buildConfirmed5 publishers Intel named Crescent Island the GPU D-Series at OCP 2026, a 350W air-cooled PCIe card with up to 480GB of LPDDR5X. Tom's Hardware reported a 1.3 PFLOPS FP8 peak target and 1.5 TB/s of bandwidth; customer sampling is in progress.
Perspective Coverage
5 publishers
- Builder
- Builder 36%
- Operator
- Operator 33%
- Investor
- Investor 31%
Reality
- Evidence40
- Adoption12
- Hype gap+20
- Incentives65
- Confidence72
Nscale's IPO filing counts $103.4 billion in active and contracted AI compute deals, with only about $2.6 billion of it active. Teams renting GPUs from smaller clouds should check whether their capacity is running yet and how the provider will pay to build the rest.
Reality
- Evidence55
- Adoption25
- Hype gap+40
- Incentives75
- Confidence60
CoreWeave rents out Vera Rubin NVL72 racks backed by one customer's 4.8x throughput claim
CoreWeave now rents Nvidia's Vera Rubin NVL72 racks, with first customer Cognition reporting up to 4.8x the token throughput on SWE-2 inference. That figure is a best case from a single inference workload, and it is the only customer result behind the launch.
Reality
- Evidence35
- Adoption22
- Hype gap+35
- Incentives75
- Confidence45
Nvidia is putting $1 billion into Nscale's $3.36 billion pre-IPO round, funding a customer still building most of its capacity. Buyers signing multi-year capacity deals should check how that build-out is paid for before they check the price.
Perspective Coverage
4 publishers
- Builder
- Builder 19%
- Operator
- Operator 35%
- Investor
- Investor 46%
Reality
- Evidence66
- Adoption22
- Hype gap+38
- Incentives72
- Confidence60
The London builder is reportedly seeking $3 billion while telling investors it can go from 831MW to about 11GW. Anyone booking 2027 compute should price the gap, not the headline.
Reality
- Evidence45
- Adoption50
- Hype gap+35
- Incentives65
- Confidence55
buildOne report1 publisher Lambda ran 19 Blackwell nodes on the power budget provisioned for 16
The one measured number behind NVIDIA's tokens-per-megawatt pitch comes from Lambda's Blackwell cluster, where power reclaimed from static provisioning ran three extra nodes. Amazon's Annapurna Labs and d-Matrix got a line each.
Reality
- Evidence33
- Adoption45
- Hype gap+44
- Incentives90
- Confidence60
Two Malaysian sites take Firmus past 900MW of contracted capacity, though five of its seven data centres are still unbuilt, which puts the scarce input in this trade on the grid side rather than the silicon side.
Reality
- Evidence52
- Adoption35
- Hype gap+34
- Incentives76
- Confidence58
Hudson River Trading's CoreWeave deal puts a quant fund at the front of the Vera Rubin queue
HRT went from new CoreWeave account to a multi-year, multi-billion-dollar commitment in about five months, and bought early access to NVIDIA's newest parts along the way.
Reality
- Evidence27
- Adoption52
- Hype gap+34
- Incentives82
- Confidence41
CoreWeave's $104bn backlog is now a margin question, not a demand question
Management says scale is starting to show up as operating leverage. The release as supplied gives a backlog, power figures and more than $14bn of fresh capital, but no margin numbers.
Publishers:investors.coreweave.com
Reality
- Evidence42
- Adoption71
- Hype gap+32
- Incentives88
- Confidence55