Enterprise AI inference is moving from shared public clouds to dedicated, purpose-built infrastructure. Rackspace and Together AI announced a multiyear strategic agreement on November 10, 2026, under which Rackspace will provide and operate dedicated NVIDIA GB300 NVL72 systems to power Together AI’s inference workloads, with deployment planned for 2027. The deal pairs one of the oldest names in managed hosting with one of the fastest-growing AI-native clouds — and it says a lot about where enterprise AI is heading.
What the Rackspace and Together AI Deal Covers
Under the initial agreement, Rackspace expects to supply and operate dedicated NVIDIA Blackwell Ultra infrastructure — including power, liquid cooling, networking, and storage — while Together AI runs its managed cloud platform on top, offering access to open-source models, inference optimizations, and developer tools. The deployment will use NVIDIA Spectrum-X Ethernet networking and sit in a purpose-built, liquid-cooled facility.
The companies also signed a separate, non-binding letter of intent to explore a joint go-to-market approach for enterprise customers, with Together AI providing inference capabilities and Rackspace contributing infrastructure, platforms, and forward-deployed engineering. Future joint offerings could combine Together AI’s inference, post-training, and reinforcement learning capabilities with Rackspace’s infrastructure for customers that need greater control, compliance, and data sovereignty.
Why the NVIDIA GB300 NVL72 Matters for Inference
The hardware at the center of the deal is NVIDIA’s flagship rack-scale system for AI reasoning. The GB300 NVL72 packs 72 Blackwell Ultra GPUs and 36 Grace CPUs into a single liquid-cooled rack, connected by fifth-generation NVLink with 130 TB/s of bandwidth. NVIDIA says the system delivers 1.5 times more FP4 Tensor Core performance and twice the attention-layer acceleration of previous Blackwell GPUs, and it is purpose-built for test-time scaling inference — the technique behind today’s reasoning models.
That focus on inference, rather than training, is the point. As reasoning models spend more compute “thinking” at query time, inference has become the workload that decides whether an AI service is fast, affordable, and reliable enough for production. Rackspace, an NVIDIA Cloud Partner, is betting that enterprises will want that inference capacity dedicated to them rather than shared. For context on how the inference hardware race is shaping up, see our coverage of Google’s new TPU systems for confidential inference.
The Enterprise Angle: Control, Compliance, and Sovereignty
Rackspace’s pitch is not raw performance alone. The company is positioning itself as “the operator of the full enterprise AI stack,” leaning on more than 25 years of running mission-critical environments in regulated industries. The announcement stresses security, trust, and governance — the requirements that keep banks, insurers, and healthcare organizations from putting production AI on shared infrastructure.
Together AI brings the other half of the equation. Founded in 2022, the company powers more than two million developers and describes itself as the AI Native Cloud, combining open-source models with frontier research in inference and scalability. Its platform offers serverless inference, dedicated deployments, fine-tuning, and reinforcement learning — the full journey from experimentation to production scale. The partnership lets each side do what it does best: Together AI focuses on the inference software, while Rackspace operates the physical infrastructure.
The sovereignty angle matters too. Rackspace says the companies will evaluate deployment configurations that bring Together AI’s cloud closer to enterprise data and workloads, potentially giving customers more flexibility over where and how their AI runs. That echoes a broader industry trend — NVIDIA’s sovereign AI partnerships, for example, show how governments and enterprises are demanding AI infrastructure they can control.
The Financing Question Hanging Over the Deal
There is an important caveat. Rackspace’s obligations under the agreement are subject to conditions, including securing financing for the GB300 NVL72 systems on acceptable terms. As of the announcement date, no definitive GPU financing commitment had been obtained, though Rackspace said it was in active discussions with potential lenders.
On November 6, 2026, Rackspace entered into a share purchase agreement with an investor group led by Voss Capital for a $66 million private placement of common stock. The company expects to use the proceeds for general corporate purposes, working capital, and growth capital, including GPU-related initiatives — but the GPU purchase itself will need separate financing. Deployment is planned for 2027, subject to those conditions being satisfied.
This is worth watching because it reflects a wider pattern in the AI infrastructure market: the hardware is expensive, the demand is real, and the companies building AI factories are financing them deal by deal. Rackspace says its approach is to seek contractual commitments for dedicated capacity and align financing with expected customer demand and the useful life of the hardware.
What This Signals for the AI Infrastructure Market
The Rackspace–Together AI agreement fits a clear trend: inference is becoming a dedicated-infrastructure business. Cloud providers, neoclouds, and managed service providers are all racing to offer GPU capacity tuned for production inference, and differentiation increasingly comes from operations — uptime, compliance, cost control, and the ability to run open-source models efficiently — rather than from owning the models themselves.
For enterprises evaluating their AI options, the deal is a reminder to look beyond model benchmarks. The questions that matter are operational: who runs the infrastructure, how is it secured, where does the data live, and what happens when a workload needs to scale. Our practical guide to AI agents covers the broader questions enterprises should ask before committing to an AI platform, and enterprise AI platforms like Snowflake’s agentic stack show how the same governance concerns are shaping the software layer too.
What to Watch Next
Three things will determine whether this partnership delivers. First, whether Rackspace closes the GPU financing and hits the 2027 deployment target. Second, whether the non-binding letter of intent turns into real joint enterprise offerings with defined data residency and compliance features. And third, whether demand for dedicated inference capacity keeps growing as reasoning models push more compute into the inference stage. If all three break the right way, Rackspace’s bet on operating the full enterprise AI stack could pay off — and Together AI gets the dedicated Blackwell Ultra capacity it needs to scale.
