A belated slap in the face is still a slap; A100 provides a perfect logic for Neocloud

Last night's CoreWeave (a representative new cloud infrastructure company in the U.S., Neocloud) conference call was truly impressive. The most striking part wasn't the CFO's comment that “we are seeing demand, pricing, and margin all expanding,” nor the $129 billion RPO accumulated (more than 3.3 times the annual CapEx), but this statement:

“We recently signed an A100 contract that extends into 2029 at an attractive price.” For example, we recently signed an A100 contract with a term extended to 2029 and the price is very compelling.

This aligns perfectly with the open letter Jensen just posted on X, where he said the A100 has an economic lifespan close to 10 years. We all know the A100 was launched in 2020—a GPU released six years ago can, in 2026, be signed to a new contract lasting until 2029, and at an “attractive” price.

This hits the core issue of the Neocloud business model squarely.

In its 2025 10-K, CoreWeave explicitly disclosed that the depreciation period for GPUs and related technological equipment is six years. This depreciation convention has been heavily questioned by the famous short-seller Michael Burry, and was Michael Burry’s key reason for being bearish on AI in November 2025.

Burry believes the true economic life of a GPU is only about 2–3 years, while Hyperscalers depreciate equipment over 5–6 years, leading to a serious underestimation of depreciation expenses and overestimation of cloud business profits.

This skepticism itself is not absurd (which is why the market was so shaken at the time). If a GPU really only delivers 2–3 years of economic value, Neocloud’s high leverage would naturally be very risky:

Borrow money to buy GPUs → GPU economic value declines rapidly → asset residual value approaches zero → before the debt is repaid, the asset is already unprofitable.

Now, CoreWeave has provided a very direct counter-example, slapping Michael Burry in the face. The 2020 A100 completes its six-year depreciation cycle in 2026, but a new contract can extend to 2029, with its economic value even still increasing.

CoreWeave explicitly stated in the call that after old GPU rental contracts expire, they can be repurposed for inference and other workloads to continue maximizing asset value.

Then, CoreWeave made a crucial statement:

“So when an initial contract ends, the cluster no longer has any leverage, and we are free to recontract that cloud infrastructure or offer it to the market.” So, when the initial contract ends, the cluster no longer carries any leverage, and we can freely recontract these cloud infrastructure assets or offer them to the market.

In plain terms, this means:

When the initial contract ends, the financing obligations associated with this Cluster (GPU cluster) are usually already covered or repaid by the revenue from the initial contract. Therefore, CoreWeave can relend this now highly deleveraged asset, and subsequent income no longer carries the heavy burden of original capital recovery.

This is exactly the most remarkable aspect of the Neocloud business model.

You can view a GPU Cluster as having two lifecycle stages: the first phase is the Initial Contract, where the customer signs a relatively long-term agreement. Rental income in this period mainly covers the leveraged portion of CapEx. The second phase is entirely different—after the contract expires, the GPU is not scrapped but continues to be rented out. Regarding this, CoreWeave specifically mentioned:

“In a market where new capacity is supply-constrained and costs are rising, AI cloud infrastructure in production is a scarce, valuable asset…increasingly we are seeing longer utilization at higher prices, offering the potential for significant further upside.” In a market where new capacity is constrained and costs are rising, AI cloud infrastructure in production is a scarce and valuable asset…we are increasingly seeing longer utilization at higher prices, offering significant further upside potential.

Simply put, this means:

After the initial contract ends, new contracts can be signed at significantly higher prices.

This explains why CoreWeave’s business model—that is, the Neocloud business model—looks much better than the market imagines. Neocloud’s most interesting aspect isn't just how much profit the GPU's first contract brings, but how much and for how many years the GPU can continue to generate profit after it is deleveraged for the first time.

Judging from the A100, a product released six years ago (in 2020, when the world was still in a pandemic and AI training was virtually unheard of), the GPU cluster's “second lifecycle” is obviously very long. Thus, Neocloud’s real ROIC is unimaginable.

Clearly, Musk, who plans to quadruple capacity and build 8GW data centers by 2027, understands this well. SpaceX is almost a Neocloud company already (note, this is not mocking Elon Musk).