Asset remarketing isn’t exclusive to mobile technology. Any IT hardware can be resold, including AI data centers. In fact, retired AI accelerators hold real secondary-market value, with H100-class units reported in the $15K to $20K range. The challenge is capturing adequate resale value in hardware that becomes obsolete so quickly. This blog post breaks down the rapid evolution of AI data center hardware, how to capture meaningful secondary market value, and how grading and timing can help move the numbers in your favor.
The Rapid Growth of AI in Data Center Hardware
The Artificial Intelligence wave hit hard with one of the fastest tech adoption curves witnessed in modern history. The spike occurred in late 2022, just as the peak of COVID-19 began to crest, and global media usage and digital screen time reached an all-time high. Now, AI is heavily utilized every day for tasks ranging from daily problems to complex business projects. Tools like ChatGPT and Claude scaled quickly and are used for a multitude of different needs. Search engines such as Google, Microsoft, and Yahoo utilize AI to process results and queries, and many websites have used chatbots to help with initial query navigation.
AI has been embraced globally, but this massive wave of usage has also led to a surge in data centers. AI data centers are popping up all over the world with mixed reception. These updated servers are different from traditional models and require a completely different cooling approach.
The Cost of AI Servers Vs. Traditional Models
The core difference between a data center and an AI data center is that an AI data center runs AI software. However, AI has completely changed how data centers are built, creating new hardware differences in power and cooling.
First, traditional data centers rely heavily on central processing units (CPUs) and are built to handle serial processing. This means one complex task after another, whereas AI data centers are built for parallel processing–millions of micro-calculations at once–and rely on GPUs like the Nvidia H100. GPU speed is specifically for training AI models and requires much more power.
A traditional data center typically requires 5 to 10 kilowatts (kW) of power per server rack. Still, AI data centers require 40 to 100+ kW per server rack because AI chips pull significant amounts of electricity for data processing. As a result, this requires a completely different cooling system.
Traditionally, data centers use massive fans and air conditioning units to keep hardware cool. Data centers running AI software generate more heat and require liquid cooling–the process of running specialized liquid directly over the chips.
Lastly, traditional data centers use standard fiber-optic and Ethernet connections for standard internet traffic, but think about how fast generative AI tools like ChatGPT respond. AI data centers require ultra-fast, ultra-low-latency networks that enable thousands of GPUs to communicate instantly while training a single AI model.
Utilize Retired Hardware to Fund AI IT Refresh Cycles
These differences are no small tweak in the system. Replacing regular data centers with AI models requires complex facility remodeling and rearranging to incorporate liquid cooling. Coupled with installation costs and constant upgrades to avoid obsolete hardware, staying on the AI train can become costly very fast.
AI is becoming an industry norm, but the problem lies in rapid aging. Data center hardware vendors are shifting from a multi-year release model to a 12-24 month cycle. This frequent hardware redesign cycle compresses redesign loops, forcing rapid compute adjustments. This forces companies into more rapid refresh cycles, and budget funds can shrink quickly. Fortunately, AI accelerators are just like any other piece of IT hardware and can be resold.
Rapid deployment of new GPU models can be stressful to keep up with. Still, companies can take advantage of rapid deployment cycles by reselling retired H100 accelerators to help fund AI IT refresh cycles through value recovery.
Capture Meaningful Residual Value with Well-Timed Resale Windows
Because AI hardware ages quickly, resale timing is critical. Rule number one is to move quickly. Idle hardware means deteriorating value, and the most crucial part is finding the right resale window before value starts to erode. The 90-day value cliff applies to AI hardware just like other IT equipment. Long storage time in an era of rapid tech deployment is a death sentence for resale value and could cost companies millions in residual value.
There are three options for increasing value in retired H100s. First, the secondary market is a goldmine for increasing ROI because there is always someone willing to save money on used hardware. H100s are highly capable and retain much of their original cash value, making them an excellent candidate for resale. Additionally, the influx of AI also means more e-waste, and keeping H100s in circulation through reuse helps keep them out of the waste stream.
Retired H100s can also be repurposed for less demanding tasks that lean away from heavy AI training and more toward standard cloud computing. This extends the lifespan and saves budget funds by utilizing hardware already in storage rather than purchasing chips for basic tasks.
Recycling is always an option for damaged hardware that cannot be repaired or resold. Even obsolete chips can be harvested for copper and other valuable components.
Improve Recovery Numbers with Grading
Value recovery doesn’t start with resale; it begins with grading. Device grading helps determine whether a device needs repair and if it is eligible for resale. Grading involves a rigorous and thorough examination and evaluation of IT assets, including AI data center hardware. Grading determines how valuable an asset is, and often how much the value can be increased. Grading is also a scale by which to measure ROI. Information collected from a thorough grading process can often provide companies with the numbers to help determine whether or not resale is the best option. If retired hardware is in good condition and fairly new, it’s immediately worth more; however, the speed of AI model deployment accelerates obsolescence, and sometimes resale prices are not enough to offset disposition costs. Fortunately, ITAD providers offer the perfect, streamlined roadmap for retired IT assets, even AI data centers.
ITAD Vendors Provide The Perfect Roadmap
With an IT asset disposition provider, the hard part is done. ITAD vendors take care of the heavy lifting, including resale window navigation, logistics, data security, and value recovery.
With more than 30 years of industry experience, HOBI is an industry leader in IT asset management and disposition. Our R2v3, RIOS, ISO 14001, and NAID AAA certifications ensure we are uniquely qualified to handle large-scale needs while protecting data and mitigating environmental liability. HOBI offers data center services, including AI hardware, and can help provide budget funds for frequent IT refresh cycles while extending the lifecycle of used equipment.
Contact HOBI today to request an AI hardware valuation at 877-814-2620 or sales@hobi.com.