and positioning as critical forward signals for the AI trade and broader equities. Rosenberg cites several indicators he sees as flashing extreme optimism:

• Mutual fund managers are reportedly down to about 1% cash, which he characterizes as exceptionally low.
• A sentiment gauge he references (he calls it "market vane sentiment") is around 78%, which he describes as near its highest reading ever.
• Consumer confidence surveys may be weak on the economy, but when respondents are asked about the stock market’s future direction, expectations are in roughly the top 5% of historical readings.
• Federal Reserve flow‑of‑funds data show about 73% of US household financial assets in equities, roughly 7% in bonds, and the remainder in cash—levels Rosenberg says have never been seen before.

He also cites a Bank of America global fund manager survey indicating that only about 2% of managers see an economic downturn in the next year, versus a long‑run average implied risk closer to 15%. He interprets this as evidence that many investors now implicitly treat both the business cycle and market cycle as "repealed," a psychology he finds dangerous.

Rosenberg summarizes his caution with a so‑called rule: when everyone is on the same side of the trade, he believes it is time to take some chips off the table. However, he reiterates that he is not in cash or short; instead, he is emphasizing diversification, lower beta, and risk‑adjusted returns.

Bernstein, channeling the legacy of former Merrill strategists he worked with, suggests that a constructive way forward is to look for improving charts outside the core AI complex. He implies that there are sectors with better technical profiles away from the most crowded AI names, though he cautions that some of those areas may still be correlated with AI and require careful risk analysis.

Going forward, both men indicate they are watching:

• Credit spreads and CDS pricing for AI‑exposed issuers.
• The relative performance of housing and rate‑sensitive sectors.
• Central bank gold purchases and movements in real interest rates.
• Fund cash levels and survey‑based recession expectations.

According to the guests on Excess Returns, these indicators will help signal whether the AI trade is topping, whether disinflation continues, and which non‑AI assets may become more attractive as the cycle evolves.