AI Infrastructure Investment Signals Drive US Growth Equity Market to USD 21.2 Billion Through Contracted Demand

The US growth equity market surged to USD 21.2 billion in October 2025, reflecting a shift towards conviction-based investing. Shaurya Mehta highlights that AI infrastructure and contracted demand serve as critical signals. As investors prioritise measurable data over narratives, long-term commitments in compute markets and repeat usage patterns now define the sector's sustainable growth.

Growth equity has started moving like a conviction market again. On Monday morning investment committee calls, the debate is less about story and more about what the spreadsheet can actually prove.

In the U.S., October 2025 saw $21.2 billon raised across 60 deals of $100m or more, up from $15.9 billion across 45 such deals in October 2024. The volume hides a tighter filter underneath. Investors are not just asking what is exciting, they are asking what can be measured. Shaurya Mehta, an AI infrastructure investor and Fast Company Board, builds his work around that distinction, preferring signals that show up in contracts, usage, and prices rather than in pitch decks. What ties the next few shifts together is simple: the cleanest clues are the ones that force a decision, whether that is a price, a contract, or repeat usage. To understand how teams are handling signal, noise, and conviction across today's AI markets, we spoke with Shaurya Mehta.

"Prediction markets are useful when they force you to price uncertainty, not narrate it," says Mehta. "When the price moves, you have to ask what information just got absorbed." That mindset carries into the rest of his work, where belief is only interesting once it shows up in behavior that can be observed.

Priced Belief As A Signal Layer

One prediction market platform was in early discussions with investors at a valuation between $12 billion and $15 billion. Mehta points to examples like this because they show what "signal" looks like in its purest form: uncertainty is forced into a number that moves when the world changes. He recalls a late night watching odds shift sharply after a single data release that most headlines treated as minor. The story did not change, but the price did. That was the signal. He applies the same instinct in AI, where the strongest signals tend to show up in what customers commit to and what they keep using.

Contracted Demand Replaces Benchmarks

If prediction markets are about priced belief, compute markets are about contracted need. Teams used to compare hourly rates. Now they sign for capacity like it is power. A recent strategic infrastructure agreement made that shift plain, with a contract value of $10+ billion and a related equity issuance of $350.0 million. For Mehta, moments like this are when the market stops debating hypotheticals and starts recording priorities in writing. "When everyone is short at once, this stops looking like a commodity," he says. "It behaves like a capacity market because demand shows up as signed commitments, not as experiments. When buyers commit for years, you learn what they will not risk."

Product Pull In The Model Layer

Compute can be booked. Models have to be adopted. That is where signal gets harder, because the product is behavior change, and behavior change is easy to overstate. One leading foundation model platform has said it reached $10 billion in annual recurring revenue and is serving about 3 million paying business customers. Mehta’s read is that adoption becomes meaningful when the tool stops being a trial and starts becoming a default, when teams return to it under time pressure and pay for it because it is embedded in real work. He mapped the ecosystem, tracked product releases, and refreshed his view as usage patterns evolved, separating durable pull from short-lived hype. His work as a judge and mentor for Stanford’s MS&E 272: Entrepreneurship without Borders without Borders also keeps him close to the moment when a prototype stops being a demo and starts becoming a habit. The signal is rarely the first use. It is the second week.

The Next Signal Arms Race

The instruments are getting sharper. More capital is flowing, but the bigger change is the competition to measure truth earlier, and with less noise. Beyond his investing work, Mehta has also written about the structural forces shaping the AI economy. In his book Scientific and Economic Acceleration: Building the Infrastructure for the Age of AI, he examines the systems required to support large scale intelligence production, including compute supply, energy availability, capital formation, and geopolitical constraints. The book frames artificial intelligence not simply as software progress but as a new infrastructure layer, comparable to electricity or the internet, where breakthroughs in science translate into physical capacity decisions. That perspective mirrors the signals Mehta tracks in markets today, where the most reliable indicators often emerge from infrastructure commitments rather than product announcements.

The GPU as a service market was valued at $4.31 billion in 2024 and is projected to grow from $5.79 billion in 2025 to $49.84 billion by 2032. Those figures describe the backdrop against which capacity decisions are being made, and why signal now often starts in contracts and repeat usage rather than in product polish. Mehta’s work has included referring potential customers to compute providers when constraints were limiting real roadmaps, not just experiments, and staying engaged after investment as markets normalized and behavior stabilized. His judging role for the Stanford x DeepMind Hackathon reinforces the same lesson in a different form: the future signal is already present, but it hides inside what builders do repeatedly when nobody is watching. "The future signal is already present," he says. "You just have to be early enough, and disciplined enough, to believe what you can measure."