What’s next for artificial intelligence after DeepSeek?

Key takeaways

DeepSeek is an AI disrupter.

This Chinese competitor has seen excellent performance for an affordable cost and operates as a public access open source model.

AI investing is not jeopardy.

But it has changed. Costs had been falling prior to DeepSeek’s entrance, and we expect they’ll bode well for AI themes going forward.

These stories are evolving.

Given we are still in the early stages, we will continue to monitor these themes and expect ebbs and flows along the way.

On Jan 27, markets woke up in a panic to a seemingly unanimous — and surprising — investment narrative. DeepSeek’s new artificial intelligence (AI) model showed cutting-edge performance alongside significant efficiency gains, challenging the assumption that AI would require greater investment in computing capabilities for power improvements. With DeepSeek coming to the forefront so quickly, what does this mean for AI?

AI has promised to revolutionize our lives at work and at home with the potential for greater economic outcomes. Big tech companies have invested billions of dollars into research and development and infrastructure to support better AI capability development. Equities of large tech companies, semiconductor designers, and even utilities have benefitted substantially from this theme, but DeepSeek’s new R1 model is reported to have required 50% less to compute versus OpenAI’s GPT-4o model while achieving virtually the same performance.¹ The directionality of incremental improvements in AI is no longer defined by a linear relationship between cost and performance. In fact, recent model advancements have maintained or improved performance while significantly reducing costs.

We believe the AI competitive landscape is changing, with a renewed focus on efficiency and new entrants (like DeepSeek) into the market. Value capture from the intellectual property of models may be shifting as open-source models become more powerful. Yet, with cheaper models, we expect adoption to accelerate, which should maintain demand for computing power and the enablers of AI that have so far benefitted from this technological boom.

Big tech underperforms after DeepSeek makes a splash

After DeepSeek entered the market, 2025 returns of AI-related companies dropped, namely hyperscalers (large companies providing computing power and infrastructure for AI applications), semiconductors, and data centers. The only category not particularly suffering being utilities. While investors are questioning if AI is forever changed, we believe the answer is more complex. We think there are new wrinkles, but many of yesterday’s winners remain well-placed to potentially continue benefitting, even in this new era. More specifically, we believe AI adopters should begin to stand out as beneficiaries of this theme as costs fall and capabilities improve.

DeepSeek disrupts the AI competitive landscape: Lower costs and better performance

DeepSeek’s release and a step up in Google offerings have changed the competitive landscape. OpenAI took a premium approach, resulting in superior AI theme performance — but at a cost. Moving forward, they may focus on managing those costs through a mass-market approach.

On the other hand, DeepSeek has seen excellent performance, affordable cost, and is from a non-US competitor. Google’s models have also stepped up performance while managing computing costs, their latest model (made public in early February) setting the bar high.

2023 and 2024 focused on improving capabilities, regardless of the cost. More recent model releases show the once-rapid growth of AI computing requirements is slowing. And now, we’re seeing a step up in efficiency gains — a good quality for future AI adoption.

DeepSeek disrupts the AI competitive landscape: Open source catches up

DeepSeek’s results are also notable because the company chose to publish it as an open source model. In other words, it’s free access. Anyone can download and run it on their computer.

This may indicate a shift in the perceived value of AI model development. To date, proprietary models have led in technology, but they tend to require costly licensing agreements (for extra bells and whistles, like dedicated support and often top-of-the-line capabilities). AI adopters may wish to apply and fine-tune models from the open-source community to avoid these costs and restrictions around licensing. In other words, DeepSeek may be eroding part of AI’s intellectual property value.

The AI investment story isn’t in trouble, in our view — but it has changed

Falling costs are most likely a good thing for the AI narrative. We’ve suggested technology exerts a deflationary impulse by both enabling expanded productivity and costs of the technology itself falling alongside increasing adoption.

Many AI hypotheses assume ever-larger models power ever-greater capabilities. However, we believe narrower, more specialized models are more likely to see integrations and use cases in the industry. Moreover, costs had been falling before DeepSeek even entered the market. Iterations of OpenAI’s GPT-4 model enabled comparable performance at lower costs to access. We anticipate further refinement in more specialized use cases.

AI adoption still needs to play out

We believe the climb up the AI adoption curve — and the fact that we remain in the early innings — represents a long-term bull case for the theme. AI capabilities are impressive, but the onus is on developers and businesses alike to deliver tangible results. As use cases are identified and we climb the adoption curve, we expect the demand for AI specialists, products, and supporting infrastructure to continue benefitting.

In our view, AI is an ongoing story that is likely to see twists and turns in the short-term. It’s also likely to play a key role in future economic growth, specifically in information-focused industries.

Is concentration set to ease? Enthusiasm shifts toward Chinese tech

So far, the AI theme has largely been a US phenomenon, as captured by the Magnificent 7. In 2024, it made up nearly a third of the S&P 500 by market capitalization and contributed more than half of the year’s gains.² Since the release of DeepSeek, Magnificent 7 stocks as a group have underperformed and Chinese technology stocks have jumped higher,³ with investors placing a renewed focus on AI projects that were in motion beforehand. It appears the sentiment towards Chinese investment has seen a leg higher.

Overall, we see competition is increasing, but the need for supporting infrastructure for AI adoption remains.

What’s next in AI after DeepSeek?

Cheaper models could help accelerate adoption of AI, begetting an ongoing investment cycle as the technology is scaled up. We believe the current buildup in computing power and the growing number of strong models set the stage for specialization and refinement of the technology for their ultimate use cases and implementation. High-profile models like GPT-4o and DeepSeek’s R1 have demonstrated the capabilities of recent AI advancements. We expect private enterprise to carry the torch on delivering AI-powered tools and services and promoting the adoption of these technologies.

The path ahead: Cheaper AI should help accelerate adoption

In conclusion, lower costs and increased competition should help accelerate AI adoption, but many businesses lack the human capital or organizational agility to integrate it, likely meaning a relatively prolonged adoption period.

Open-source AI offerings are becoming increasingly competitive with proprietary developers after lagging in the past by 12-18 months. However, DeepSeek and others⁴ appear to be closing this gap.

We expect computing demand to continue to surge, which may be good for semiconductor designers and manufacturers. AI adopters, however, may prefer open-source models over some proprietary offerings.

Elevated valuations in mega-cap US tech stocks have left them exposed to shifts in sentiment around the theme. Meanwhile, Chinese tech stocks have received a boost in sentiment since the DeepSeek news. Our team will continue to monitor DeepSeek as it prepares to become a main character in the ongoing AI theme story.

—

Originally Posted March 3, 2025

What’s next for artificial intelligence after DeepSeek? by Invesco US

Footnotes

• Compute comparison based on floating point operations (FLOPs) reported by OpenAI and DeepSeek. Performance based on elo ranking across all tasks reported by LMsys Chatbot Arena. Sources: Bloomberg and Invesco, as of Feb. 14, 2025.
• Sources: Bloomberg and Invesco, as of Feb. 14, 2025.
• Sources: Bloomberg and Invesco, as of Feb. 14, 2025. Based on the MSCI China Information Technology Index.
• Others include Llama from Meta and some models from Mistral, which are offered open source.

Important information

NA4271478

All investing involves risk, including the risk of loss.

Past performance does not guarantee future results.

Investments cannot be made directly in an index.

This does not constitute a recommendation of any investment strategy or product for a particular investor. Investors should consult a financial professional before making any investment decisions.

Fixed income investments are subject to credit risk of the issuer and the effects of changing interest rates. Interest rate risk refers to the risk that bond prices generally fall as interest rates rise and vice versa. An issuer may be unable to meet interest and/or principal payments, thereby causing its instruments to decrease in value and lowering the issuer’s credit rating.

A coupon is the annual interest rate paid on a bond, expressed as a percentage of the face value.

Disinflation, a slowing in the rate of price inflation, describes instances when the inflation rate has reduced marginally over the short term.

Monetary easing refers to the lowering of interest rates and deposit ratios by central banks.

In general, stock values fluctuate, sometimes widely, in response to activities specific to the company as well as general market, economic and political conditions.

The risks of investing in securities of foreign issuers, including emerging market issuers, can include fluctuations in foreign currencies, political and economic instability, and foreign taxation issues.

The federal funds rate is the rate at which banks lend balances to each other overnight.

The Federal Open Market Committee (FOMC) is a committee of the Federal Reserve Board that meets regularly to set monetary policy, including the interest rates that are charged to banks.

Inflation is the rate at which the general price level for goods and services is increasing.

Personal consumption expenditures (PCE), or the PCE Index, measures price changes in consumer goods and services. Expenditures included in the index are actual US household expenditures. Core PCE excludes food and energy prices.

The S&P 500® Index is an unmanaged index considered representative of the US stock market.

The NASDAQ Composite Index is the market-capitalization-weighted index of approximately 3,000 common equities listed on the Nasdaq stock exchange.

A policy rate is the rate used by central banks to implement or signal their monetary policy stance.

Stagflation is an economic condition marked by a combination of slow economic growth and rising prices.

The Survey of Consumers is a monthly telephone survey conducted by the University of Michigan that provides indexes of consumer sentiment and inflation expectations.

A trade deficit is an economic measure of international trade in which a country’s imports exceed its exports. Therefore, a trade deficit represents an outflow of domestic currency to foreign markets.

The opinions referenced above are those of the author as of Feb. 24, 2025. These comments should not be construed as recommendations, but as an illustration of broader themes. Forward-looking statements are not guarantees of future results. They involve risks, uncertainties and assumptions; there can be no assurance that actual results will not differ materially from expectations.