HPE just made a $1.2 billion bet that the real money in AI infrastructure isn't in the chips — it's in the wires connecting them.
The Summary
- HPE won a $1.2 billion order from Vultr to supply AMD server racks bundled with HPE networking gear, signaling a shift in how cloud companies are buying AI infrastructure
- HPE boosted its networking revenue forecast on the strength of this deal, betting that AI networking can deliver faster revenue growth and higher margins than commodity hardware
- The deal positions HPE's networking strategy as the differentiator in a market where everyone is chasing AI infrastructure dollars
The Signal
HPE isn't trying to beat Nvidia at the chip game. It's selling the plumbing. The Vultr deal bundles AMD server racks with HPE's networking equipment, a move that says more about where AI infrastructure is heading than any chip benchmark. Vultr, a smaller cloud provider competing against AWS and Azure, needs to build inference clusters that don't bankrupt them on interconnect costs.
The economics matter here. Training a large language model is a one-time expense. Running inference on it is forever. Every query, every API call, every chatbot interaction flows through network fabric. If that fabric is slow or expensive, your margin evaporates. HPE is betting that cloud providers now understand this, and will pay a premium for networking that doesn't bottleneck their AMD chips.
"HPE is betting that AI networking can turn surging demand into faster revenue growth and higher margins."
What makes this deal different from a typical hardware order:
- It's not just servers. It's servers plus the networking stack, sold as integrated infrastructure
- Vultr is choosing AMD over Nvidia, which only works if the networking can handle non-CUDA workloads efficiently
- HPE raised its networking forecast specifically because of this deal, meaning they see a pattern, not a one-off
The timing tells you something too. HPE boosted its networking revenue forecast right after announcing this order, which means their sales pipeline is seeing more deals structured this way. Cloud providers are no longer buying compute and networking separately. They're buying AI-ready clusters where the network is purpose-built for the workload.
This is the move from "AI infrastructure" to "AI systems." Vultr isn't buying parts. They're buying a thing that works out of the box for inference at scale. HPE is assembling the pieces and taking a margin on the integration. That margin is higher than what they'd make selling commodity servers alone, which is why Fortune notes HPE's focus on faster revenue growth and higher margins from this strategy.
The Implication
Watch for more bundled deals where the network is the differentiator, not the chip. If HPE can prove that their networking turns mid-tier AMD hardware into competitive inference infrastructure, every cloud provider not locked into Nvidia's ecosystem has a playbook to follow. For builders, this means inference costs might finally start dropping as network efficiency catches up to compute power.
The real test comes when Vultr starts running production workloads. If HPE's networking delivers on the promise, expect AMD to win more cloud deals and expect networking to become a first-class concern in AI infrastructure procurement. If it doesn't, this becomes a very expensive lesson about why everyone just buys Nvidia and calls it a day.