Elon Musk's xAI shifts to the role of "cloud provider," opening tens of thousands of GPU computing power to AI programming unicorn Cursor

robot
Abstract generation in progress

ME News, April 17 (UTC+8). According to Beating monitoring, xAI plans to provide large-scale computing power support to the AI programming startup Cursor, enabling it to train the latest programming model Composer 2.5 on xAI’s infrastructure. According to Business Insider, Cursor will use tens of thousands of GPUs from xAI’s “Colossus” data center.

This arrangement marks a major strategic shift for xAI: by leasing out redundant computing power, xAI is expanding from a pure model R&D company into the role of a cloud service provider, similar to AWS and CoreWeave.

The background of this cooperation is complex. Earlier this March, xAI just poached two product engineering leads from Cursor to run its product team. In addition, an internal memo at xAI shows that its current GPU utilization rate (MFU) is only 11%, far below the industry average of 35% to 45%. xAI President Michael Nicolls has asked the team to raise utilization to 50% within the next few months. Opening up computing power to external unicorns not only helps share the costly data center operating expenses, but also allows xAI to obtain valuable engineering feedback by providing top-tier programming agent services.

Currently, Cursor is in funding negotiations at a valuation of $50 billion. Against the backdrop of OpenAI and Anthropic aggressively moving into the programming assistant track, tying its competitive advantage to xAI’s computing resources has become a key factor.

(Source: BlockBeats)

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • 11
  • 8
  • Share
Comment
Add a comment
Add a comment
GoldfishOnIce
· 2h ago
Sharing costs is a minor matter; the main thing is to make use of idle cards.
View OriginalReply0
MirrorBallReflection
· 11h ago
MFU doubling means the same cost produces double the output, and Cursor's valuation can still increase.
View OriginalReply0
TheProphetOfToast
· 13h ago
Open computing power for feedback, xAI, this calculation is so loud I can hear it from Mars.
View OriginalReply0
TreatMemesAsBeliefs
· 13h ago
The 50 billion valuation is still raising funds; Cursor's cash flow pressure isn't small, right?
View OriginalReply0
ThetaSideEye
· 13h ago
Composer 2.5 is coming, Cursor users are ecstatic
View OriginalReply0
CollateralCora
· 13h ago
35-45% industry average, 11% is indeed a bit hard to sustain
View OriginalReply0
SaveABitOnGasFees
· 13h ago
Nicolls has been investing in 50% MFU for several months, hitting the KPI target.
View OriginalReply0
FragmentedSilverStarMap
· 13h ago
Switching from model vendors to cloud service providers, Musk's move is quite pragmatic.
View OriginalReply0
OrderCancellerAfterTheRain
· 13h ago
Colossus has tens of thousands of cards, just thinking about this electricity bill is exciting.
View OriginalReply0
On-ChainCheatSheetKing
· 13h ago
xAI is going the AWS route; training models ourselves isn't as profitable as selling computing power.
View OriginalReply0
View More
  • Pinned