✨⏩ What DeepSeek means for faster AI-driven economic growth

'The potential for a faster buildout of AI platforms and applications raises the prospect of a more optimistic adoption and productivity boost timeline.' - Goldman Sachs in a new analysis

Jan 30, 2025

My fellow pro-growth Up Wingers,

The big story in this space (and certainly in this newsletter) has been how Chinese startup DeepSeek has shockingly unsettled the American technology sector by demonstrating that superior artificial intelligence need not require lots and lots of superior chips.

The economics are especially uncomfortable for semiconductor makers. DeepSeek reports spending a modest $5.6 million to train its model (kind of), a fraction of the amount cited by players such as OpenAI, Google, and Anthropic in the USA. That both of DeepSeek’s models rank among the top performers on Chatbot Arena, a respected testing ground, adds considerable heft to its claims.

The news sent tremors through Wall Street trading floors, as investors contemplated whether the future of AI might require fewer of the expensive chips that have propelled Nvidia, notably, to a stratospheric market capitalization and a place as the most valuable public company on Earth. Perhaps frugality, not excess, might define the next chapter of computing.

Productivity potential

I’ve been waiting for the economics team at the bank Goldman Sachs to weigh in here since it’s consistently been offering some of the most comprehensive and widely cited analysis of the emerging generative AI revolution. The story up until now:

In March 2023, GS said GenAI could boost US labor productivity growth by approximately 1.5 percentage points over a 10-year period (equating to about $4. 5 trillion) once the technology becomes firmly embedded in corporate life — though this impact may vary based on AI's ability to handle different tasks and the extent of automation. So a productivity doubling, basically, and a big boost to GDP growth of the sort seen during the late 1990s and early 2000s boom. (See chart below.)
The implications for global prosperity are equally striking with GS reckoning the technology could eventually plump up worldwide GDP by seven percent —roughly equivalent to adding another France to the global economy. (Keep in mind that forecasts such as this one do not assume AI progressing to human-level artificial general intelligence or short AGI timelines. That’s a whole other ballgame.)
Finally, the bank sees the investment wave as having two distinct pulses. The first surge, which could reach two percent of GDP, stems from that voracious appetite for hardware to train and run AI models. The second, more enduring ripple flows from software investment, which should swell steadily as businesses embrace these handy digital tools.

“The Potentially Large Effects of Artificial Intelligence on Economic Growth” March 2023

(As an aside, I urge interested readers to check out my past essays covering productivity estimates, but also the extensive National Academies report “Artificial Intelligence and the Future of Work,” which digs into how economists like those at Goldman Sachs create these forecasts.)

The DeepSeek dividend

Setting aside the short-term stock market reaction, GS number crunchers have now offered a broad first-take on the macroeconomic implications of the DeepSeek breakthrough (although they add “there are valid reasons to question whether DeepSeek’s reported $5.6mn training cost fully reflects the cost of development or the hardware it was trained on.”)

First, their top line take in “DeepSeek Raises Micro Risks, Macro Upside,” which was just published:

From a macro perspective, the emergence of DeepSeek’s lower-cost models does not affect our view that the largest aggregate economic gains will come from the productivity boost enabled by generative AI. … While we are sympathetic to the view that these developments raise company-level micro risk, they have also, if anything, added to our confidence that AI-enabled productivity gains will be a major macroeconomic story in coming years.

Good to hear that, although it’s not as if DeepSeek changes nothing, as the initial market reaction suggests. GS goes on to explain that more affordable AI models raise important questions about how the economic value created by GenAI will be shared among different players in the industry. If high-end hardware becomes less critical for developing powerful AI systems, companies focused on building physical AI infrastructure “would likely capture a smaller share of the overall economic gains as profits.”

Again, look at the recent action of Nvidia stock. (To illustrate the point, the GS analysis notes that software contributed significantly to US economic growth from 1981–2012, but software companies captured only about 25 percent of the value created, despite the rise of some superstar mega-companies.)

Let the investment flow

While questions may continue to swirl about who will capture the economic spoils, such distributional concerns matter little for GDP, which measures total production regardless of its beneficiaries. Yes, more efficient AI training could, in theory, dampen capital investment as many investors worry, an amount projected to reach $325 billion by late 2025.

But GS economists “view this risk as limited” for two reasons. First, AI investment has thus far barely registered in official GDP figures, despite firms' increased spending. Second, DeepSeek's advances might actually spur greater hardware investment as established players scramble to maintain their lead. GS (bold by me):

More fundamentally, if the novel computational techniques employed by DeepSeek’s models indeed increase competition and lower costs, they could catalyze a faster buildout of AI platforms and applications that have so far posed a bottleneck to adoption and the impact on productivity, thereby raising macroeconomic upside. … The main macroeconomic impact from generative AI will come from the efficiency gains from AI-driven automation as companies incorporate the technology into regular production. … The potential for a faster buildout of AI platforms and applications—which we continue to see as the necessary step to facilitate adoption across a wide swath of companies—raises the prospect of a more optimistic adoption and productivity boost timeline. Our forecasts currently assume that US adoption reaches levels necessary to impact aggregate productivity statistics in 2027 with a peak impact in the early 2030s, with other DMs and major EMs lagging this timeline by a few years. The recent DeepSeek reports suggest adoption could happen sooner. … We also see DeepSeek’s reported breakthrough as posing upside risk to global GDP. The emergence of a credible competitor to US-based AI leaders could provide an uplift to global adoption and productivity through two channels.

On that global point, GS finds potential AI automation benefits to be similar across advanced economies due to “similarities in industry composition of employment.” And while the US may lead in adoption due to its recent AI development advantage, emerging non-US platforms like DeepSeek could accelerate global adoption.

Finally, rising geopolitical competition, especially between the US and China, may drive governments to boost domestic AI capabilities through increased investment and “lower regulatory barriers to encourage AI development and adoption.” The geopolitical tailwind factor figures prominently in the bullish economic thesis outlined in my 2023 book.

“DeepSeek Raises Micro Risks, Macro Upside” (January 2025)

Again, early days — not just with the DeepSeek news but with the Age of AI overall. As the analysis notes, AI adoption by business remains quite limited, with only six percent of companies currently using AI in regular activities, up marginally from four percent in late 2023. Most companies are waiting for "plug and play" solutions rather than just cost reductions before implementing AI across their operations.

Still, the bank adds, given that potential cost savings from generative AI are large and the marginal cost of deployment once applications are developed will likely be very small, we see adoption of generative AI as more of a question of ‘when’ rather than ‘if .’

This is a great quote from a new Economist piece that neatly sums up the DeepSeek impact: “For AI to transform society, it needs to be cheap, ubiquitous and out of the control of any one country or company. DeepSeek’s success suggests that such a world is imaginable.”

If you were confident before DeepSeek that GenAI was an economic accelerant — whatever the impact on specific companies — you should be at least as confident today, though probably more.