The Innovation Economy Is Getting Smaller. Its Partnerships Are Getting More Important.

In 2025, roughly 40 cents of every dollar invested in startups on Carta's platform went to an AI company. By early 2026 it was 54 cents. Average headcount at Series D has fallen 29 percent from its 2023 peak. The median seed-stage team is now four people.

The standard read is that AI lets startups do more with fewer people. That is true, but it is too narrow. Smaller does not mean more self-contained. In many cases it means the opposite. An AI-native company may carry fewer employees, but it leans harder on everything around it: model providers, chip suppliers, sponsor banks, payment processors, enterprise customers.

The company is smaller. The dependency map is larger.

That matters most in financial services, where few companies scale by software alone. Banking access, payments connectivity, lending capacity, compliance operations, and customer trust are usually delivered through partnerships, and regulators treat them accordingly. The federal banking agencies' third-party risk guidance frames these relationships as lifecycle risks, with due diligence, ongoing monitoring, documentation, and board oversight where the risk is material.

Which raises a question the industry has mostly avoided: how should those partnerships be priced?

For decades, the innovation economy used equity to align employees and investors. Employees received options because they built the company. Investors received shares because they supplied capital and took risk. But the partners who delivered cloud capacity, banking infrastructure, payments access, or enterprise adoption got paid in cash. Cash is necessary. But cash alone often under-aligns the relationship. A supplier committing scarce capacity to an early-stage company is making a judgment about future value. A bank supporting a fintech program is contributing credibility, controls, operational infrastructure, and balance-sheet access. A strategic customer that drives adoption can change the issuer's trajectory. In each case, the partner may help create enterprise value that goes well beyond the invoice.

That is where warrants enter.

A warrant is the right to buy shares at a set price in the future. Economically it sits close to an employee stock option, except the holder is a commercial partner rather than an employee. Used well, it lets a partner participate in the upside it helps create. Used badly, it becomes a hidden discount, a governance problem, a dilution leak, or a piece of legal clutter nobody manages after signing.

I have spent a career in and around banks that held these instruments, through cycles where the warrant book quietly mattered more than anyone had planned. The warrants that caused trouble were almost never the ones that were negotiated hardest. They were the ones nobody looked at again after signing, until an audit, a financing, or an acquisition forced the question of what exactly had been granted, and why, and what it was now worth.

The disciplined version is simple. The cash terms stand on their own. The warrant is additive, tied to a real commercial relationship, sized against expected value, with a defensible strike price, a hard cap, objective vesting, and no control rights that distort governance. It rewards the partner only if value is created. The bad version is what happens when any one of those is missing. This is the same lesson boards learned with employee stock options: powerful because they align incentives over time, dangerous when treated as free.

The largest technology companies are running this playbook in plain sight. In October 2025, AMD issued OpenAI a warrant for up to 160 million shares as part of a multi-year agreement to deploy 6 gigawatts of AMD GPUs, with vesting tied to deployment and purchase milestones and AMD share-price targets, and exercise further conditioned on OpenAI meeting defined technical and commercial milestones. In February 2026, AMD struck a similar structure with Meta: another performance-based warrant for up to 160 million shares, vesting as purchases scale to six gigawatts. In August 2026, Marvell disclosed a warrant issued to Google for roughly 59 million shares at $206.58 per share in connection with a custom semiconductor agreement, most of it vesting in 240 equal tranches, one for each $500 million in custom product revenue through fiscal 2033.

Fintech has run the same playbook for years. Affirm granted Shopify warrants exercisable at a penny per share alongside their installment program agreement. Adyen issued eBay warrants for up to 5 percent of its fully diluted share capital, vesting on processing-volume milestones. When the first tranche vested in 2021, eBay paid roughly $110 million in cash for Adyen shares valued at approximately $1.1 billion. Marqeta has disclosed customer warrants tied to annual transaction-count thresholds and new-cardholder milestones.

The pattern is consistent. When one party's commitment can materially shape another party's growth, equity-linked alignment is more precise than a discount and more durable than a handshake.

That does not mean every partnership should include warrants. Most should not. Equity is too important to become a routine concession. A warrant makes sense when the partner can materially affect company value, the outcome can be measured, the dilution can be bounded, and the company can explain why the economics are fair. It is strongest where the partner controls distribution, capacity, transaction volume, or a strategic customer path the company could not easily buy elsewhere. It is weakest as a vague sweetener.

For the institutions on the other side of the table, the discipline is a mirror image. A bank or enterprise that holds warrants should be able to show that its credit and pricing decisions hold up on their own, that the warrant added upside rather than filled a gap, and that someone owns the position for its entire life: the vesting triggers, the corporate actions, the financing three years from now that changes everything. A warrant is not a sweetener on that side either. It is an asset with a lifecycle, and it deserves the same rigor as any other asset on the books.

AI will keep making teams smaller and dependency maps denser. More of the value created in the next decade will sit across the edges of companies rather than inside them.

The firms that understand that will stop treating partners as vendors only.

They will treat alignment as infrastructure.