This week, Big Tech’s biggest players—Apple, Meta, and Microsoft—released their earnings, and in the middle of the DeepSeek AI drama, we got a rare glimpse into how trillion-dollar companies are thinking about AI’s future. Their key takeaways? AI isn’t slowing down, compute demand is shifting, and the real winners aren’t necessarily the best models but the best business models.
1. DeepSeek Is a Serious Contender
Each CEO acknowledged DeepSeek’s breakthroughs, signaling that even major Western AI players see its potential impact. Mark Zuckerberg put it plainly—AI follows a predictable cycle: proprietary breakthroughs become open-source, adoption increases, costs collapse, and AI becomes more accessible. Whether it’s a Chinese model like DeepSeek or a homegrown one, AI development follows this same pattern.
2. Compute Is Moving from Training to Inference
A common misconception after DeepSeek’s announcement was that AI compute demands would decline due to more efficient training. However, even before DeepSeek, a shift was underway: compute resources are moving from training to inference. Mark Zuckerberg noted, “This doesn’t mean you need less compute. Instead, more compute at inference time means higher intelligence output.” AI isn’t getting cheaper to run; it’s just changing where and how compute power is applied.
3. Capital Expenditure Is Still Rising
Big Tech isn’t scaling back AI investment—they’re doubling down. If AI costs drop, usage rises, and demand increases. Satya Nadella made a historical comparison, likening AI to cloud computing—when servers became cheaper, they didn’t disappear; instead, they became ubiquitous. Similarly, AI’s efficiency gains mean more applications, not fewer investments. Expect more billion-dollar data centers packed with GPUs.
4. AI Models Are Commodities—Distribution Is the Moat
Each company emphasized the same idea in different ways: AI models are interchangeable, but distribution is everything.
- Meta: 3.3 billion daily users.
- Apple: 2.35 billion active devices.
- Microsoft: Owns enterprise software and cloud infrastructure. DeepSeek, OpenAI, and Anthropic might develop top-tier AI, but without control over distribution, they remain vendors. Microsoft integrating DeepSeek R1 into its model catalog shows its intent to aggregate intelligence rather than just build it.
5. Business Model Resilience Is the Real Advantage
Mark Zuckerberg emphasized that Meta’s AI edge isn’t just about its models but how AI strengthens its core business. AI-driven ad targeting enhances revenue without increasing costs—a win for investors. Apple, meanwhile, takes a more calculated approach, waiting for AI costs to rationalize before integrating AI at scale, ensuring maximum profit margins.
The Future of AI: It’s About Business, Not Just Models
AI breakthroughs will keep happening, and costs will continue falling. But as these earnings calls reveal, the real winners won’t be the companies with the best AI models—they’ll be the ones with the strongest business models. In the end, AI is just another tool. The companies that own the data, the distribution, and the customers will be the ones reaping the rewards.
