--- title: CoreWeave's A100 contracts complicate the GPU... description: CoreWeave signed a multi-year, take-or-pay fixed-price contract for NVIDIA A100 chips running through 2029, with customers specifically requesting the now... canonical: https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: CoreWeave's A100 contracts complicate the GPU depreciation debate | daily.dev og:description: CoreWeave signed a multi-year, take-or-pay fixed-price contract for NVIDIA A100 chips running through 2029, with customers specifically requesting the now... og:url: https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj og:image: https://api.daily.dev/og/posts/2M7VOGjaj.png og:image:alt: CoreWeave's A100 contracts complicate the GPU depreciation debate og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # CoreWeave's A100 contracts complicate the GPU depreciation debate **[Collections](https://daily.dev/sources/collections)** · 2 min read · 0 upvotes · 0 comments ## Summary CoreWeave signed a multi-year, take-or-pay fixed-price contract for NVIDIA A100 chips running through 2029, with customers specifically requesting the now 9-year-old Ampere architecture for certain workloads rather than accepting it as a fallback. Founder Brannin McBee says A100 pricing has held steady since early 2025 and argues the chip remains the best fit for specific workload sizes rather than newer GPUs always being superior. CoreWeave uses a 6-year useful-life estimate for GPUs, raised from 5 years in 2023 due to hardware, software, and data center improvements, and notes that older fleets often shift into managed inference work after initial contracts expire rather than sitting idle. ## Content There's a narrative in AI infrastructure that goes something like this: GPUs are perishable goods, useful for maybe 2-3 years before the next NVIDIA generation makes them obsolete. CoreWeave's latest numbers don't really support that story. The company just signed a multi-year contract for A100 chips that runs through 2029. NVIDIA launched the A100 back in 2020, which means by the time this contract ends, customers will have been actively requesting a 9-year-old chip. Not settling for it because nothing else was available - specifically asking for it. CoreWeave founder Brannin McBee put it plainly on a recent show appearance: > "We still have clients coming in, asking specifically for that SKU. It's not like they're coming in asking for Hopper, and it's like, 'Oh, we only have Ampere,' or they're asking for Blackwell, and it's like, 'Oh, we only have Ampere.' They are saying, 'No, we want Ampere for workloads,' because that is the most performant platform for their workload." That's the part that catches me off guard. This isn't a case of customers taking older hardware as a fallback - they're picking it because it fits their specific workload best. Turns out "most powerful chip available" and "right chip for the job" aren't always the same thing. **The pricing angle is just as interesting.** McBee noted that A100 pricing has held steady since early 2025, and this new deal is a take-or-pay, fixed-price contract. That's not the pattern you'd expect from equipment supposedly circling the drain toward obsolescence. On the useful-life question, CoreWeave has been consistent: 6 years. The company actually raised this estimate from 5 years back in 2023, citing improvements in hardware, software, and data center design. McBee argues even that might be conservative: > "There isn't one GPU to rule them all. It's just this matrix of different sizes of workloads relative to different sizes of GPUs." CoreWeave also mentioned that once initial contracts on older GPU fleets expire, that hardware doesn't just sit idle - it often shifts into managed inference work, squeezing more paid use out of equipment that's already installed and powered. Given how much capital and energy infrastructure goes into standing up a GPU cluster, that's a meaningful difference in the economics. There's a related point worth raising, made separately by Adam (thdxr): the ## Questions this post answers ### How long is CoreWeave's new contract for NVIDIA A100 GPUs? CoreWeave signed a multi-year, take-or-pay, fixed-price contract for A100 chips that runs through 2029. Since NVIDIA launched the A100 in 2020, customers under this contract will be actively requesting a 9-year-old chip by the time it ends, rather than settling for it due to lack of alternatives. _Anyone weighing GPU procurement timelines can track infrastructure economics debates like this one on daily.dev._ ### What useful-life estimate does CoreWeave use for its GPU fleet? CoreWeave estimates a 6-year useful life for its GPUs, an estimate it raised from 5 years back in 2023 citing improvements in hardware, software, and data center design. Founder Brannin McBee suggests even 6 years might be conservative, arguing there's no single best GPU since workload sizes vary against different GPU sizes. _Teams modeling GPU depreciation and total cost of ownership can follow this kind of infrastructure analysis on daily.dev._ ### Why would customers request older NVIDIA A100 GPUs instead of newer Hopper or Blackwell chips? Customers specifically request A100 GPUs because the chip is the most performant platform for their particular workload, not because newer chips are unavailable. CoreWeave founder Brannin McBee noted clients explicitly ask for Ampere-generation A100s over Hopper or Blackwell, since matching workload size to GPU size matters more than raw chip generation. _Engineers choosing between GPU generations for specific workloads can dig into cases like this on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 4 discussions and 66 comments across x (as of 2026-08-14). **TL;DR:** Reactions are split between people finding it plausible that A100s still hold value for specific workloads and skeptics who see the long-term GPU commitments as a financing/duration-mismatch problem or hype-driven pricing rather than genuine scarcity. **Sentiment:** 25% positive · 45% mixed · 30% skeptical **The case for** - Some argue matching workloads to older, cheaper hardware is smart infrastructure strategy rather than always chasing the newest chip. - A few note that keeping older GPUs in service could help bring inference costs down over time. **The pushback** - Several frame multi-year commitments as really a financing/duration-mismatch problem, not a true supply shortage. - Some suspect pricing pressure is being used as an excuse to inflate prices rather than reflecting real scarcity. - Skeptics question whether locking into 3-5 year contracts makes sense given how fast workloads and hardware needs change. - One person disputes that older architectures remain competitive, pointing out most inference has already shifted to newer generations. **By community** - x (mixed): Replies range from framing the A100 deal as smart, durable infrastructure economics to dismissing multi-year commitments as a symptom of financing rigidity or artificial scarcity, with no single view dominating. **Hottest debate:** Whether the real GPU 'shortage' is a genuine capacity constraint or actually a flexibility/financing mismatch caused by long-term contract commitments. **Open questions** - What is the actual ratio of committed vs. uncommitted GPU capacity across the industry, and how hedged is that against a potential bubble? - Which specific A100 workloads still outperform newer architectures like Blackwell, and at what utilization rates are the 2029 contracts priced? - Would more flexible, shorter-term or resellable contracts fix the perceived shortage better than long-term lock-ins? **Highlights** > @thdxr coreweave, lambda, and the rest of the neocloud layer exist to eat that 3-5 year commitment risk and resell it back out by the minute. someone still has to hold the term, it just isn't you anymore. > — [JensHonack on x](https://x.com/JensHonack/status/2088162266348958194) > @thdxr the 3-5 year commit is the tell that it is a financing problem, not a supply one. the silicon exists, but nobody wants to lend against a depreciating asset on a spot basis. the shortage is really a duration mismatch between how fast the chips age and how long the contract locks > — [rusabuilds on x](https://x.com/rusabuilds/status/2087951888599957700) > @tbpn @branninmcbee Which A100 workloads still beat Blackwell for those clients, and what utilization rate are those 2029 contracts priced at? > — [cozytomcat on x](https://x.com/cozytomcat/status/2088064399923638494) > @blankfella @thdxr You realise most of the inference currently runs on Blackwell and companies are already stocking Vera Rubin, H100s are basically legacy at this point. We are not talking about consumer cards. > — [MichaelArnaldi on x](https://x.com/MichaelArnaldi/status/2088191823931904168) **Source threads** - [x](https://x.com/rohanpaul_ai/status/2088000292692595009) · 0 points · 0 comments - [x](https://x.com/rohanpaul_ai/status/2087815879652483122) · 0 points · 10 comments - [x](https://x.com/thdxr/status/2087919860919513099) · 0 points · 47 comments - [x](https://x.com/tbpn/status/2088039416279535640) · 0 points · 9 comments ## Similar posts on daily.dev - [CoreWeave signs multi-year Anthropic deal as nine of ten top AI model providers join its platform](https://daily.dev/posts/coreweave-signs-multi-year-anthropic-deal-as-nine-of-ten-top-ai-model-providers-join-its-platform-znfy3mszu) · The Next Web · 0 upvotes · 0 comments - [Nvidia invests $2B to help debt-ridden CoreWeave add 5GW of AI compute](https://daily.dev/posts/nvidia-invests-2b-to-help-debt-ridden-coreweave-add-5gw-of-ai-compute-suyuzcb2u) · TechCrunch · 0 upvotes · 0 comments --- Tags: [#cloud](https://daily.dev/tags/cloud), [#gpu](https://daily.dev/tags/gpu), [#ai-infrastructure](https://daily.dev/tags/ai-infrastructure) [View this post on daily.dev](https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"CoreWeave's A100 contracts complicate the GPU depreciation debate","url":"https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj"},"datePublished":"2026-08-13T23:06:32.296Z","dateModified":"2026-08-14T11:07:43.525Z","description":"CoreWeave signed a multi-year, take-or-pay fixed-price contract for NVIDIA A100 chips running through 2029, with customers specifically requesting the now...","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"cloud,gpu,ai-infrastructure","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"CoreWeave's A100 contracts complicate the GPU depreciation debate"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/coreweave-s-a100-contracts-complicate-the-gpu-depreciation-debate-2m7vogjaj#faq","mainEntity":[{"@type":"Question","name":"How long is CoreWeave's new contract for NVIDIA A100 GPUs?","acceptedAnswer":{"@type":"Answer","text":"CoreWeave signed a multi-year, take-or-pay, fixed-price contract for A100 chips that runs through 2029. Since NVIDIA launched the A100 in 2020, customers under this contract will be actively requesting a 9-year-old chip by the time it ends, rather than settling for it due to lack of alternatives. Anyone weighing GPU procurement timelines can track infrastructure economics debates like this one on daily.dev."}},{"@type":"Question","name":"What useful-life estimate does CoreWeave use for its GPU fleet?","acceptedAnswer":{"@type":"Answer","text":"CoreWeave estimates a 6-year useful life for its GPUs, an estimate it raised from 5 years back in 2023 citing improvements in hardware, software, and data center design. Founder Brannin McBee suggests even 6 years might be conservative, arguing there's no single best GPU since workload sizes vary against different GPU sizes. Teams modeling GPU depreciation and total cost of ownership can follow this kind of infrastructure analysis on daily.dev."}},{"@type":"Question","name":"Why would customers request older NVIDIA A100 GPUs instead of newer Hopper or Blackwell chips?","acceptedAnswer":{"@type":"Answer","text":"Customers specifically request A100 GPUs because the chip is the most performant platform for their particular workload, not because newer chips are unavailable. CoreWeave founder Brannin McBee noted clients explicitly ask for Ampere-generation A100s over Hopper or Blackwell, since matching workload size to GPU size matters more than raw chip generation. Engineers choosing between GPU generations for specific workloads can dig into cases like this on daily.dev."}}]} ```