Two Business Models for AI Compute Leasing: What CoreWeave and Applied Digital's Revenue Gap Tells Us
As AI compute demand surges, the business models and profitability of GPU cloud providers are under scrutiny. This article breaks down the underlying logic of AI infrastructure investment by comparing the revenues of CoreWeave and Applied Digital.
What Happened?
Recently, the revenue performance of two U.S. AI infrastructure companies—CoreWeave and Applied Digital—has drawn market attention. According to a report by The Motley Fool, both companies lease GPU compute power, yet their revenue scales have diverged significantly. This gap reflects two distinct business models in the AI compute market.
CoreWeave operates an asset-light model. Rather than building its own data centers from scratch, it leases space and deploys GPU clusters in partnership with existing facilities. This approach allows rapid scaling and flexibility, though it creates heavy dependence on upstream chip suppliers (primarily NVIDIA) and may limit profit margins.
Applied Digital, by contrast, is a more traditional compute infrastructure company. It started in cryptocurrency mining and later pivoted to AI compute provision. Its model is capital-intensive: it builds its own data centers, purchases hardware, and constructs networking infrastructure. This requires longer investment cycles and slower returns.
In terms of revenue, CoreWeave has grown markedly faster than Applied Digital. This reflects a "fast fish eats slow fish" dynamic in the AI compute market—speed to capacity matters.
How Does This Affect You?
You might wonder: what does the revenue gap between these two companies have to do with me?
In fact, the evolution of the AI compute leasing market is quietly changing how we use AI services. The AI assistants, image-generation tools, and autonomous driving systems you interact with all require massive compute resources behind the scenes. Much of that compute comes from GPU cloud providers like CoreWeave.
The supply efficiency of AI compute directly determines how quickly you can access smarter, cheaper AI products.
Consider these examples:
- If you're a startup founder building an AI app but can't afford expensive GPUs, you can rent compute from a provider like CoreWeave on a pay-as-you-go basis—lower cost, faster time to market.
- If you're an everyday user noticing that your AI assistant responds faster or costs less, that improvement may stem from gains in compute leasing market efficiency.
Viewed from another angle, the competitive landscape of AI compute leasing is shaping how quickly AI products spread and how good the user experience is.
How Should Everyday People Think About This?
Given the rapid development of the AI compute leasing market, here are a few ways to understand and respond:
1. Understand the "Compute-as-a-Service" Trend
In the past, companies that wanted to run AI models had to buy their own servers and build data centers. Today, more and more companies choose to "rent compute," paying on demand much like a utility bill. This model lowers the barrier to AI adoption and enables broader participation in AI innovation.
2. Pay Attention to the Investment Logic Behind AI Infrastructure
If you follow investing, note this shift: the AI compute market is moving from a "parameter race" to a "deployment race." Investors at major AI conferences are now more focused on inference cost, mass production timelines, and time-to-market—not just how large a model's parameter count is.
This means the AI companies that will be profitable in the future are not necessarily those with the most powerful models, but those that can drive compute costs down and bring products to market fastest.
3. Beware of Overhyped "Compute Anxiety"
While AI compute demand is growing, that doesn't mean a persistent "compute shortage" is inevitable. For example, when Chinese AI startup Moonshot AI (maker of the Kimi chatbot) temporarily paused new consumer subscriptions due to surging user demand, it simultaneously announced it was accelerating capacity expansion at full speed.
This illustrates that compute shortages are a phase, not a permanent bottleneck. As more capital and technology flow into the sector, supply will gradually catch up with demand.
Deep Dive: The "Light" vs. "Heavy" Approaches to AI Compute Leasing
Comparing CoreWeave and Applied Digital reveals two typical paths in the AI compute market:
- Asset-light model (e.g., CoreWeave): Does not build data centers; instead leases or co-deploys GPU clusters to respond quickly to customer needs. Advantages include rapid expansion and high flexibility. Drawbacks include heavy reliance on upstream chip makers and potentially constrained profit margins.
- Asset-heavy model (e.g., Applied Digital): Builds its own data centers, purchases equipment, and constructs networking. Requires large upfront investment and long lead times, but once operational, marginal costs are lower and long-term profitability is stronger.
This means future competition in the AI compute market will be about more than technology—it will also be a contest of capital efficiency and operational execution.
A Unique Angle: The "Second-Order Effects" of AI Compute Leasing
The competition in AI compute leasing isn't just about who rents the most or the fastest. It also produces a series of second-order effects:
- Lower AI application costs: As compute supply increases, the inference cost of AI applications will gradually decline, making AI services affordable for more small and mid-size businesses and developers.
- Shift in AI product forms: When compute is no longer the bottleneck, AI products may shift from "large general-purpose models" toward "smaller models tailored to specific use cases," prioritizing practical deployment and efficiency.
- Industry landscape reshaping: In the future, whoever masters efficient compute scheduling and optimization will hold a stronger position in the AI value chain.
Broader Perspective: From "Training Models" to "Shipping Products"
The investment logic around AI is shifting: practical applications and deployment are replacing raw model parameters as the core basis for project valuation.
This is closely tied to the development of the AI compute leasing market. Over the past two years, large-model companies competed on parameter scale and fundraising records. Now, the market cares more about inference cost, mass production and delivery, and commercialization capability.
In other words, the AI industry is shifting from "technology-driven" to "business-driven."
If compute supply can keep pace with demand, AI applications will enter a period of rapid growth. But if compute costs remain stubbornly high, many AI projects may struggle to turn a profit, and the industry will enter a consolidation phase.
Points Worth Watching
While the AI compute leasing market has broad prospects, there are also areas that warrant caution:
- Don't over-interpret "compute shortages": Supply gaps are transitional. As capital and technology enter the space, supply will gradually catch up.
- Beware of "compute anxiety" being used as a marketing tool: Some companies may exploit "tight compute supply" narratives to raise prices or create a sense of scarcity. Stay rational.
- Invest with caution: The AI compute leasing market is still in its early stages, and business models are not yet fully mature. When evaluating related companies, focus on cash flow and profitability—not just revenue growth.
The above analysis is for informational purposes only and does not constitute professional investment advice.
Summary & Discussion
One-sentence takeaway worth sharing:
The AI compute leasing market is shifting from a "parameter race" to a "deployment race." Whoever can make compute cheaper and more efficient will win the future.
Question for discussion:
Have you experienced slow response times or high prices when using AI products? What do you think is the biggest bottleneck for wider AI adoption—compute, data, or finding the right use cases? Share your thoughts in the comments.