Home / Directory / AI infrastructure / Bare-metal clusters / TensorWave
TensorWave
AMD Instinct GPU cloud offering dedicated bare-metal AI clusters (MI300X and peers) for training and inference teams that want an alternative to NVIDIA fleets. TensorWave raised a $350 million Series B in June 2026, co-led by Magnetar and AMD Ventures.
Platform teams standardizing on AMD Instinct bare-metal clusters that can own ROCm tooling and want dedicated multi-GPU servers.
Teams locked to CUDA-only stacks, or buyers who need instant self-serve NVIDIA pods with a public rate card.
Verdict
TensorWave fits when you deliberately want AMD MI300X-class bare metal instead of another NVIDIA neocloud. High-memory OAM servers and dedicated clusters are the pitch; ROCm readiness is on you. Public self-serve dollar cards are thin, so treat third-party GPU-hour listings as directional and get a written quote. Skip it if your stack is CUDA-only or you need same-day NVIDIA pods.
Score Breakdown
How TensorWave scores in the categories that matter to its buyers.
Pricing
TensorWave commercializes dedicated AMD GPU clusters primarily through sales quotes rather than a full self-serve rate card. Third-party GPU price indexes have listed MI300X around the low-$2/GPU-hour band for 8-GPU bare-metal nodes; treat those as directional only. As of September 2026.
Model: quote-led bare-metal clusters. Typical packaging is multi-GPU AMD Instinct nodes (for example 8x MI300X OAM) reserved for training or inference, with commitment terms negotiated per deal. Confirm ROCm image support, networking fabric, and egress in the quote. Do not treat aggregator listings as a public TensorWave card.
The Field at a Glance
Where TensorWave ranks among Bare-metal clusters vendors we reviewed, by Overall Score and relative typical engagement cost.
TensorWave scores 6.7, behind Lambda (7.3) and RunPod (8.0). Its relative cost is mid-band for dedicated AMD bare metal compared with public pod meters.
Use-case matrix
| Use case | Fit | Notes |
|---|---|---|
| AMD Instinct bare-metal clusters | Strong | AMD hardware is what sets it apart from NVIDIA neoclouds. |
| High-memory LLM training nodes | Strong | The MI300X memory capacity is the draw. |
| Self-serve hourly NVIDIA pods | Poor | This is a different product, and RunPod and Lambda sell it. |
| CUDA-only production stacks | Poor | Porting from CUDA to ROCm takes engineering time. |
| Reserved multi-node AMD fleets | Strong | Dedicated clusters are sold through sales quotes. |
| Retail hobby GPU rental | Weak | The business is built around clusters and enterprise contracts. |
Review Excerpts
Excerpts from public reviews and forum discussions of TensorWave, plus representative buyer views drawn from public information. Paid reviews and pay-for-play sites were excluded.
“Aggregator listings describe 8-GPU MI300X bare-metal configurations with roughly 192 GB HBM per GPU, sold as dedicated cluster capacity rather than noisy shared VMs.”
“I suggest taking the report with a grain of salt.”
“The MI300X memory capacity let our larger models fit on fewer GPUs, and we got dedicated nodes instead of shared VMs.”
Representative buyer view
“Our CUDA code needed real porting work. ROCm versions and framework pins took longer to sort out than the GPU setup did.”
Representative buyer view
Where customer reviews couldn't be found, our analysts attempted to give a voice to buyer sentiment, drawing on real information from review sites, forums, and product documentation.
Based on: TensorWave’s documentation and SemiAnalysis ClusterMAX review of TensorWave
Company Facts
- Founded
- 2023 (CB Insights)
- Headquarters
- Las Vegas, Nevada, United States (CB Insights)
- Total funding raised
- $498M (CB Insights)
- Key investors
- Nexus Venture Partners, AMD Ventures, Magnetar Capital, DataPower Ventures, Maverick Silicon (CB Insights)
- Customers
- Modular (CB Insights)
- Website
- tensorwave.com
Key Leadership
Darrick Horton
Founder and CEO
Methodology
This page is an independent evaluation of TensorWave for buyers comparing options in bare-metal clusters. AI Industry Reviews accepts no sponsorships, advertising, or pay-for-placement fees. TensorWave did not pay for this review.
What we scored
The headline number is an Overall Score on a 0-10 scale. Eight criteria fall under it in two groups.
- Buyer outcomes (for bare-metal clusters)
- AMD MI300X density
- Bare-metal cluster control
- Software stack maturity
- Availability / lead time
- Company & commercial
- Innovation & product leadership
- Project management & communication
- Pricing
- Contract fairness
Pricing measures whether the price looks fair for the value delivered, including packaging and renewal friction that show up in real buying cycles.
Score Composition
| Input | Weight | What it covers |
|---|---|---|
| Reviews | 40% | A proprietary read of what practitioners say about likes, complaints, and day-to-day use, including public review sites, forums, and private chat rooms. Paid reviews and pay-for-play sites are out of scope. |
| Product | 35% | Hands-on look at screens and workflows. |
| Pricing | 15% | Whether the price looks fair for what you get. |
| Docs & training | 10% | Docs, tutorials, and training material. |
How we balanced the evidence
The Overall Score is the simple average of the eight criteria. Recommendation language follows that score and the use-case matrix above.
Scope
TensorWave is graded here as bare-metal clusters. Criteria scores can move as more review volume and product checks are added.