DeepSeek has raised a hard question for the AI industry: if powerful models can be built or run more efficiently, will the biggest technology companies still need to spend aggressively on chips and data centers?
For Meta, the answer is still yes. Mark Zuckerberg used Meta's first-quarter earnings call on Wednesday to make clear that the company is not stepping away from its long-term AI infrastructure plans.
Markets reacted first, but Meta stayed the course
The immediate backdrop was a sharp market reaction. U.S. markets panicked on Monday over speculation that DeepSeek's AI models would reduce demand for GPUs. Nvidia's stock dropped almost 20% as investors weighed what that could mean for the companies supplying the hardware behind the AI boom.
Meta did not respond by signaling a retreat. Instead, Zuckerberg said the company would invest "very heavily" in AI over the long term. He also said that could mean "hundreds of billions of dollars" over time.
That message builds on a spending plan Zuckerberg had already announced last week. Meta expects to spend more than $60 billion in 2025 on capital expenditures, with that money going primarily toward data centers.
The key point is that Meta sees infrastructure not only as a cost, but as a way to compete. Data centers, chips and the ability to run large-scale AI services are being treated as part of the company's core strategy.
DeepSeek is a competitor, not a reason to stop spending
Zuckerberg did not dismiss DeepSeek as irrelevant. According to the source, Meta considers DeepSeek a new competitor and is learning from it. But he argued that it is "way too early" to decide that demand for chips will stop rising.
His reasoning centers on inference, the process of using AI models after they have been trained. Even if AI development changes, Meta still needs to serve AI features to billions of users. That scale keeps chips important in Zuckerberg's view.
In response to an analyst's question about DeepSeek's possible effect on Meta's AI spending, Zuckerberg said that heavy spending on AI infrastructure would continue to be a "strategic advantage" for the company.
He put the bet plainly: "At this point, I would bet that the ability to build out that kind of infrastructure is going to be a major advantage for both the quality of the service and being able to serve the scale that we want to," Zuckerberg said.
That statement shows how Meta is framing the issue. The company is not only chasing better models. It is also trying to make sure it can deliver those models widely and reliably across its products.
Llama 4 raises the stakes for open models
Meta's AI strategy is also tied to its Llama models. Zuckerberg said the company's goal for its next model, Llama 4, is to make it the world's most competitive, even against closed models such as ChatGPT.
That is a shift in ambition from the previous generation. Zuckerberg said, "Our goal with Llama 3 was to make open source competitive with closed models," and added, "And our goal for Llama 4 is to lead."
The source also says Zuckerberg expects Llama 4 to have agentic capabilities, an area OpenAI and Anthropic have moved into, along with multimodal capabilities. In plain terms, Meta wants its next model to compete not just on raw output, but on the kinds of tasks and inputs that advanced AI systems are increasingly expected to handle.
That makes the infrastructure question more central. If Meta wants Llama 4 to lead, and if it wants to serve AI tools at the scale of billions of users, then cutting back on the systems that power those services would conflict with the company's stated direction.
What the spending signals
Meta's position after DeepSeek is not that efficiency does not matter. The company is watching DeepSeek, treating it as competition and learning from it. But the company is also saying that efficiency does not automatically erase the need for scale.
The spending plan signals several priorities:
- AI infrastructure remains central. Meta expects data centers and chips to support the quality and reach of its services.
- Inference demand matters. Zuckerberg pointed to billions of users as a reason chips remain crucial.
- Llama 4 is a major target. Meta wants the next model to lead, including against closed models.
- DeepSeek has not changed the long-term bet. Meta sees a new competitor, but not a reason to reverse course.
The broader implication is that Meta views AI as a scale contest as much as a model contest. Better techniques may change how companies build and run AI, but Meta's public stance is that large infrastructure will still help determine who can deliver the strongest services to the most users.
For now, DeepSeek has intensified the debate over AI spending. Meta's answer is to keep building.