
An AI Efficiency Breakthrough: DeepSeek’s Impact
A sudden release from a Chinese AI start-up rocked markets last week. DeepSeek, a new large-language model (LLM), has demonstrated performance comparable to OpenAI’s ChatGPT while dramatically reducing compute and power costs through innovative design and optimizations. This news has the potential to upend the current AI narratives and surrounding technology ecosystem that have been driving financial markets this cycle.
There are a number of technological, financial market, and portfolio construction implications from this AI-related news:
1. Benefits of free(r) markets at work
This drastic improvement in LLM efficiency highlights both the classic pattern of technological innovation (delivering better performance at lower costs) and the unintended consequences of tariffs (when faced with barriers, competitors often respond with extreme innovation that can leapfrog existing technology).
This pattern emerged dramatically in post-WWII Japan, when Sony (then Tokyo Telecommunications Engineering Corporation) faced severe resource constraints and limited access to Western technology. Rather than trying to compete with American vacuum tube technology, Sony's co-founder Masaru Ibuka invested heavily in developing transistor-based electronics. While initially limited by Japan's post-war economy, this constraint pushed Sony to innovate in miniaturization and power efficiency. The result was the world's first pocket-sized transistor radio in 1957, the TR-63, which revolutionized consumer electronics and established Sony as a global technology leader. The company turned its initial limitations into advantages by focusing on smaller, more efficient devices—-a strategy that would define consumer electronics for decades to come.
2. Likely higher returns on investment from AI capex:
Early reports suggest the cost of training the DeepSeek model was an order of magnitude less than what incumbents OpenAI or Meta had been spending, in part because they were able to train their model with fewer and less advanced graphic processing units (GPUs). While these reports are somewhat misleading as they may not account for the full costs incurred by DeepSeek, the future direction is clear—better output with less resources thanks to ongoing innovation including DeepSeek’s techniques that others will surely adopt.
3. Potentially lower amounts of AI capex spend in the short to medium-term
Many AI-driven companies were down sharply on the news as the market is worried that potentially greater output from capex spend will mean less capex spending overall, at least in the short-to-medium term, which also would mean less GPUs from Nvidia, less electricity from utilities, etc.
On the one hand, efficiency improvements could lead to lower total compute demand in the medium term. Hyperscaler capex looks pretty strong for at least 2025-26, but orders could fall off after that if there is a time lag for demand to catch up to supply (i.e., perhaps an air pocket in 2027-28 as the 2023-24 spending boom comes due for replacement after 4 years).
On the other hand, Jevon’s Paradox is a real possibility here; if customers are getting better ROI, they will find more use cases and therefore total demand could increase more rapidly.
In either scenario, it appears like the long-term direction leans towards more AI use cases, higher ROI, higher capex, but depending on the speed of use case adoption, the path could be a lot bumpier in the first scenario and smoother in the second.
4. Competitive moats of the big tech platforms
Technology can be highly valuable for society but not always for the companies that provide it. Consider mobile networks: despite massive infrastructure investments, telecom companies saw modest returns while the real winners were consumers (through productivity gains), Apple (hardware), and companies like Google and Meta (advertising through mobile platforms).
The tricky question for big tech investors is whether the big capex spenders (Microsoft, OpenAI, Meta, Google) will be the big winners this time or if it will once again be the consumer of that product and/or companies built on top of the infrastructure.
On the one hand, these latest developments could potentially be quite positive for the major tech platforms (e.g., Amazon, Apple, Google, Meta, Microsoft, ByteDance, and Tencent). A key to the investment case of these platforms is they have the distribution to monetize technological improvements at scale. This has been a tailwind they’ve enjoyed for years and LLMs may be no different. In fact, improved efficiency may reduce the risk of overspending on capex at low ROI.
On the other hand, competition and new entrants like OpenAI, DeepSeek, and others down the road (think: quantum computing) may significantly reshape competitive dynamics, even if AI may be a major driver in society over the next 20-30 years. While obviously imperfect, investors should be wary of the Lindy effect.
5. Likely positive impact for consumers
End users have been net beneficiaries of increased competition in the AI space and this will likely continue. Five years ago, Google, Microsoft and Meta’s lead in cloud and AI seemed unassailable. Since then, competitors like OpenAI and DeepSeek have emerged with comparable or superior technology, driving rapid product improvement and pricing competition.
6. Businesses that can effectively use AI should flourish
Beyond consumers, companies that can effectively implement AI into their businesses may see opportunities to increase revenues and/or lower costs. Mawer portfolios contain a plethora of these businesses, including creative advertising (Publicis), helping lawyers to draft legal documents (RELX), providing faster and more efficient travel reservations (Booking), and in better pricing risk (Marsh & McLennan).
7. A reminder to stay diversified
Last Monday’s big tech selloff of some of the largest names in the equity market, combined with the quickly shifting technological landscape and resulting market narratives are a great reminder of why it’s important for investors to stay diversified (across region, business models, market caps, etc.).
While it’s still a little early to know the full implications of DeepSeek’s arrival onto the world stage, it does seem to support the economic rationale for AI capex, the ability of start-ups to break into markets that appear controlled by big tech incumbents, and the risks of overly concentrated portfolios.
This blog post is solely intended for informational purposes and should not be construed as individualized investment advice, research, or a recommendation to buy, sell or hold specific securities. Information provided reflects current views based on data available at the time or writing and may change without notice. Mawer Investment Management Ltd. and/or its clients may hold positions in the securities mentioned, which may create a potential conflict of interest. While efforts are made to ensure accuracy, Mawer Investment Management Ltd. does not guarantee the completeness or accuracy of this information and disclaims liability for any reliance placed on the publication. Mawer Investment Management Ltd. is not liable for any damages arising out of, or in any way connected with, its use or misuse.