Axios Senior AI Reporter Madison Mills laid out an investor debate on CNBC on Thursday, August 13, framing NVIDIA’s newly announced $500 billion financing agreement with top Wall Street firms as either the largest asset-class creation event of the AI era or the most sophisticated circular-financing structure yet. Her reporting places NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) at the center of a market that is racing to securitize compute itself.
Mills said investors she spoke with believe the $500 billion package isn’t big enough yet: “Some of them say the $500 billion isn’t even enough. It’s going to put online about ten gigawatts, which is what we’re expected to need next year alone,” she told CNBC.
Nvidia reported Q1 FY2027 revenue of $81.615 billion, Data Center revenue of $75.25 billion (up 92% year over year), and disclosed total supply-related commitments of $119.0 billion. CEO Jensen Huang called the current buildout “the largest infrastructure expansion in human history.”
Wall Street Is Turning Compute Into a New Asset Class
Mills discussed how compute is quickly becoming a tradable asset. “A year ago I covered this startup… which is these 20-somethings in an apartment who are doing $1 billion worth of GPU trading from this small townhouse in the course of just a month. And we know that CME, ICE, all these companies are trying to get in on GPU futures spot trading. OpenAI is hiring finance team leaders that are going to work on GPU securities,” she said.
CME Group (NASDAQ:CME) CEO Terry Duffy specifically flagged “compute futures” alongside Single-Stock futures and Treasury clearing as an innovation area in the company’s Q2 2026 remarks. CME posted Q2 EPS of $2.99 on revenue of $1.71 billion, with record market data revenue of $238 million, up 20% year over year.
Intercontinental Exchange (NYSE:ICE) delivered adjusted EPS of $1.90 on $2.67 billion in revenue, with CEO Jeff Sprecher noting that “markets become more global, digital and continuous.”
The pitch to investors is that securitizing GPU capacity effectively turns compute into a tradable commodity comparable to crude or power spreads, and would spread financing risk across pension funds, retirement accounts, and hedge funds.
The Bear Case: What Happens if Better Models Need Less Compute?
Mills laid out the counterargument that could crush compute futures while the asset class is still in its infancy: “One investor who I talked to yesterday told me he’s worried about something called diminishing model returns. This idea is that we already have these really amazing models. And it’s not clear that the AI labs are going to need as much compute going forward to train better and better models,“ she said. If frontier model improvements plateau, the collateral value of leading-edge chips could compress well before the debt underlying compute securities matures.
She pointed to a live example. “Look at what’s happening with Google. They’ve already decided to kind of, for better or worse, roll back their AI ambitions because they went free cash flow negative,“ Mills said. A hyperscaler pulling back on capex is the kind of signal that would erode long-dated compute demand assumptions.
What the Market Is Pricing
The market clearly believes the demand for AI compute is real. The larger unanswered question is whether NVIDIA and Wall Street are creating a durable new asset class or using increasingly complex financing structures to sustain a buildout whose economics remain unproven. Either way, some experts expect that even a $500 billion financing package may cover only a fraction of what the AI industry will need.
Contact [email protected] for any questions or corrections.
Top Gaining Stocks
Top Losing Stocks