The $1.17 Billion Data Deal Exposes Cracks in AI’s Infrastructure Boom
CoreWeave’s massive $1.17 billion partnership with Vast Data sent shockwaves through the AI infrastructure world late in 2025. On the surface, it looked like a win for everyone. Vast became the main storage backbone for CoreWeave’s booming GPU cloud, and believers in the “AI gold rush” cheered. To them, the deal confirmed that specialized AI clouds were thriving, that data bottlenecks needed smarter engineering, and that money would keep pouring toward those solving these problems.
But look a little deeper, and the story gets trickier. Beneath the excitement lies a tangled web of financing and expectations—one that might be stretching further than real cash flow can reach. The issue isn’t whether AI workloads are real; they absolutely are. The question is whether companies are scaling their infrastructure faster than their revenue streams can sustain.
Inside AI’s Data Dilemma
There’s no denying the technical logic behind this deal. Training modern AI models demands enormous datasets—exabyte scale—and lightning-fast access speeds. We’re talking 7 GB per second per GPU. Traditional tiered storage systems, where “cold” data sits on old hard drives or even tape, simply can’t keep up. Pulling that data back into use can stall a massive 512-GPU cluster for hours, burning through $30,000 to $50,000 in wasted opportunity each time it happens.
Vast Data steps in with a simple but powerful fix. Its all-flash object store merges all storage into a single, high-speed layer. Add NVIDIA’s BlueField DPUs—which handle storage operations on dedicated chips—and you cut CPU overhead in half while keeping everything secure with hardware encryption. For CoreWeave, which manages more than 250,000 NVIDIA GPUs across 32 data centers, this means one thing: better efficiency. Utilization rates could rise from the industry average of around 75% to something much closer to 90%.
That boost couldn’t come at a better moment. CoreWeave’s been growing like wildfire—$1.9 billion in revenue in 2024, an eightfold jump from the year before. Major contracts, including a $14 billion deal with Meta and over $22 billion with OpenAI, are fueling this expansion. Handling those mountains of data isn’t just a technical challenge; it’s a make-or-break factor that determines whether those expensive GPUs earn their keep or sit idle.
The Speed Trap of AI Investing
Still, not every investor is popping champagne. The structure of this deal raises eyebrows. That eye-catching $1.17 billion? It’s a multi-year commitment, not a single check. CoreWeave will pay over three to five years, depending on how quickly it scales and how much capacity it actually uses. That means Vast’s revenue depends on CoreWeave’s ability to grow—and stay profitable—well into the next decade.
Here’s where the loop starts. Meta and OpenAI commit huge sums to CoreWeave for GPU capacity they’ll need later. CoreWeave, in turn, signs contracts with Vast to ensure it can deliver that capacity. Vast then uses those contracts to justify a higher valuation—jumping from $9.1 billion in 2023 to as high as $30 billion in 2025—despite only around $200 million in recurring revenue. Each company’s story props up the next.
And at the center of it all? NVIDIA. The chip titan holds equity in CoreWeave, reportedly eyes investments in Vast, and remains the indispensable supplier tying the ecosystem together. NVIDIA’s influence gives everyone confidence—but it also concentrates risk. When one company anchors multiple layers of the supply chain, the whole system can start echoing its own optimism. Demand projections sound enormous because everyone believes NVIDIA-backed capacity will always find buyers.
That’s not fraud—it’s a familiar pattern in capital-heavy industries. During adoption booms, companies build for a future that’s just a little too far ahead. It’s how bubbles quietly form: reasonable assumptions made at unreasonable speed.
If this sounds familiar, it should. The late-’90s fiber-optic boom followed the same script. Demand for internet bandwidth was real, just not as immediate as investors hoped. Billions were poured into cables that sat dark for years, and when the music stopped, plenty of vendors vanished.
The Hidden Risk Few Are Pricing In
For Vast, the CoreWeave deal could represent 20–30% of its near-term revenue if the company spreads $1.17 billion over four years against a $200–250 million yearly baseline. That’s a lot of eggs in one basket. If CoreWeave hits delays—power shortages, permitting roadblocks, or even slower-than-expected customer growth—Vast’s revenue curve could dip overnight.
CoreWeave’s juggling plenty of its own challenges. It’s spending billions on GPUs, buying software startups like Marimo to move up the value chain, and expanding overseas. Its $7.5 billion debt raise in 2025 fuels all this, but that’s fixed money chasing variable demand.
Meanwhile, the big cloud players—AWS, Azure, Google Cloud—aren’t sitting idle. They’re rolling out their own AI-optimized storage layers with single-zone, low-latency designs meant for exactly this kind of workload. These giants can afford to accept thinner margins just to win market share, and that leverage could bite CoreWeave when it comes time to renegotiate with Vast in a few years.
When the Hype Outruns the Clock
Let’s be clear: none of this means the technology isn’t real. Unified flash storage genuinely solves pain points for AI workloads. The bigger question is about timing. Are infrastructure investments being made faster than actual AI monetization can justify?
If big enterprises take longer to deploy those $100 million language model projects than expected, we’ll see a ripple effect. Overcapacity builds up, vendors face slower growth, and valuations start to wobble. When that happens, the companies with concentrated customer bases—those relying heavily on just a few contracts—are usually the first to feel the pain.
The smart money may still flow toward NVIDIA, which wins no matter who else stumbles, or toward memory manufacturers who benefit from the broader shift to flash storage. The riskier bets? Late-stage private infrastructure firms that keep pricing themselves as if every new megadeal is guaranteed. That’s the classic red flag of a late-cycle market—when deals are expected, not celebrated.
The CoreWeave–Vast partnership will probably deliver solid results over the next few years. The question that lingers isn’t whether the AI future arrives—it’s what happens if it just takes a little longer than everyone planned.
NOT INVESTMENT ADVICE