The market today faces a unique and dangerous confluence of risks spanning three critical categories. They are:
Macroeconomic risk: Growth is sputtering: Real GDP is decelerating, credit stress is growing, consumer sentiment hovers near a 70-year low, and factory surveys flash bright red. These cracks, formed before the latest policy shocks, have been split wide open.
Policy risk: Washington has flipped the script on globalization: Tariffs have halted Chinese goods and alienated allies, while shocking ports, supply chains and prices. The fiscal impulse has reversed.
Market risk: The foundation is brittle: Valuations assume earnings growth that no sober model supports. Leverage in ETFs and options will serve to shatter the glass house they helped construct. Complacency sets the stage for a slide worse than 2022’s shake-out.
Every bear market tells a story. The narrative often crystallizes around a central theme, a dominant fear that grips markets and drives asset prices lower. We saw it in 2022, as the specter of runaway inflation and an aggressive Federal Reserve ended the post-COVID euphoria. We saw it in 2020, when a novel virus (extremely briefly) brought the global economy to a standstill. Go back further: the 2018 taper tantrum, the 2008 Global Financial Crisis born from subprime excess, the 2001 bursting of the dot-com bubble. Each downturn had its defining characteristic, its unique signature of risk.
Today, as markets navigate treacherous crosscurrents, the question arises: what is the theme of this potential bear market?
It is hard to put a neat box around the innumerable and cross correlated factors of market pricing. But, across the landscape, several undercurrents keep appearing – the three pillars of the new bear market.
First, is the Capital Cycle
The market has benefited from both a public and private capital cycle, driven by new technology and stimulative policy. The effects on estimates and earnings alike has been impossible to miss. But the cycle is turning.
Second, is the Trade War.
Without exaggeration, we are witnessing the most dramatic rejection of the free-trade ethos that has defined the post-Cold War era – an ethos that has greatly benefitted corporate valuations and profits.
Finally, there is the Market Mirage.
Driven by crowded trades and the proliferation of leverage, many stock have completely disconnected from fundamentals. What’s more, an overwhelming sense of investor complacency has emerged out of a virtuous cycle of reassurance from the market’s consistent rebounds. Signal after signal, sometimes dire, are ignored.
Calling a bear market is one thing, the real question is how to trade it?
This is my first comprehensive analysis of asymmetric opportunities as we enter what is sure to be one of the most hated bear markets in history. Complacency may last for longer than is comfortable on the short side, it always has. Take, for example, 2008…
The writing was on the wall as early as 2007. Bear Stearns collapsed in March 2008. Still, by mid-May, SPX was a meager 9% off of its highs.
So, as we prepare for the inevitable conclusion to this saga, here’s my cross-sector, high-level takes on the micro aspect of a very macro slowdown. I explore how each of these risks is likely to manifest in single names and discuss setups I feel will serve well as the market weakens. I. Macro Risk: Tercio de Varas
Even before tariffs throttled trade and policy chaos cracked the confidence of CEOs, the underlying economy was flashing warning signs. Consumers — especially low and middle-income consumers — have been struggling for years though widely ignored. The employment market has been deteriorating since 2022.
Consumers polled by the University of Michigan expecting worse business conditions in a year is at its highest level since the survey began, which began rising in Q1 2025:
Outside of the lucky few, businesses aren’t faring much better. Borrowing costs have remained elevated for nearly the years, dashing the hopes for a refi for those struggling with debt service coverage, and small businesses are reporting the worst outlook on the economy that they have in years. PMIs have been in contraction for five months, trucking volumes are already down ~20% YoY and the wave of negative earnings guidance revisions is surging higher. Long before the recent collapse in new orders, manufacturing optimism had peaked most surveys:
And while we are nowhere near the default rates that were present even before the GFC, it’s evident that any sort of buoyancy to the consumer’s position provided by COVID-era stimulus has been completely exhausted, as evidenced by high levels of serious delinquencies.
Sometimes, it’s as simple as listening to the companies. Downward EPS guidance revisions have increased to the highest level since the 2022 downturn.
Earnings calls have taken a dour turn. Just over the past week: “Saving money because of concerns around the economy was the overwhelming reason consumers were reducing the frequency of restaurant visits.” — Chipotle CEO Scott Boatwright
“Relative to where we were three months ago, we probably aren’t feeling as good about the consumer now” — Pepsi CFO Jamie Caulfield
“I don’t care if you call it a recession or not, in this industry, that’s a recession” — Southwest Airlines CEO Robert Jordan
Back in February, Equifax flagged a slowdown in hiring that had not yet translated to a meaningful increase in unemployment claims — but it’s hard to imagine that doesn’t come to fruition after what’s occurred since. The labor market has been cyclically tight, but not secularly in the way some would have you think.
Put simply: the real economy can’t hold up given the massive headwinds it faces. The market may have rebounded, but the consumer is strained and the corporate sector is cautious. The sugar highs of fiscal spending are wearing off. And with cost pressure suddenly rising, the Fed is boxed in and government policy backstops will see none of the bipartisan support nor the speed of the COVID bills (and to some degree the IRA) that saved us from recession in 2022. The Capital Cycle
Capital investment is at the heart of economic growth but also highly cyclical. This duality explains how massive capital buildouts, from railroads in the 1850s to internet infrastructure in the last 1999s, often coincide with market booms and busts. The underlying technology is often a long-term winner, but the pattern of optimism leading to overhype and overcapacity is repeated again and again.
The nature of accounting also adds to these swings. In the early phases, there is a widespread overstatement of earnings as the spender’s costs are capitalized while the providers immediately recognize these sums in earnings. As the expense shows up in lagging depreciation earnings it depressed earnings for future periods.
This cycle we’ve just lived through was driven by a surge in capital goods, fiscal spending on infrastructure, new residential construction, and a massive buildout AI. The timeline looks something like this.
We’ve already seen the boom… AI: Shifting Narratives
After leading the market for much of 2023 and 2024, sentiment for the AI capex trade has clearly already peaked. Early signs of wavering were amplified by the DeepSeek scare — the first true moment of disbelief.
The AI buildout, a key driver of both economic and market activity over the past two years, is no longer a clean macro story. Capex remains high, but efficiency demands are rising. As LLM commodification concerns lead to more prudent and exacting capital plans, many of the “AI winners” are being reset. The next leg requires proof of ROI, not just ambition.
Hyperscalers, which hold the handle of the bullwhip, are increasingly cautious on excessive capex, after already committing to some of the massive spending ever undertaken in the tech industry. Now, architectural improvement and commodification of the LLM foundational models — not to mention the sheer scale of current spending — threatens the ROI on investment while a wave of depreciation expense is only beginning to drag on earnings.
To be clear, I’m not arguing that AI capex is going to collapse. Rather, I’m arguing that it’s impossible to maintain the rate of change that has occurred over the past two years. For picks and shovels, the rate of change is critical, as they require increasing investment for ongoing growth.
Moreso, market perception and valuation is likely to overshoot fundamentals on both the upside and downside. If there is a moderation, pause or decline in hyperscaler capex, the bullwhip will sting through a long value chain of suppliers and beneficiaries, from chips, to hardware, to integrators, and datacenter hosts.
All of these companies, and their suppliers, owners, and employees, have a significant exposure to the direction of spending.
Betting against the capex cycle is not a bet against AI. It is a bet based on the psychology of the hype cycle, a moderation of spending growth rates, and an evolution of more efficient AI development across both training and inference combined with the ongoing rapid deployment of incremental GPU capacity.
Here are the names most at risk.
First up, after collaborating on the Haves vs. Have Nots, I’ve asked CitriniResearch to weigh in on the risks to the King Kong of the AI trade: Nvidia. NVIDIA Corporation (NVDA)
The semiconductor complex, led by Nvidia, has driven markets for much of the 2020s, but cracks are showing. Nvidia still walks on water at 20x next-year sales and a 75% gross margin, yet markets rarely let one firm own a gold rush forever. Five cracks already spider across the surface: customers rolling their own chips, insurgents routing around Mellanox, software abstractions flattening CUDA, efficiency breakthroughs bending the compute curve, and a product road map that keeps skipping beats. When perfection is priced in, even a hairline fracture can wobble the story.
First: the hardware. After years of writing blank checks to Nvidia, hyperscalers now parade TPUs, Trainiums, and next-gen ASICs that hit “good-enough” throughput at cost. Cerebras welds an entire 300mm wafer into one die and Groq’s deterministic TPUs run 1,400 tokens per second. Both dodge the bandwidth chokepoint that props up Nvidia’s ASP. If a data-center manager can clear the same inference load with fewer boxes, the H-series premium looks fragile.
Second: the efficiency gains. While the parade of Jevon’s Paradox explainers has subsided, it still seems as though there is a disconnect between market understanding and the reality of exponential efficiency improvements. My first interaction with The Last Bear Standing was after he called the top in the AI Power Trade, and those insights (that power efficiency will likely result in power demand not expanding due to AI at the pace the market expects) apply across the technology - not just in the power theme.
Third: the cycle. A refrain I’ve found myself repeating all too often over the past 5 years or so: Just because AI or is a real thing does not mean semiconductors aren’t cyclical anymore. Nvidia just ate a 5.5 billion-dollar writedown on Hopper-derived H20 parts that never found a home. Blackwell, a stop-gap tuned for jumbo context windows, arrives as customers digest excess gear. Management is already teasing “Rubin,” a lower-precision workhorse aimed at robotics. Three overlapping architectures, swelling inventories, and a softening order book rhyme with the margin punch-out that clipped the stock in 2022. It’s starting to feel like we’ve seen this movie before.
Put this all together and the asymmetry tilts bearish, although NVDA is certainly an excellent operator and a company I’d want to own, the fact is I can’t underwrite these risks currently with the potential for such minimal near-term reward. A modest dip in compute demand or a modest nick to pricing cascades straight through a high-operating-leverage model. Cracks in the armor of NVDA’s CUDA moat or a significant series of breakthroughs on ASICs or TPUs from Amazon or Google. More open-sourced efficiency improvements a la DeepSeek.
Now, NVDA is not just training, and inference demand accelerating can be a big tailwind (although AMD may be better positioned and have more favorable base effects for that occurrence). Rubin could ship cleanly while buybacks plug the valuation gap.
Still, when every pixel of the canvas is priced for brilliance, “good” can trade like “bad” and “not perfect” can still pay. Oracle Corporation (ORCL)
As early mover Microsoft takes a step back, Oracle is beginning to fill its shoes taking on much of pre-training capex through its Stargate partnership with OpenAI and Softbank. But unlike Microsoft, whose massive growth in profitability over the past several years has provided plenty of organic funding for capex, Oracle is going “all-in”.
Even as Oracle’s current capex spending is smaller than the other hyperscalers, it is much larger relative to its cash flow generation. In the most recent quarter, its $5.8 billion capex spend was equal to its entire cash from operations – reinvesting the entire business’ cash flow into the buildout.
In short, it’s leveraging the company on this bet. The spending will certainly provide incremental growth uplift, but if return on investment falters, forward earnings will be blanketed with the long and lagging drag of depreciation for years with little to show for it. If Microsoft’s hesitance proves wise, there is significant downside risk in the stock.
Historically speaking, ORCL hasn’t needed to rely on aggressive capex, so the current bevy of moves smacks of desperation. I see this aggressive business pivot towards growth-at-big-cost to be a misguided departure from the mature tech-giant playbook which generally includes inorganic growth at reasonable hurdle rates and a lot of buybacks. Broadcom Inc. (AVGO)
As a major component supplier to datacenters, Broadcom has ridden the AI capex wave, but unlike NVDA which has seen its forward valuation metrics rapidly compress to cycle lows, AVGO’s forward metrics expanded significantly since 2023. On a NTM basis, AVGO trades at double its historical multiple and a 20% premium to NVDA.
While much of the company’s recent growth has been driven by US datacenter builds, AVGO’s business is much broader, with exposure across a range of applications and end uses. The company also has significant revenue exposure to China (and Singapore as a proxy), which could present challenges given the geopolitical landscape. Meanwhile AVGO faces competitive pressures within the AI market as recent news shows Google shifting towards MediaTek to help design its Tensor Processing Units (beginning with next generation designs in 2026).
Long-time CEO/Architect Hock Tan and AVGO management are well aware of their saturated position in the communications chip and embedded interconnect markets, so much so that they have made it a point to use cash generated by the chip business to expand into other verticals. The company went on an acquisition spree between 2018 and 2021, spending more than $110 billion on three software companies, with the bulk of that being on virtualization giant VMWare.
While virtualization is clearly a winning theme in the distributed, heterogeneous computing world, Broadcom is running the subsidiary as if it were a private equity firm, raising unit prices by 1,200% in some instances and increasing minimum core purchases from 12 to 72.
A slowdown in AI spending or sentiment could lead to significant multiple contraction, while its broader business portfolio demonstrates significantly slower growth with global macroeconomic sensitivity.
