AI-pocalypse?...Not so Fast!
- SWS
- Jan 28
- 7 min read
Updated: Sep 10
Energy Policy Perspectives Vol. 2 - January 28, 2025
DeepSeek. The Chinese startup released its DeepSeek R1 Large Language Model (LLM) AI assistant app, suggesting that it created the very capable app using older, less capable GPUs, building its LLM with the aid of existing open AI models, while utilizing expenditures of only about $6 million. Marc Andreessen called it AI’s “sputnik moment.” However, AI pioneer Yann LeCun suggested that it showed the power of open AI development and not necessarily that Chinese AI developers have surpassed its U.S. counterparts.
The big declines. You have seen all the dismal figures that followed in Monday’s trading. The NASDAQ Composite declined by over 600 points yesterday, or nearly 3.1%. NVIDIA Corporation (NVDA, $118.42, NR) declined nearly 17%, knocking nearly $600 billion off the stock’s market cap. Tech in general was weak. However, the stock market decline spread in multiple directions while yielding gains in some more defensive industries. While the S&P 500 declined 1.5%, NuScale Power Corporation (SMR, $20.27, NR) declined 27.5%, Constellation Energy Corporation (CEG, $275.00, NR) declined 20.9%, and Vistra Corp. (VST, $137.08, NR) declined 28.3%. The E&P sector did not fare as bad, but the gas weighted equities materially underperformed yesterday (down 7.2% for our gassy weighted coverage vs. 1.1% for our oily E&Ps), due mainly to concerns natural gas (natural gas was down 8.2% yesterday with the 2025-2028 curve down 3.1%) demand projections are potentially high from AI/power generation due to more efficient AI ahead, requiring significantly less computing power. While it is natural for large uncertainties like DeepSeek to lead to a decline in the equity market, we believe the huge run-up in some of the AI/power generation theme names in recent months is partly to blame. We believe that the DeepSeek news was a convenient excuse to reduce positions in some very large recent gains. We do not expect yesterday to represent a piercing of an AI bubble by any means but would not be surprised by a correction.
Fear & loathing in Silicon Valley. The DeepSeek news raised a plethora of AI questions for investors. Are the Chinese AI developers catching up to Silicon Valley? Is big tech spending too much to achieve its AI advances? Will fewer NVIDIA GPUs be required if the Chinese can develop big tech quality AI on a shoestring budget while utilizing inferior chips, reducing prospective AI spending? Will less electricity be needed than expected as a result of smaller GPU clusters and data centers? The fears related to the unanswered questions led to the market sell-off while some confidence in Silicon Valley’s American AI supremacy was also lost.
Over-done loathing. While the quality of the DeepSeek model was confirmed, many questions remain. While DeepSeek may have produced a quality model at low cost, it does not necessarily spell doom or lost supremacy for American AI developers. The number of GPUs and data centers prospectively required by big tech will ultimately depend on the numbers of parameters in their models, the processing speed required by their various types of AI (not everything is an LLM after all), the size of the data sets employed, the number of models being deployed and the level of ongoing research and training. While some important questions remain from the DeepSeek news, there are many kinds of AI and the resources required for the ongoing development of Artificial General Intelligence (AGI) is not the same as required for discreet LLM development.
Silicon Valley’s response? Silicon Valley AI developers have made it abundantly clear that they have every intention of being the first to develop AGI so they can be a dominant force, not unlike all of their historical declarations of planned dominance whether it be in search, social media, or e-commerce. What if the big tech titans did believe that the Chinese were indeed catching up and challenging their AI supremacy? We imagine that their response would be more research, more money, and more GPU clusters thrown at the problem. We do not imagine a decline in prospective AI development expenditures at all.
Jevons paradox. Jevons Paradox essentially suggests that as technological innovation reduces the resources and costs required for producing and using a particular good or service, more of those goods or services are demanded, eventually utilizing even more of the resources. At the end of the day, Jevons Paradox has often held for energy and technology. The paradox has been working on data centers thus far, with net energy demand additions for years exceeding efficiency gains despite some data center owners cutting data center energy use by up to 40% over the years. The same principle could be applied to AI. Should AI development become cheaper, akin to DeepSeek, we expect the new paradigm would merely accelerate the propagation of more narrow AI use cases/algorithms, and more AI/AGI research, which would itself also propagate more data at a rapid rate, necessitating more electricity demand, more NVIDIA chips, and more data center capacity demand overall.
The secular electricity growth thesis should remain strong despite the DeepSeek news, in our view. Even notoriously conservative U.S. DOE electricity forecasts expect about a 75% increase in electricity demand by 2050. Other forecasts do expect digital power demand growth to be a large factor in near- term electricity load growth, increasing U.S. power usage by perhaps 5% by 2028. One December 2024 Lawrence Berkely National Laboratory projection expects nameplate U.S. electricity generation resources required by 2050 to nearly triple. Frankly, our estimates suggest this was a conservative scenario. The undertaking is a gigantic one. Some reduced AI/data center electricity demand would not derail the power demand growth thesis. For one, the electricity demand growth expected over the next few decades is not entirely a result of AI or data center energy usage. In addition to growing digital usage, broader economic electrification is a strong growth driver for electricity requirements. Part of that electrification trend is vehicle transportation that we actually expect to be a larger source of electricity demand growth than digital growth. We worry more about the rapid pace of electricity demand growth and the strains on electricity infrastructure in the near-er term than the ultimate outcome or magnitude of the growth.
Potential for further weakness near-term in the gassy E&Ps with the risk natural gas demand expectations need to reset. Again, the gas weighted equities materially underperformed yesterday, due mainly to concerns natural gas demand projections are potentially high from AI/power generation due to more efficient AI ahead, requiring significantly less computing power. While a tremendous amount of uncertainty remains around demand in the coming years, some expectations calling for up to 18 Bcf/d of incremental AI and power generation demand by the end of the decade are a high bar, in our opinion. Accordingly, we see the potential for expectations for natural gas demand to begin resetting in the coming weeks as this latest news is digested by the market, which is likely to continue weighing on the gassy names, in our view.
The bullish natural gas thesis remains intact... While we have been more cautious the natural gas weighted E&Ps with mostly Hold ratings due to valuation and expect some weakness to continue near-term, the longer-term demand thesis (including increased LNG demand looming) remains intact but with demand expectations potentially requiring a reset on the AI/power generation side. While we are unable to quantify the magnitude at this early stage, the uncertainty may drive some investors to sell in the near-term. That said, we think the natural gas narrative will remain strong despite a potential softer AI/power generation demand.
…but limited upside at current strip prices. Unfortunately, even after yesterday’s steep selloff, we think most of the gassy E&Ps are discounting close to our long-term price deck of $4.00/Mcf, if not higher (vs. the 2025-2028 strip of $3.81/Mcf), with elevated valuations over the past year driven by investor inflows from generalist/tech investors. Currently, we see a better risk/reward opportunity with greater upside in the oily E&Ps in a mid-cycle $70/bbl WTI oil price environment but will monitor the natural gas names for an attractive entry point if the weakness continues, as we suspect it may.
Utility & clean energy stocks. While most of the benefits of the AI power demand growth story have accrued to nuclear generators thus far, most of the capital to be deployed will actually be invested by electric utilities. With the sell-off yesterday, many utility stocks benefitted with some defensive shift in the market evident through the day. Clean energy technology and nuclear technology names followed the large independent generators materially lower as investors worried about the implications of DeepSeek on the AI power demand thesis. We expect the move lower to be temporary for the power demand names. While utility stocks will remain partially tied to the direction of interest rates and inflation over the intermediate-term, they may enjoy a defensive benefit in the near-term, in our opinion.
Trump administration policy response? In a speech yesterday, President Trump said the DeepSeek news “should be a wake-up call,” but also called it “a good thing” that a cheaper AI might be possible. From a policy perspective, since President Trump often gauges his performance based on the stock market, we think there is now greater risk to more new regulations on China. Even tighter technology export controls, higher tariffs, and reduced immigration and higher education access could all be targets of a policy response. We believe some reflexive response from the administration to a perceived risk to the President’s vision of a new American golden age, as evidenced in the executive order, “The First 100 Hours: Historic Action to Kick Off America’s Golden Age” of January 24, 2025, is likely. After all, in the same executive order, one of the President’s greatest claims of the first 100 hours is his “securing” of the $500 billion Stargate AI deal.
New executive orders this week. On Monday, the Trump administration issued a number of new executive orders, principally defense related. “The Iron Dome for America”, January 27, 2025, executive order suggests that missile attack “…remains the most catastrophic threat facing the United States.” While perhaps graver threats exist, the new executive order is seeking a DOD plan for deploying a new missile defense system last attempted with the Reagan era Star Wars Strategic Defense Initiative (SDI). In addition to the geopolitical destabilizing potential of such a program, we believe that a new major defense program of this scale could worry inflation hawks and roil markets further. We believe that this exogenous market risk bears watching closely.
PLEASE SEE BELOW FOR IMPORTANT DISCLOSURES, REG AC ANALYST CERTIFICATION AND DISCLAIMERS
Siebert Williams Shank & Co., LLC or its Affiliates do and seek too business with companies covered in its research reports. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision.
Analyst Certification
We, Christopher R. Ellinghaus and Gabriele Sorbara hereby certify that the views expressed in this research report accurately reflect our personal views about the subject companies and their securities. We further certify that no part of our compensation was, is, or will be directly, or indirectly, related to the specific recommendations or views contained in this research report.
Financial Interests: Neither we, Christopher R. Ellinghaus or Gabriele Sorbara, nor any member of our households own securities in any of the subject companies mentioned in this research report. Neither we, nor a member of our households is an officer, director, or advisory board member of the subject company or has another significant affiliation with the subject company. We do not know or have reason to know at the time of this publication of any other material conflict of interest.
Analyst Compensation: The authors compensation is based upon the value attributed to research services by Siebert Williams Shank institutional brokerage clients. The authors of this report are compensated based on the performance of the firm, and have not received any compensation in the past 12 months from any of the subject companies mentioned in this report. The performance of the firm is driven by its secondary trading revenues, investment banking revenues, and asset management revenues.
Siebert Williams Shank Equity Research Ratings Key
BUY: In the analyst's opinion, the stock will outperform the S&P 500 on a total return basis over the next 12 months.
HOLD: In the analyst's opinion, the stock will perform in line with the S&P 500 on a total return basis over the next 12 months.
SELL: In the analyst's opinion, the stock will underperform the S&P 500 on a total return basis over the next 12 months.
Distribution of Equity Research Ratings as of: January 27, 2025
IBC (Investment Banking Clients) is defined as companies in respect of which Siebert Williams Shank (the “firm”) or its affiliates have received or are entitled to receive compensation for investment banking services in connection with transactions that were publicly announced in the past 12 months.
Other Important Disclosures
Investment Banking Disclosures: Within the past 12 months, the research analyst authoring this report has not participated in a solicitation of any subject company mentioned within this report, with or at the request of investment bankers, for investment banking business. Within the past 12 months, the firm and its affiliates have not managed or co-managed a public offering of the securities of any subject company mentioned within this report, nor has the firm received compensation for investment banking products or services from those companies.
Firm Compensation: Within the past 12 months, the firm and its affiliates have not received compensation for any non-investment banking products or services for subject companies mentioned in this report, and none of these subject companies have been a client of the firm during the past 12 months.
Stock Ownership: The firm and its affiliates do not own 1% or more of any class of equity security in this report, and do not make a market in any such securities.
Disclaimers: The information and opinions contained in this report were prepared by the firm and have been derived from sources believed to be reliable, but no representation or warranty, expressed or implied, can be made as to their accuracy. All opinions expressed herein are subject to change without notice. This report is for information purposes only and should not be construed as an offer to buy or sell any securities. The firm makes every effort to use valuation methodologies that it believes to be reasonable in the derivation of price targets, but we do not guarantee that such methodologies are accurate.
Additional Information: To receive any additional information upon which this report is based, please contact:
Siebert Williams Shank, Research Department
100 Wall Street, 18th Floor New York, NY 10015
212-830-4500
