The recent advent of DeepSeek's large language model has sent tremors throughout the technology sector, raising a multitude of questions among investors and analystsThis open-source model, which boasts performance capabilities comparable to giants like Google, OpenAI, and Meta, yet comes with reduced operational costs, has been a hot topic in financial circlesFollowing its introduction, stocks of leading technology companies such as Nvidia saw significant dips, before staging a somewhat recovery, igniting a renewed interest in artificial intelligence and its commercial viability.

Major financial institutions including JPMorgan Chase, Morgan Stanley, and Citigroup have jumped into the fray, releasing comprehensive reports that dissect the implications of DeepSeek's disruptive capabilitiesAmongst these analyses, a standout observation has emerged: the potential reduction of training and inference costs could reshape the operational landscape of artificial intelligenceThis could lead to a noteworthy balance between two previously conflicting dynamics: the decreasing demand for large-scale investments in AI application and the rapid adoption of freshly minted AI applications that were once deemed economically unfeasible.

Moreover, the findings suggest a dual directional shift in valueAs inference costs move downward, there will be a parallel potential rise in the value seen at both application and infrastructure layersIn the short term, major data center operators are expected to continue to lead the charge in AI hardware investmentsThe analysts at JPMorgan have developed two distinct investment baskets based on the companies that will either gain or lose from the rise of DeepSeek's technologyThe **positive impact basket (JPDSPOSI)** includes notable entities such as Alphabet, Amazon, Broadcom, and C3.ai, which span multiple sectors including software, semiconductor, and internet industriesThese companies are poised to benefit significantly from the technological advancements that DeepSeek promised, translating into reduced costs and expanded business opportunities.

Conversely, the **negative impact basket (JPDSNEGI)** features companies like Amphenol and Caterpillar, predominantly from the metals and mining, mechanical engineering, and construction sectors

Advertisements

These companies could face reduced business demand or restricted growth due to the impacts of DeepSeek.

Delving deeper, it becomes evident that the ramifications of DeepSeek extend beyond financial markets into various industriesIn the software sector, for example, the lowering inference costs facilitated by DeepSeek are anticipated to accelerate the applications of AgenticAIHere, the ongoing dialogue surrounding the speed of AI model evolution and the associated costs continues to prevail, primarily manifesting within the realm of softwareAs deployment and productivity become increasingly reliant on software capabilities, companies that support these advanced systems—such as Elastic and GitLab—are likely to reap corresponding benefits.

In the data space, firms such as Confluent and Informatica may also find themselves on the winning side, while major software companies like Snowflake and Salesforce could enjoy ancillary advantagesMeanwhile, Oracle may face risks pertaining to adjustments in their data center construction plans in light of these industry shifts.

The internet sector is not left untouched eitherThough giants like Meta, Amazon, and Alphabet continue to face the need for substantial capital expenditures, opportunities for growth arising from open-source innovations are plentifulMeta stands to integrate the advancements from new models into its Llama4 framework, while Amazon is expected to amplify the use of generative AI via the introduction of the DeepSeek-RI modelAlphabet's position, however, rests on a more precarious balance as it navigates the innovation landscape, attempting to speed up the development and distribution of AI agents.

In the semiconductor industry, the push for efficiency is set to catalyze increased demand for AI-focused semiconductorsCompanies such as Broadcom, Marvell Technologies, and Nvidia are expected to benefit from this trendHowever, Intel may find itself hampered, as rising computational complexities could affect their demand for server CPUs

Advertisements

The hardware and networking landscape will also face changes, with companies like Dell Technologies, Cisco, and NetApp likely to gain as the inference phase gains realism and relevance in the long termHowever, firms such as Fabrinet and Amphenol could encounter challenges given the fluctuations in Nvidia’s associated business streams.

In parallel, Morgan Stanley has also made its analytical mark in this discussionThey view the declining computational costs brought forth by DeepSeek through a macroeconomic lens, reminiscent of the transformative landscape of the 1990sThe overarching narrative suggests that reduced resource costs might stimulate corporate expenditure and bolster productivityThis hypothesis finds grounding in Jevons Paradox—the principle that advancements in technology leading to lowered resource consumption tend to increase overall demand, spiraling upward the total resource consumption while catalyzing broader applications of AI and facilitating product innovations.

Nevertheless, caution is warranted as Morgan Stanley's semiconductor research team led by Joseph Moore posits that the evolution surrounding DeepSeek may not significantly curtail associated capital expenditures related to AI infrastructureInvestment strategies appear firm, with major language model developers maintaining strong allocations toward infrastructure development.

In a striking affirmation from industry leaders, teams led by Stephen Byrd have underscored established commitments among language model developers to hefty infrastructure budgetsMicrosoft has projected upwards of $80 billion by fiscal year 2025 for data center expenditures, while Meta's CEO Mark Zuckerberg hinted at a planned investment of $60-65 billion by 2025, marking a 50% increase from 2024. Furthermore, the recently announced Project Stargate, an initiative linked with OpenAI, aims to funnel an astounding $500 billion into new infrastructure geared for AI in the U.S.

Despite these sizable forecasts, an analysis of these investments highlights a broader trend; most recognized construction goals largely cater to AI inference and other applications rather than solely focusing on AI training processes.

On the financing side, data center-related asset-backed securities (ABS) mainly serve cloud applications rather than direct AI training, yet projections indicate that the momentum for these ABS will remain robust.

In closing, Citigroup has also released strategic insights underscoring the recent sell-off in tech stocks, largely influenced by DeepSeek's release and tariff anxieties

Advertisements

Advertisements

Advertisements