March 2026: Where $122B Went and Why

March 2026 did not just break records. It rewrote the frame entirely. OpenAI closed a $122 billion funding round, the largest private raise in history, while Nscale secured $2 billion in a Series C and Shield AI locked in another $2 billion of its own. Q1 2026 ended with $297 billion in total venture capital deployed globally, a number that makes the 2021 peak look restrained. The message from institutional capital is not subtle: the infrastructure era of AI is here, and the cheque sizes reflect it.

Context matters. OpenAI's scale is partly a function of a structural shift in how hyperscalers commit capital. Microsoft's multi-year Azure capacity agreement, mirrored by Amazon and Google through AWS and Google Cloud for Anthropic, means that the largest rounds are no longer pure equity plays. They are compute commitments dressed as funding. This reframes what a "raise" even means at the frontier tier, and it raises legitimate questions about how LPs should model exposure when cloud credits constitute a meaningful portion of round value.

Below the headline number, the pattern is equally telling. Nscale, a London-based full-stack AI infrastructure provider, raised $2 billion at Series C, signaling that compute and data center infrastructure is attracting growth-stage conviction, not just strategic backing. AMI Labs pulled in $1 billion at seed, which is a figure that would have been absurd in any prior cycle and now barely registers as unusual. Across 20 or more verified mega-rounds in a single month, capital is clustering tightly around AI compute, infrastructure, and the picks-and-shovels layer that sits beneath the model builders.

For founders and investors outside the frontier tier, the concentration has real consequences. When institutional LPs deploy at this velocity, portfolio construction changes. Allocation that might have seeded 50 early-stage companies in 2022 is now anchoring one infrastructure round. Smaller funds face pressure on their LP base as capital gravitates toward perceived certainty. The market is not broadly hot. It is hot in a narrow band, and that band is defined almost entirely by proximity to AI compute and model development.

The next 12 months will test whether the concentration holds or begins to diffuse. If frontier model performance plateaus or compute costs compress faster than expected, the infrastructure premium built into these valuations will face pressure. More likely, a second wave of capital begins flowing into applied AI, companies using the infrastructure already funded to build in healthcare, defense, and climate. That is where early-stage founders should be positioning now, not competing for the compute layer, but building what becomes valuable once it is cheap and abundant.