The Deal

TensorWave, a data center company based in Phoenix, just raised $350 million in funding. This values the company at $1.55 billion, according to The Wall Street Journal.

Here's what TensorWave does: it builds warehouses full of specialized computer chips used to train and run artificial intelligence systems. The key difference is that TensorWave uses chips made by AMD instead of Nvidia, which is what most other AI companies use. Even AMD itself has invested in TensorWave's funding round.

TensorWave plans to use the money to build more data centers. This means buying more computer servers, signing long-term contracts for electricity, and adding more AMD chips to its facilities.

Why AMD? Why Now?

For the past few years, there has been a shortage of Nvidia chips. Companies that want to use Nvidia's most powerful GPUs (the H100 and H200 models) often have to wait months to get them. Smaller companies upgrading to Nvidia's newer Blackwell chips face long delays and higher costs.

AMD's chips offer an alternative. They're not better at every task, but they're good enough for many jobs — especially when you're trying to save money. And AMD chips are actually available in the quantity companies need right now.

TensorWave's bet is simple: some customers will choose cheaper, available chips over the most famous brand name. As AI projects move from experiments to actual production use, cost matters more. Companies are asking, "Can we do this more cheaply?"

AMD being an investor in TensorWave is significant. When a chip maker takes an ownership stake in a company that uses its products, it's not just about the money. It creates a formal partnership. AMD gets a reference customer to point to when talking to other potential buyers. TensorWave gets priority access to chips and direct technical support from AMD's engineers. Both sides benefit from the arrangement.

The Market Landscape

The AI computing market has split into three layers.

At the top are the giants: Amazon Web Services, Microsoft Azure, and Google Cloud. They build their own custom chips to use alongside Nvidia purchases.

In the middle is a newer group of specialist companies — CoreWeave, Lambda Labs, and now TensorWave — that rent computing power to customers. They compete on price, how quickly they can deliver, and what hardware options they offer.

Below that is a long tail of smaller, regional operators.

TensorWave occupies a specific position in that middle tier. CoreWeave, which became a public company earlier this year, depends almost entirely on Nvidia chips. CoreWeave is betting that Nvidia will remain the standard and that chip prices hold their value. TensorWave is, in effect, betting the opposite: that AMD's software ecosystem will improve fast enough and that cost-conscious customers will switch.

We've seen this pattern before. In the 2000s, AMD's Opteron chips briefly won significant market share from Intel in data centers, simply because Intel's supply was tight and expensive. Once Intel responded with better pricing and faster innovation, that advantage disappeared. AMD never solidified the gains. The GPU market is different — companies can get locked into Nvidia's software ecosystem, which makes switching harder — but the basic pattern might repeat. Companies facing tight margins look for credible alternatives, and someone has to build the infrastructure around them.

The Valuation Question

TensorWave raised $350 million and is now valued at $1.55 billion. That means investors valued the company at roughly $1.2 billion before this funding round.

For a business that rents computing power, this is a modest valuation. Unlike software companies that can scale with low costs, data center companies face heavy, ongoing expenses: real estate, electricity, hardware depreciation, and the challenge of keeping their machines running at high utilization rates. These businesses are typically priced based on current earnings before interest, taxes, depreciation, and amortization — or EBITDA, a measure of actual cash generation — rather than on future revenue growth alone.

AMD's decision to invest as a strategic partner likely helped TensorWave's valuation. Companies with supply chain relationships often attract higher prices from investors looking to build positioning. But the round's terms also reflect a market cooling off. In 2023 and early 2024, GPU cloud companies were attracting sky-high valuations based on the assumption that chip shortages would last forever. That fever has broken.

What the Money Builds

A modern AMD data center with state-of-the-art AI chips needs tens of megawatts of electrical power. To put that in perspective: imagine a small city's worth of electricity demand, all in one building. At current construction costs, $350 million can likely pay for one to three major new data centers, depending on where they are, how much electricity costs locally, and whether TensorWave is building from scratch or leasing existing space.

It's also likely that TensorWave will borrow additional money to fund construction. Infrastructure businesses typically do this — they secure loans against future revenue from customers, using the equity funding as the down payment.

One thing worth noting: AMD's stake may come with a practical advantage. In a market where chip supplies are still tight, preferential access to AMD's inventory could be valuable. This advantage isn't explicitly stated in any reporting, but it's a reasonable inference.

What This Means

TensorWave's funding is part of a broader pattern: money keeps flowing into AI data center companies, even as electricity costs rise, construction takes longer, and uncertainty grows about whether companies will keep spending on AI infrastructure at current levels.

The fact that a major chip maker believes AMD's technology is ready for large-scale deployment is worth noting. It's a vote of confidence that AMD's software ecosystem is maturing fast enough to compete.

The real test will come next. Data centers succeed or fail on utilization — how much of the time their expensive equipment is actually being used by paying customers. A cluster running at 60% utilization has very different economics from one running at 85%. That number, more than anything else, will determine whether TensorWave's business makes sense. We don't know it yet.