Since ChatGPT went mainstream on November 30, 2022, the stock market has delivered strong gains, especially among the large-cap companies building and powering today’s AI ecosystem.
At the same time, investors are asking a fair question: Is AI becoming a bubble?
As asset managers, we understand why many investors feel both excitement and unease. AI is still early in its lifecycle, yet capital markets have already turned it into a defining investment narrative.
Here’s how we at Waldron Partners are thinking about AI’s long-term implications and what today’s investment environment may mean for portfolios.
1) Capital Expenditures: AI Is Consuming Cash (and Debt Markets Too)
The companies at the center of the AI ecosystem (hyperscalers and AI accelerators) have produced substantial free cash flow since late 2022.
But what’s most notable is how much of that cash is being recycled right back into the AI buildout, primarily through:
- Semiconductors and advanced chips
- Data centers and physical infrastructure
- Cloud compute capacity
- Research talent and engineering teams
In other words: this is a rapidly expanding compute arms race, and the cost of staying competitive is rising.
Big Tech is issuing debt to fund elevated capex
One notable development: several large technology companies have increased debt issuance to support AI spending while also continuing large share buyback programs, often used to offset dilution from stock-based compensation.
In 2025, Amazon, Alphabet, Meta, Microsoft, and Oracle collectively issued approximately $121 billion in investment-grade debt, more than 4x their five-year annual average of roughly $28 billion.
Does this create a credit risk?
At first glance, these capital structure changes can look concerning. But many of these firms still have:
- Strong balance sheets
- Meaningful recurring cash generation
- Opperating leverage in core businesses
The bigger risk isn’t “debt today”, it’s “capex forever.”
If AI spending stays elevated for years without a durable payoff, expectations may eventually reset. Which leads to the key issue…
2) Return on Investment: AI Payoffs Could Take Longer Than Markets Expect
Market enthusiasm has moved fast. Butreal-world buildout moves slower.
Data centers are not plug-and-play projects. The timeline often includes:
- Securing land, zoning, approvals, licensing
- Construction and facility outfitting
- Scaling power and networking capacity
- Ramping utilization to efficient levels
This can takemultiple years before AI is fully monetized at scale.
The power constraint is real
A growing open question is simple: where will the power come from?
AI compute requires massive energy, and meeting those demands could:
- Strain electrical grids
- Increase energy prices
- Trigger regulatory or political backlash
Even if AI revenue and productivity ultimately justify the investment, timing still matters.
Why timeline matters for valuations
In corporate finance, the longer cash flows take to arrive, the lower the present value of a project.
Extended timelines reduce:
- Net present value (NPV)
- Justified equity value
- Investor patience (especially in momentum-driven trades)
Our view: the market may be underestimating how long the full AI buildout will take.
3) Circular Investment: When AI Funding Becomes Self-Reinforcing
One of the most interesting dynamics in AI investing is what we’d call “circular investment.”
A high-profile example: Nvidia’s reported $100 billion investment in OpenAI, one of its most important customers.
The strategic logic
The optimistic interpretation is straightforward: if Nvidia helps OpenAI scale faster and OpenAI attracts more capital this could create a loop:
- OpenAI grows usage
- More computing is needed
- Demand for Nvidia chips accelerates
- AI investment momentum continues
The risk: demand becomes self-referential
The more skeptical view is that AI spending may increase because AI exists, and because growth must be demonstrated to justify the next funding round, whether or not end-demand is truly sustainable.
If AI ROI fails to materialize at scale, that feedback loop can break.
4) Valuations: AI Leaders Are Priced for Continued Success
As of today, several hyperscalers and AI accelerators trade at meaningfully elevated valuation multiples.
Forward P/E of selected AI hyperscalers/accelerators
Average forward P/E: 32
- GOOGL – 29
- AMZN – 32
- AVGO – 33
- META – 27
- MSFT – 30
- NVDA – 39
- ORCL – 36
Compared to broader U.S. equities
- S&P 500 forward P/E: 22.5
- RSP (Equal Weight S&P 500): 17
(a proxy for the “average” large U.S. stock)
So far, earnings growth has supported the narrative
Despite major stock gains, forward multiples haven’t expanded dramatically over the past year, suggesting earnings expectations have risen alongside prices.
- Average YTD return of the seven names above: 25%
- S&P 500 YTD return: 17%
Since 11/30/2022, those seven stocks have generated:
- ~58% annualized returns
vs.
- ~20% annualized for the S&P 500
The tradeoff: higher multiples can mean higher volatility
Elevated valuations aren’t inherently “wrong,” but they often:
- Amplify price swings
- Increase sensitivity to changing growth expectations
- Raise the penalty for execution missteps
In a fast-moving landscape, volatility may remain a feature, not a flaw.
5) Depreciation Schedules: A Quiet Accounting Issue That Could Get Loud
Michael Burry has recently highlighted a topic investors rarely debate: depreciation schedules.
Many large tech firms depreciate chips over roughly six years. The concern is that AI hardware may become obsolete faster due to rapid innovation.
Why depreciation assumptions matter
If useful life is overstated, earnings may appear:
- Artificially high
- Smoother and less volatile than reality
- Less reflective of true replacement needs
Potential consequences include:
- Higher “true” capex needs on shorter cycles
- Future impairment charges
- Investor confidence shocks if assumptions change
The counterpoint
NVIDIA’s Jensen Huang argues that older chips can still be productive and may be repurposed for less demanding workloads over time, supporting longer useful life assumptions.
Either way, the depreciation debate is a reminder: AI isn’t only about revenue growth. It’s also about the durability of the underlying asset base.
What May Come Next: AI Adoption Beyond Big Tech
So far, AI investing has been dominated by:
- The enablers (chips)
- The hyperscalers (compute)
But the next phase may look different.
Over time, we expect AI to become:
- Cheaper
- Faster
- More efficient
- More widely distributed across industries
That could shift attention toward adopters, companies using AI to improve:
- Productivity
- Cost structure
- Margins
- Speed of decision-making
- Operational resilience
Expect leadership shifts
As AI capabilities evolve, the market’s perceived winners are likely to rotate.
A recent example: since the release of Alphabet’s Gemini 3 model on November 18, Alphabet has outperformed Nvidia (+6.5% vs. -4%), a reminder of how quickly leadership can change when investors sense disruption.
Our Takeaway: Diversification Still Matters in an AI-Driven Market
AI may reshape the economy in ways that benefit far more than the companies selling chips or building data centers.
For long-term investors, we continue to believe the strongest approach is maintaining a diversified portfolio, rather than trying to identify the single “ultimate winner” in an AI arms race that remains highly dynamic.
We do believe the long-term outcome may be winner-takes-most, consistent with historic patterns in technology. But we also believe investors can pursue compelling upside without concentrating risk in a narrow slice of the market.