Hyperscale design-manufacturersVendor profile

Flex

Verified2026-07-15
Systems listed4

Flex's genuinely in-house engineering is power and cooling, not GPU compute nodes: the 800 VDC power rack co-developed with NVIDIA for Vera Rubin, JetCool microjet cold plates, and grid-to-chip rack integration. It touches nearly every AI datacenter without ever badging the server. The datacenter segment spins off as a standalone company in 2027.

What they design in-house01
  • Vertically integrated grid-to-chip datacenter model: compute integration plus power distribution plus liquid cooling as complete rack-level solutions, racks rated to 120kW with a stated path to 300kW
  • Acquired JetCool (2024) for in-house direct-to-chip microjet liquid cooling: SmartPlate cold plates rated for over 3,000W TDP
  • In-house power engineering: 800 VDC Power Rack co-developed with NVIDIA for Vera Rubin, 110kW power shelf, capacitive energy storage
  • Turnkey liquid-cooled server integrations on Dell PowerEdge platforms with unified warranty
  • Cloud and Power Infrastructure segment being spun off into a standalone datacenter company, expected Q1 2027
NVIDIA systems02
ModelPlatformForm factorCooling
800 VDC Power Rack
Co-developed with NVIDIA, scaling toward 880kW per rack on future platforms
Vera Rubin (power)power rack sidecarair
110kW power shelf
Shown COMPUTEX 2026 with 30kW capacitive energy storage
Vera Rubin NVL72 (power)rack power shelfair
JetCool SmartSense CDU
300kW per unit, scalable to 2.1MW per row
MGX (cooling)6U in-rack CDUDLC
JetCool SmartPlate System
Microjet cooling for Dell PowerEdge GPU servers, turnkey with unified warranty
PCIe (cooling)direct-to-chip cold platesDLC
NVIDIA relationship03
  • Listed among the 40+ datacenter infrastructure ecosystem partners NVIDIA cites on Blackwell / GB200 NVL72 (power and cooling role)
  • Launched the 800 VDC Power Rack developed with NVIDIA for the Vera Rubin platform
  • No Flex-branded HGX or NVL72 compute server SKU confirmed; its disclosed role concentrates on power, cooling and rack subsystems
Manufacturing
  • Dallas, Texas: new 400,000 sq ft plant for grid-to-chip datacenter power
  • Columbia, South Carolina: dedicated datacenter power capacity
  • 100+ facilities in ~30 countries
Field notes
  • Buyer-facing role is power and cooling infrastructure plus Dell-partnered liquid-cooled servers, not Flex-branded GPU compute nodes
  • CPI segment spin-off into a standalone datacenter company expected Q1 2027
  • Do not confuse Flex Ltd with Flexential, a colocation operator
Questions we get on this vendor

Is Flex an OEM, ODM or integrator for NVIDIA AI servers?

Flex sits in the "Hyperscale design-manufacturers" tier of our index. Vertically integrated grid-to-chip datacenter model: compute integration plus power distribution plus liquid cooling as complete rack-level solutions, racks rated to 120kW with a stated path to 300kW

Does Flex design its NVIDIA servers in-house?

Flex's genuinely in-house engineering is power and cooling, not GPU compute nodes: the 800 VDC power rack co-developed with NVIDIA for Vera Rubin, JetCool microjet cold plates, and grid-to-chip rack integration. It touches nearly every AI datacenter without ever badging the server. The datacenter segment spins off as a standalone company in 2027.

Which NVIDIA platforms does Flex ship?

Flex ships systems on Vera Rubin (power), Vera Rubin NVL72 (power), MGX (cooling), PCIe (cooling). The systems table on this page lists the specific models.

Where does Flex manufacture its AI servers?

Dallas, Texas: new 400,000 sq ft plant for grid-to-chip datacenter power. Columbia, South Carolina: dedicated datacenter power capacity. 100+ facilities in ~30 countries