CoreWeave's own CFO put a number on it: an "approximately 25% increase across SKUs," implemented in July 2026 and disclosed on the company's Q2 2026 earnings call on August 11 (The Motley Fool's transcript). This CoreWeave price increase 2026 is real, it's company-wide, and if you rent H100 or H200 capacity and haven't seen your bill move, that's not a mistake on your end, it's landing somewhere you probably aren't looking: negotiated contract pricing, not the public rate card. This post breaks down exactly what changed, why the on-demand numbers look untouched, and what it means for a renewal that's coming up.
Quick Answer: The CoreWeave Price Increase 2026, What Changed and When
CoreWeave raised prices roughly 25% across its SKU lineup in July 2026, a change it didn't disclose publicly until its Q2 earnings call on August 11. CFO Nitin Agrawal attributed it to "the current demand environment and the increasing ROI our customers are observing" as workloads shift toward inference, and CEO Michael Intrator confirmed the hike lands hardest on Blackwell and Vera Rubin SKUs. CoreWeave's published on-demand H100 and H200 rates haven't moved, because the increase is showing up in negotiated and reserved contract pricing, which is where most of CoreWeave's revenue actually comes from.
What CoreWeave Actually Raised: SKU-by-SKU Breakdown
The 25% Figure, Straight From the Q2 2026 Earnings Call
The exact line, from Agrawal's prepared remarks: CoreWeave's operating margin improvement "came before our July pricing changes which included an approximately 25% increase across SKUs in response to the current demand environment and the increasing ROI our customers are observing from their investments in the CoreWeave platform as they shift to inference" (earnings call transcript). He added that CoreWeave is "also passing through component price increases," tying part of the hike directly to what CoreWeave itself pays for hardware.
Agrawal explained why the increase is sticking rather than just passing costs through when an analyst pressed him on supply chain factors later in the call: "the increase in the value of output of the CoreWeave cloud has outpaced the value of the input increases" in the supply chain, according to the same transcript. Put plainly, CoreWeave's own read is that demand for what its GPUs produce has grown faster than what it costs CoreWeave to buy and run them, and the July repricing is the company capturing that gap, not just covering higher component costs.
On-Demand Rate Card vs Contract SKU Pricing (Why the Public Numbers Didn't Move)
Here's the piece that trips people up. CoreWeave's self-serve pricing page still lists HGX H100 at $49.24/hr ($6.16/GPU-hr) and HGX H200 at $50.44/hr ($6.31/GPU-hr), the same figures published on the page as of July 5, 2026, according to our own earlier CoreWeave H100 and H200 pricing breakdown. Checking the same page again in August, the numbers are identical. If the company raised prices 25% in July, why does the public page look untouched?
Because the public rate card isn't where most of CoreWeave's dollars flow. Most CoreWeave revenue is sold through multi-year committed contracts negotiated directly with an account team, not through the checkout-style page anyone can load. That structure is exactly why the on-demand numbers can sit still while a real 25% increase moves through the business: the increase is landing in renewal terms, new contract quotes, and reserved-capacity negotiations that never touch the public page. Our deeper look at CoreWeave's contract-first pricing model covers why the rate card functions more like a ceiling than a price most customers actually pay.
Hopper (H100, H200) vs Blackwell/Vera Rubin: Where the Increase Actually Lands
The hike isn't evenly distributed across generations. Intrator was direct about this on the call: "Pricing and margins for our Blackwell and Vera Rubin SKUs are setting new highs, while pricing for prior generation SKUs is at or above where it was years ago." Translation: the steepest pricing power sits with the newest hardware, where demand is least satisfied. Hopper-generation H100 and H200 aren't exempt, but they're described as elevated relative to their own history rather than "setting new highs" the way Blackwell and Vera Rubin are.
That distinction matters if you're renewing an H100 contract specifically. You're not necessarily facing the same pricing pressure a team locking in Vera Rubin capacity is facing, but you're not facing a discount either. Ask directly which tier your renewal falls into before you accept a number.
GPU Cloud Price Increase 2026: Why Now, and Why CoreWeave Isn't Alone
CoreWeave isn't the only GPU cloud provider raising prices in 2026, and understanding why helps explain whether this is a CoreWeave-specific move or a market-wide one. AWS raised its EC2 Capacity Block rates twice this year, roughly 15% in January and another 20% in July, covering P6-B300, P6-B200, P5, P5e, P5en, and P4de, according to our AWS Capacity Blocks price hike breakdown. Two of the largest names in GPU cloud raised reserved-capacity prices in the same window, on the same underlying story: not enough Nvidia hardware for the demand chasing it.
For a fuller picture of where CoreWeave's new numbers sit against the rest of the market, see our GPU cloud pricing comparison across providers, which tracks on-demand rates from AWS, Azure, Lambda Labs, and neoclouds side by side.
Sold-Out Capacity and the Vera Rubin Demand Wall
CoreWeave says it isn't sitting on idle inventory. "Our near-term capacity remains effectively sold out," Intrator told investors. "That is translating into signed commitments on increasingly favorable terms from a broadening set of customers" (reporting via Yahoo Finance). That's not a new theme; on the Q1 2026 call, Agrawal said much the same thing: CoreWeave was "largely sold out of our 2026 capacity, with prices increasing across the board from Ampere to Hopper to Blackwell" (Q1 2026 earnings call transcript). The July SKU increase isn't a one-off; it's the continuation of a pricing trajectory that's been building since at least the start of the year.
Layered on top of raw sold-out capacity is demand for NVIDIA's next platform. Vera Rubin capacity is already commanding, in Intrator's words, "new highs" even ahead of broad general availability, which tells you CoreWeave is fielding forward demand for hardware that isn't fully deployed yet. Our NVIDIA Vera Rubin NVL72 guide has the spec and architecture background if you're evaluating whether to wait for it or lock in Hopper or Blackwell capacity now.
Even 2020-era hardware isn't loosening up. Intrator disclosed that CoreWeave "recently signed an A100 contract that extends into 2029 at an attractive price," a six-year-old GPU generation still commanding multi-year commitments. That's the clearest signal in the whole call that this isn't a Blackwell-only supply story; every generation CoreWeave sells is tight.
Component Costs: HBM and GPU Supply Feeding Into the Hike
Agrawal's line about "passing through component price increases" isn't rhetorical. Industry analysts describe the current shortage as "the most prolonged memory shortage on record," with the bottleneck having shifted from chip fabrication to HBM memory supply; HBM costs rose roughly 30% in Q4 2025 alone, and data center GPU lead times are still running 36-52 weeks against an estimated industry backlog of roughly 3.6 million units as of April 2026 (ValueAdd VC's analysis of the 2026 chip shortage). Nvidia reportedly cut RTX 50-series consumer GPU production 30-40% in the first half of 2026 just to free up memory manufacturing lines for data center chips, a sign of how tight HBM supply is across the whole GPU market, not just at CoreWeave. Our GPU shortage 2026 guide covers the wider supply picture and what teams are doing to secure capacity around it.
That input-cost pressure is one half of the story CoreWeave told investors. The other half, as Agrawal put it earlier on the same call, is that demand for CoreWeave's cloud has grown in value faster than the company's own input costs, meaning it believes it can pass those costs through and still expand margin, not just cover them.
CoreWeave's Backlog and Margin Math Behind the Decision
The numbers back up that margin claim. Agrawal said "recent contracts carry contribution margins 5 to 10 percentage points above those added in previous quarters," direct evidence the price increase is flowing to the bottom line rather than getting absorbed by rising costs. Q2 2026 revenue came in at $2.58 billion, up 112% year over year and beating the $2.56 billion analyst estimate, and the stock jumped more than 14% in after-hours trading on the news (Yahoo Finance).
The backlog is the bigger tell. CoreWeave's revenue backlog hit $104.2 billion at the end of Q2, up 246% year over year, with roughly $25 billion in new commitments signed early in Q3 alone. Full-year 2026 revenue guidance was raised to $12.4-13.2 billion, and Q3 guidance came in at $3.45-3.6 billion (Investing.com's Q2 earnings analysis). Managed inference, a newer and higher-margin business line, grew its ARR from $1 million to more than $100 million in a single quarter, with a $250 million-plus target by year-end. A company with that kind of pipeline and that kind of margin expansion on new contracts has very little incentive to negotiate a 25% increase back down.
CoreWeave's Two Price Lists: Current vs Legacy Classic Page
There's a second pricing wrinkle worth flagging before you compare numbers you find online. CoreWeave still runs a legacy "Classic" pricing page alongside its current one, and the two list different numbers for nominally the same hardware. The Classic page prices NVIDIA H100 PCIe at $4.25/hr and HGX H100 at $4.76/hr. CoreWeave's current pricing page prices HGX H100 at $6.16/GPU-hr, a gap of roughly 29% for the same silicon, and that gap exists independent of the 25% SKU increase disclosed in August.
If you're pricing out a renewal and an old review, forum post, or cached page quotes you a CoreWeave H100 rate that looks unusually low, check which pricing page it came from before you use it as a negotiating baseline. It may be pulling from Classic, not current.
| Pricing surface | HGX H100 rate | HGX H200 rate | What it actually reflects |
|---|---|---|---|
| Current on-demand page | $49.24/hr ($6.16/GPU-hr) | $50.44/hr ($6.31/GPU-hr) | Public self-serve rate, unchanged since at least July 5, 2026 |
| Legacy Classic page | $4.76/hr (per GPU) | Not listed | Older pricing tier, roughly 29% below the current H100 rate |
| Negotiated contracts | Not published | Not published | Where the disclosed 25% SKU increase is actually landing |
Pricing fluctuates based on GPU availability. The prices above are based on 21 Aug 2026 and may have changed. Check current GPU pricing → for live rates.
What This Means for Your Contract Renewal
If You're Negotiating a Reserved Renewal Right Now
Walk in with the 25% figure as your reference point, not your worst-case scenario. A few things worth doing before you sign anything:
- Ask which SKU tier your renewal falls into. Intrator's own comments suggest Blackwell and Vera Rubin capacity is where pricing is "setting new highs." If you're renewing H100 or H200 capacity, ask directly whether your specific tier saw the full 25% or something closer to the "elevated but not new highs" language used for prior-generation SKUs.
- Get the current on-demand rate in writing. CoreWeave's published rate ($6.16/GPU-hr for H100, $6.31/GPU-hr for H200) is your comparison anchor even though most volume moves through contracts. If your quoted renewal rate is close to or above the on-demand number, the reserved discount you're supposedly getting has mostly disappeared.
- Price a walk-away alternative before you're mid-negotiation. Knowing what the same capacity costs elsewhere is the single biggest lever in a renewal conversation. See our CoreWeave alternatives roundup if you want the full field, or Spheron vs CoreWeave for a direct head-to-head.
- Check whether you actually need H100 or H200. If your renewal is up for both, our H100 vs H200 comparison breaks down where the extra H200 memory bandwidth actually pays for itself versus where H100 is still the better economics.
What the Same H100/H200 Capacity Costs on Spheron
CoreWeave's H100 and H200 tiers sell exclusively as 8-GPU nodes, so the comparison that matters is the per-GPU rate, not the sticker price on the node. Here's how CoreWeave's current on-demand rate compares to Spheron's live rate, checked the same day this post was published:
| Metric | CoreWeave H100 (HGX, 8x) | Spheron H100 SXM5 | CoreWeave H200 (HGX, 8x) | Spheron H200 SXM5 |
|---|---|---|---|---|
| On-demand $/GPU-hr | $6.16 | $3.38 | $6.31 | $4.22 |
| Spot $/GPU-hr | $2.46 | $2.91 (SXM5) / $2.20 (PCIe) | $2.62 | $2.53 |
| Minimum node size | 8 GPUs | 1 GPU | 8 GPUs | 1 GPU |
| Billing granularity | Per hour | Per minute | Per hour | Per minute |
Spheron's H100 SXM5 on-demand rate runs about 45% below CoreWeave's per-GPU H100 rate, and H100 PCIe on-demand runs closer to 57% below it. H200 SXM5 on-demand is roughly 33% cheaper. Spot is more mixed: Spheron's H200 spot rate is modestly cheaper than CoreWeave's, while Spheron's H100 SXM5 spot rate currently runs slightly above CoreWeave's H100 spot, though Spheron's H100 PCIe spot still comes in below it. Rates move with availability on both sides, so treat any spot comparison as a snapshot, not a fixed gap.
A worked example: an 8-GPU H100 node run continuously for a 720-hour month costs roughly $35,453 on CoreWeave's current on-demand rate. The equivalent 8 GPUs of H100 SXM5 on Spheron run about $19,469 for the same month, a difference of roughly $15,984, without a multi-year contract or an 8-GPU minimum forcing you to pay for capacity you don't need. The same math on H200 GPU rental puts CoreWeave's node at roughly $36,317/month against Spheron's roughly $24,307, about $12,010 cheaper.
Spheron aggregates capacity from 5+ providers on the backend, which is part of why on-demand availability holds up without requiring a fixed reservation window or a multi-year commitment to get a competitive rate. Documentation for standing up instances is in the Spheron docs.
Pricing fluctuates based on GPU availability. The prices above are based on 21 Aug 2026 and may have changed. Check current GPU pricing → for live rates.
If your CoreWeave renewal is still weeks out, it's worth tracking whether this is the last adjustment of the year or the first of another round. Our weekly GPU cloud pricing digest logs hikes and cuts across providers as they're reported, so the next move doesn't catch your budget flat-footed.
Renewing a CoreWeave H100 or H200 contract into this pricing environment? Spheron runs the same hardware on-demand with per-minute billing, single-GPU granularity, and no multi-year commitment required.
Spheron H100 instances → | Check H200 GPU pricing → | Get started on Spheron →
Frequently Asked Questions
On its Q2 2026 earnings call (August 11, 2026), CFO Nitin Agrawal disclosed that management's July pricing changes included 'an approximately 25% increase across SKUs in response to the current demand environment and the increasing ROI our customers are observing from their investments in the CoreWeave platform as they shift to inference.' It was a company-wide SKU repricing, not a single-product adjustment, and it landed in July even though it wasn't disclosed publicly until the August earnings call.
Because the public, self-serve rate card and the 25% SKU increase are two different pricing surfaces. CoreWeave's published on-demand numbers for HGX H100 ($49.24/hr) and HGX H200 ($50.44/hr) have held at those exact figures since at least July 5, 2026, and were still unchanged when checked again in August. Most of CoreWeave's revenue runs through multi-year committed contracts negotiated directly with an account team, not the checkout-page rate. The 25% increase is landing primarily in those negotiated renewals and new contracts, which is why the on-demand page looks untouched.
Both, but unevenly. CEO Michael Intrator said on the same call that 'pricing and margins for our Blackwell and Vera Rubin SKUs are setting new highs, while pricing for prior generation SKUs is at or above where it was years ago.' Blackwell and Vera Rubin capacity is commanding the steepest pricing. Hopper-generation H100 and H200 aren't exempt, they're priced above their own historical baseline, but the newest SKUs are where CoreWeave's pricing power is most visible.
Three things are compounding: CoreWeave says its near-term capacity is 'effectively sold out,' so there's no idle inventory pushing prices down; HBM memory costs rose roughly 30% in a single quarter (Q4 2025) with data center GPU-server lead times running 36-52 weeks, so CoreWeave's own input costs are rising; and demand for the newest Blackwell and Vera Rubin platforms is outrunning available supply, giving CoreWeave negotiating leverage on every renewal that touches those SKUs.
Go in expecting the 25% figure to be the opening reference point, not a ceiling you can talk down to zero. Get the exact current on-demand and reserved rate in writing before you negotiate, ask whether the increase applies to your specific SKU tier or only to newer capacity, and price out an on-demand alternative like Spheron as a walk-away option, since Spheron's H100 and H200 rates are billed per minute at single-GPU granularity with no multi-year commitment required.
