Everyone's Building AI Agents. Nobody's Figured Out How They Get Paid.

By Susana Khan, CMO, TODAQ February 27, 2026 · 6 min read

It’s been a noisy week in AI. And not the usual noise — breathless product launches and model benchmarks. This week’s headlines felt more structural, like the ground shifting underneath the whole industry.

OpenAI closed what is being called the largest private fundraise in history: $110 billion, valuing the company somewhere between $730 and $840 billion depending on who’s doing the counting. Amazon, Nvidia, and SoftBank are among those writing the cheques. Meanwhile, Block announced it’s cutting roughly 40% of its headcount to lean aggressively into AI-driven operations. Anthropic’s ongoing conversations with the Pentagon are generating debate about where AI’s ethical red lines should sit. Nvidia continues to print money as every major AI lab races to secure compute.

Each of these stories is significant on its own. Together, they point at a single underlying question that the industry hasn’t fully answered yet: when AI agents are transacting at scale, what does the payment infrastructure actually look like?

It’s the question we’ve been building toward for nearly a decade.

$110 Billion Buys a Lot of Compute. It Doesn’t Buy a Payment Rail.

OpenAI’s raise is, at its core, a bet on Agentic AI — autonomous systems that don’t just respond to prompts but take actions, make decisions, and execute transactions without a human in the loop. The capital goes toward 3 gigawatts of inference capacity, 2 gigawatts of training infrastructure, and deeper integration with cloud providers.

That is a staggering amount of computational capacity being directed at autonomous agents. And here is what that implies at a practical level: those agents will be transacting constantly. Script generation, data retrieval, model invocations, image synthesis, audio transcription; every interaction carries a cost, and every cost requires settlement.

Two types of infrastructure are being positioned as the answer. Neither actually solves the problem.

The first is legacy payments: credit cards, batch billing, net-30 invoicing. The industry has tried to paper over their limitations with batching: aggregate thousands of micro-interactions, settle them together, reduce the per-unit fee burden. It’s a workaround, not a solution. Batching still isn’t true pay-per-use. It re-introduces latency, requires reconciliation, and forces agents into settlement cycles that have nothing to do with the speed at which they operate. The underlying infrastructure was designed for monthly human billing, and no amount of aggregation changes that architecture.

The second option gaining traction is crypto-native rails: protocols like Coinbase’s x402, which use Bitcoin’s Lightning Network or Stablecoin layers to enable machine-to-machine payments. The instinct is sound: the problem is real, and something genuinely new is needed. But gas fees on every transaction, however small individually, become a structural levy across billions of calls. More fundamentally, most enterprises, most developers, and most AI companies aren’t crypto-native. They run on commercial banking infrastructure and settle in dollars, and requiring them to onboard an entirely separate financial layer to access payment rails is friction that most won’t accept.

TAPP was designed from the ground up for neither of these camps. It settles in USD, connects natively to commercial banking rails, and requires no crypto wallet, no token, no gas fee. The payment is embedded directly in the API request which means settlement and service delivery happen simultaneously, in real-time. No batch cycle. No reconciliation. No intermediary taking a cut at every step.

For the businesses using our infrastructure, that translates into payment costs running at a fraction, up to 95% less, than what they’d face on any alternative rail. That’s not a theoretical efficiency; it’s what we’ve measured in production. A three-cent transaction becomes genuinely economical rather than a loss absorbed in the hope of volume.

One of our early customers came in to test the pipes with a few dozen API calls per hour. Within days, they added more assets, more APIs, and were scaling 40-50x. Machines settling with other machines in real time, automatically, with no human involvement. Money moved when value moved. The infrastructure held because it was built for exactly that load. No ceiling, no workarounds required.

That is what the AI economy’s payment layer should look like. Not legacy infrastructure adapted under pressure, and not crypto rails with the roughest edges filed down. Something that treats real-time machine commerce as the primary use case, not an edge case to be accommodated.

Block’s Cuts and the Embedded Finance Signal

Block shedding 40% of its workforce while doubling down on AI-driven operations reads, on the surface, as a cost story. But it’s also a signal about where fintech is heading: toward leaner, more automated stacks where AI handles fraud detection, transaction optimization, and operational intelligence at a fraction of the traditional headcount cost.

The irony is that as payments infrastructure becomes more automated, the transactions themselves are becoming more granular. AI systems making per-query fraud checks, agents executing real-time cost arbitrage across cloud providers, machine-to-machine settlements for compute resources. These are not the kind of flows that legacy infrastructure was designed for.

The shift toward embedded, usage-based payments isn’t just a startup opportunity. It’s becoming a structural reality for every company in the payments space, large or small.

Traction, and What Comes Next

We’ve had a busy few weeks ourselves, and we want to be transparent about that.

Interest in TODAQ’s infrastructure has accelerated meaningfully; from AI companies looking to monetize their services at the API level, fintech builders exploring real-time micropayment rails, and from investors who see the same convergence we do. We’re not announcing a round today, but we’re in active conversations, and we’ll have more to say soon.

What we can say now is that the production evidence continues to build. The same infrastructure we validated in video streaming, where 90% of transaction volume became AI-to-AI with no human initiation, is expanding across new verticals. Each one with the same underlying requirement: settlement that keeps pace with delivery, at any transaction size, with no floor.

This is what we mean when we say TAPP is a foundational layer rather than a vertical product. The architecture is the same regardless of industry. The problem it solves is always the same: the moment value changes hands, payment should too.

The Quieter Infrastructure Story

The loudest AI headlines tend to centre on models, funding, and geopolitical friction; Anthropic’s Pentagon discussions, OpenAI’s valuation, the compute arms race, and so on. These are unquestionably important stories.

The less glamorous story, the one that tends to emerge later, is always about infrastructure. Who built the rails that all of this runs on? Visa didn’t make news when it was scaling card rails in the 1970s. AWS didn’t generate headlines proportional to its eventual importance when it launched in 2006. Stripe was considered a developer curiosity before it was considered foundational.

The AI economy is generating billions of machine-to-machine transactions today, and that number grows by an order of magnitude with every major deployment cycle. The window for building the infrastructure layer for this economy is not permanently open. It closes when a dominant architecture becomes entrenched.

We’re not pitching that architecture. We’re running it. The video market gave us the proof. The AI economy is where it scales.

TODAQ is a pre-seed US startup building internet-native payment infrastructure for the AI economy. If you’re an AI company, creator, or investor who wants to understand what we’re building — reach out at hello@todaq.net.

Follow along here for updates as we expand across industries in the weeks ahead.