The landscape of artificial intelligence (AI) has become increasingly complex and competitive, particularly with the emergence of companies like DeepSeekRecently, some investors have begun to question whether the massive investments made by US tech giants in AI are yielding commensurate returnsAs a counterpoint, certain Asian stocks related to AI and semiconductors have surged, with the Hang Seng Index climbing by 15% in the past month alone, making it one of the top performers among major global indicesThis uptick piques global interest, especially as DeepSeek continues to disrupt the traditional paradigms of AI development.
In response to a flurry of inquiries from clients, major investment banks like Goldman Sachs, Morgan Stanley, and UBS have issued multiple reports focusing on DeepSeekWhile excitement builds, analysts caution that it's too early to ascertain how the innovations reported by DeepSeek will impact the broader AI ecosystem, especially concerning cost structures and demand for leading models like OpenAI’s GPT-4 or Meta’s Llama3.2. There appears to be a prevailing sentiment among these institutions that the applications and platforms will benefit from an environment where model competition drives down computational costs, a trend DeepSeek seems to embody.
Experts point out that the growing capital expenditures in AI, alongside the investment in larger models, align well with American investors’ appetitesZack Kass, formerly OpenAI's global market application head, elucidates this trendHe highlighted a new $500 billion AI infrastructure initiative named "Stargate" announced by the US President, exemplifying the narrative US investors love—despite DeepSeek not being a groundbreaking innovation, its arrival forces the market to reconsider its storytelling around AI development and funding.
The DeepSeek model is noted for its innovative architecture, particularly its Mixture of Experts (MoE) design that efficiently scales model capacity by implementing multiple specialized routing experts alongside a general shared expert—drawing parallels to a “divide and conquer” strategy
Advertisements
This efficiency not only enhances performance but challenges the entrenched belief in the need for exorbitant spending to build bigger models in the future.
Goldman Sachs points out that the capital expenditures of tech giants, including Google, Microsoft, and Apple, ballooned to around $160 billion in 2023, with projections of $200 billion for 2024. Yet, such spending largely consumes the free cash flow of these behemoths, raising questions about sustainabilityFor instance, Microsoft anticipates aligning its annual cash flow closely with an expected $80 billion in capital expenditure, a situation not unique to it.
UBS echoes concerns regarding the wave of uncertainty DeepSeek has unleashed on the AI trading market, noting that its R1 release has provoked doubts about what it calls a supercycle in AI computational investmentThe breakthrough characteristics of DeepSeek—successful training of competitive foundational models with minimal computational resources—has led to a contraction in AI computing costs across the sectorFor example, DeepSeek charges only 1.4 cents for each million tokens generated, significantly less than Meta's charge of $2.8 for equivalent output, showcasing a stark contrast in the pricing structures that could redefine AI resource allocations.
However, while DeepSeek showcases some promising strategies, there remains a cautious sentiment amongst analysts regarding its long-term viability in seamlessly integrating within the established AI ecosystemConcerns remain about its dependencies on advanced techniques like Multi-Head Potential Attention (MLA) and MoE, which, while effective for smaller models, might not translate well to larger-scale modelsFurthermore, despite adopting an open-source approach, the feasibility of its integration remains in question, yet the methodology might still inspire iterations across existing systems.
According to Zack Kass, the phenomenon can be explained through Jevons Paradox, which posits that as resources become more efficient, overall consumption experiences a rise
Advertisements
He speculates that in the future, AI could become as ubiquitous and affordable as the Internet, implying a significant increase in computing power demand but a collectively lower unit cost, allowing for a more egalitarian distribution of AI technology.
With turbulent market conditions, especially among Wall Street investment managers, concerns surrounding the implications of DeepSeek have permeated the psyche of institutional investors, particularly for those heavily invested in tech stocksThe US tech giants contribute significantly to the 2024 S&P 500 total return, showcasing their entwined relationship with investor sentiment.
UBS notes a nuanced impact from DeepSeek on internet firms like Amazon and Google, which are both consumers and providers of AI models through their services like Amazon's Bedrock and Google's Vertex AIShould the trends towards more efficiency and lesser capital requirements persist, operational and capital expenditures for these giants could dim, fostering further investor interest.
In assessing the risks associated with AI revenue forecasts, analysts observe that Meta could experience the least impact, followed by Amazon and GoogleMeta has yet to generate substantial income through its open-source model Llama, while Amazon relies on various external models for its offeringsGoogle, too, focuses on its proprietary Gemini models, positioning it in a complex competitive landscape.
Goldman Sachs suggests that Google and Meta currently occupy a particularly advantageous position among tech giants due to their advancements in AI application layersYet, emerging companies also stand to benefit from this wave of AI transformations, resembling how the rise of 5G technology permeated various sectorsEnterprises such as Canva, Adobe, and Gitlab, yet to go public, may find extensive monetization possibilities as they leverage generative AI technology to enhance their offerings.
Meanwhile, in the semiconductor domain, analysts still advocate for a buy-low strategy despite recent sell-offs impacting companies like Nvidia and Broadcom significantly
Advertisements
UBS emphasizes that compute power remains the cornerstone for enhancing AI performance, projecting that demand for AI training will continue to drive growth even with new algorithms surfacingThey conclude that AI computing is still in its infancy stage and has ample room for expansion.
Encouragingly, emotions across the Chinese market seem buoyed by DeepSeek's advancementsThe Shanghai Composite Index has crossed the 3300 mark, with the Hang Seng Index nearing a technical bull marketSpecifically, the Hang Seng Technology Index has surged approximately 23% since its January low, reflecting renewed confidence from overseas investors.
Goldman Sachs perceives that companies like Tencent and Alibaba are reaping rewards from the AI fervorWith Tencent's WeChat forming a robust ecosystem combining social and transactional functionalities, it is uniquely positioned to capitalize on AI-related applicationsFurthermore, Alibaba’s status as a colossal public cloud computing entity positions it advantageously in the steadily growing AI application landscape.
Firms like Century Interconnect and GDS Holding reflect the data center theme, with expectations that long-term demand for AI computing power will foster growth in public cloud and AI infrastructure investmentsThe consensus leans towards greater emphasis on application layers across sectors—from internet applications to manufacturing and autonomous driving—suggesting a reshuffling of perceived company capabilities as firms like Kingsoft and Yonyou are beginning to garner renewed attention.
However, Morgan Stanley's Asian tech team warns against macroeconomic risksThey posit that the technological breakthroughs represented by DeepSeek could lead to overvaluations along the Asian AI supply chainIn contrast, traditional non-AI tech firms currently face low valuations due to geopolitical uncertainties and sluggish global demand, likely inviting renewed interest from investorsThe uncertainty surrounding tariffs poses a lingering concern, as accelerated and broad tariff implementations could adversely impact corporate profits, global liquidity, risk appetite, and market valuations.
Advertisements
Advertisements