JPMorgan has identified eight stocks facing significant headwinds following the market disruption caused by China’s DeepSeek AI model, which demonstrated that advanced artificial intelligence can be developed more efficiently and cost-effectively than previously thought. While mega-cap technology companies continue their massive AI spending commitments, the investment bank warns that certain sectors could face substantial challenges as the industry adapts to new training methodologies.
The DeepSeek revelation centers on more efficient AI training and inference costs, potentially reducing the need for large-scale infrastructure investments while simultaneously enabling new AI applications that were previously economically unviable. This creates a complex balancing act between decreased demand for current high-density solutions and increased demand for inference workloads.
Among the biggest potential losers, Fabrinet leads with a 17.2% decline since DeepSeek’s emergence. The manufacturing solutions provider faces reduced demand tied to Nvidia’s vertically integrated solutions, which may be less favored in inference workloads as hyperscalers increasingly adopt custom ASICs. Caterpillar has dropped 11.9%, with JPMorgan estimating the company’s data center revenue exposure at high single-digit percentage of total Machinery, Energy and Transportation sales through backup power generators and Solar turbines.
Infrastructure and component suppliers face particular pressure. Amphenol, which sells electrical and fiber optic connectors for data centers, could see disruptions in demand as the industry shifts away from high-density solutions. Oracle has declined 7.2%, with JPMorgan suggesting DeepSeek increases the likelihood of overbuilding based on overly aggressive data center plans. Intel dropped 5.5% as the continued shift to accelerated computing could negatively impact server CPU demand.
Power generation and heavy machinery companies are also vulnerable. Cummins, a manufacturer of engines and generators with significant data center sales exposure, faces lower estimates under “pre-DeepSeek power generation models.” Custom Truck One Source, with 60% revenue tied to transmission and distribution markets, could see growth curtailed by efficiency gains from DeepSeek-style models.
Interestingly, CS Disco has gained 5.1%, as the legal services AI provider could benefit from greater disintermediation opportunities as customers utilize efficient, low-cost AI models more frequently. This highlights how DeepSeek’s impact creates both winners and losers across the AI ecosystem.
Key Quotes
A key theme to come out of their analyses is the potential for reduced training and inference costs, which could lead to balancing acts between (1) the decreased need for large-scale investment in current applications versus the rapid propagation of new AI applications that were previously not economically viable; and (2) a potential value shift both up to the application layer (inferencing) and down to the infrastructure layer (training)
JPMorgan analyst Claudia Hueston explained the core tension created by DeepSeek’s efficiency breakthrough, highlighting how cost reductions could simultaneously decrease infrastructure demand while enabling new AI use cases, fundamentally reshaping value distribution across the AI stack.
Content increase for connectors is tied to the increasing density of the vertically integrated solutions from Nvidia from one generation to another, and lower demand for the high density solutions could be a potential headwind to the bull case despite offsets from the increase in Inferencing demands
JPMorgan’s rationale for Amphenol’s vulnerability illustrates how component suppliers built business models around ever-increasing density requirements may face demand disruption as the industry shifts toward more efficient AI architectures.
As compute complexity continues to grow, we believe there will be a continued shift to accelerated computing, which would negatively impact server CPU demand
The bank’s assessment of Intel’s challenges highlights how the move toward specialized AI processors and accelerated computing could further erode traditional CPU market share, compounding existing competitive pressures.
Our Take
JPMorgan’s analysis captures a crucial moment where AI efficiency threatens established infrastructure assumptions. The irony is striking: DeepSeek’s breakthrough in creating powerful AI more cheaply could paradoxically slow the AI infrastructure boom that has driven massive stock gains. This represents a classic technology disruption pattern where innovation makes previous solutions obsolete.
The real insight lies in understanding this isn’t about AI slowing down—it’s about AI becoming more economically accessible. Companies positioned for the “old” AI economics of massive data centers and power-hungry training runs face reassessment, while those enabling distributed, efficient inference could benefit. The 17% decline in Fabrinet versus CS Disco’s 5% gain perfectly illustrates this divergence.
Investors should watch whether DeepSeek’s efficiency gains prove replicable at scale and whether Western companies can successfully adopt similar methodologies. The answer will determine whether this represents a temporary correction or a fundamental reset in AI infrastructure economics.
Why This Matters
This analysis reveals how quickly AI innovation can reshape entire supply chains and investment theses. DeepSeek’s demonstration that advanced AI models can be built more efficiently challenges the prevailing assumption that AI progress requires exponentially increasing infrastructure investments. This has profound implications for the estimated $1 trillion in planned data center spending over the next several years.
The ripple effects extend far beyond chip manufacturers to encompass power generation, electrical components, heavy machinery, and enterprise software companies. JPMorgan’s identification of these “second derivative” trades shows how AI efficiency gains could reduce demand for backup generators, fiber optic connectors, and high-density computing solutions while potentially accelerating adoption in cost-sensitive applications.
For investors and businesses, this represents a critical inflection point in understanding AI’s economic model. The shift from training-intensive to inference-optimized workloads, combined with the potential for custom ASICs to displace general-purpose solutions, could fundamentally alter competitive dynamics and capital allocation across the technology sector. Companies that built strategies around ever-increasing AI infrastructure demands may need to reassess their growth projections and investment plans.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- Wall Street Asks Big Tech: Will AI Ever Make Money?
- The AI Hype Cycle: Reality Check and Future Expectations
- How AI Can Help Make Buildings More Energy Efficient
- Encharge AI Secures $21.7M Series A Funding to Revolutionize AI Chip Efficiency
- Jensen Huang: TSMC Helped Fix Design Flaw with Nvidia’s Blackwell AI Chip