DeepSeek's Low-Cost Model Reshapes AI Landscape: Experts Predict Shift To 'Application Layer'

The artificial intelligence boom that has powered tech stocks for over a year is facing fresh scrutiny after DeepSeek, a Chinese AI firm, unveiled a model that could challenge the capital-intensive spending narrative driving the U.S. AI industry.

Investors are questioning whether the staggering sums poured into AI infrastructure, exceeding $500 billion in recent announcements from major U.S. tech firms, will effectively generate the anticipated returns.

According to a note from Goldman Sachs shared Tuesday, analyst Eric Sheridan said DeepSeek's R1 model has triggered market concerns over "levels of spend, returns on spend, and the sustainability of current trend lines" in AI investment.

AI Stocks Face Pressure

Investors are now grappling with a critical question: If effective AI models can be built at significantly lower costs, does this diminish the competitive edge of firms investing heavily in GPUs, networking equipment, and data centers?

The uncertainty has weighed on semiconductor stocks, which saw a knee-jerk selloff on Monday following the news.

DeepSeek's R1 model, which builds on open-source architectures like Meta Platforms Inc.'s META Llama and Alibaba Group Holding Ltd.'s BABA Qwen, has demonstrated strong reasoning capabilities while reportedly costing just $5.6 million to train.

By contrast, some of the most advanced AI models from U.S. companies require billions in computing power and proprietary datasets to develop.

Sheridan said, "DeepSeek's reported training cost likely does not represent the fully loaded expense of developing its models," as the firm appears to have leveraged existing open-source frameworks built on significantly higher infrastructure spending.

He added that the lack of transparency around DeepSeek's hardware and data sources adds another layer of complexity for investors.

Nonetheless, the emergence of a competitive model at lower cost has raised the question of whether companies like Nvidia Corp. NVDA, Advanced Micro Devices Inc. AMD, and Broadcom Inc. AVGO can sustain their current AI-driven valuation multiples.

"The market reaction is currently being driven by questions around the collective capital expenditures needed for the AI theme to scale," Sheridan said.

A Shift From Infrastructure To Applications

Goldman Sachs' analysis suggests that AI investments may soon shift from the infrastructure layer—dominated by semiconductor and cloud providers—to the “application layer,” where consumer and enterprise AI adoption could accelerate.

Companies developing AI-powered assistants, automation tools, and enterprise solutions may see a greater share of AI-related revenues in the future.

"The next level of evolution in the AI theme is likely to shift from infrastructure to applications, as AI agents, enterprise use cases, and rising consumer utility take center stage,” Sheridan said.

The note also highlights that AI assistants, such as Alphabet Inc.’s GOOGL Gemini and OpenAI’s ChatGPT, are likely to play a central role in consumer interactions with AI over the next 12 to 18 months.

The rising capabilities of these tools—such as booking reservations and browsing the web—are expected to shape how users engage with AI-powered services.

"DeepSeek's innovations reinforce our thesis that the application and platform layers will benefit as revenue moves away from the infrastructure layer,” Sheridan added.

What's Next for AI Investors?

The debate over AI spending efficiency is far from over, and the next major catalyst will likely come during fourth-quarter earnings season. Investors will be looking for tech executives to address whether AI investments are generating meaningful returns and how competition from lower-cost AI models may influence future spending.

Sheridan cautioned that the uncertainty surrounding DeepSeek's cost advantages may weigh on valuations for AI-related stocks until greater clarity emerges.

For now, the sector remains at an inflection point with DeepSeek challenging the status quo. The next phase may be defined not just by who spends the most, but by who spends the smartest.

Read now: