A frenzy for power and GPUs can create scarcity, but the more consequential shortage is institutional. The AI economy still lacks the commercial and financial rails that make compute measurable, priceable, billable, benchmarkable and financeable.

This scarcity, both real and perceived, can justify vast capital expenditure, drive breathless valuations and make every available megawatt feel strategic. But none of that, by itself, creates a mature market. Markets do not emerge simply because demand explodes. They emerge when participants can agree on what is being sold, how it is measured, how it is priced and whether the resulting cash flows can be trusted.

While the world marvels at the sheer power of GPUs and the scale of new data centers, all too often the financial infrastructure that underpins them is still running on spreadsheets and PDFs. It is this systemic friction that has resulted in a landscape where billing disputes and revenue leakage have simply become the norm.

This is the challenge that is being addressed by Internet Backyard – a San Francisco-based firm that’s automating quoting, billing, payments and financial workflows for the AI industry. Its bet is that whoever builds the financial layer will help turn compute from a land rush into a real market.

Legacy billing systems

Inevitably, the core of the problem lies in the transition from traditional cloud computing to the much more specialized demands of AI. For years, the industry relied on relatively static billing models. However, the move towards neoclouds and high-performance inference systems has exposed deep flaws in the legacy financial stack.

Certainly, traditional ERP (Enterprise Resource Planning) and CRM (Customer Relationship Management) systems are ill-equipped to handle the granular, millisecond-level usage data generated by modern GPUs. Statistics suggest up to 80 percent of data center operators experience recurring billing disputes, reflecting a total breakdown in the order-to-cash lifecycle.

“The weakness in price discovery isn’t just confined to sales, it spills directly into billing, forecasting, financing and valuation,” says Mai Trinh, CEO and co-founder of Internet Backyard. “When commercial terms live in PDFs, the deal history lives in the CRM, invoices are generated in another system and usage data arrives on different clocks and in different formats, revenue stops behaving like infrastructure-grade cashflow.”

According to Trinh, because compute infrastructure is capital intensive and often debt financed, operational ambiguity becomes a financial risk, resulting in lenders pricing in uncertainty. “If you need to demonstrate predictable cash flow to your financiers or lenders and all of your systems are manual, you have a lot of noise and risks in your financial model, which makes it harder to secure capital.” She adds: “In infrastructure, the quality of the cashflow matters nearly as much as the quantity.”

Preventing revenue leakage

When billing is inaccurate or simply incomplete, the consequences extend well beyond simple administrative headaches. For data center operators, revenue leakage – the uncaptured value of utilized resources – can significantly impact margins in an industry where capital expenditure (Capex) is measured in billions.

According to Trinh, there are really two layers to revenue leakage. Firstly, there are disputes over energy usage, which often ‘require customers to dig through their operating systems to prove to their clients that they really did use that much energy’. Then there’s another issue around untapped revenue.

“If you already have the network to power the GPU and you’re not using it, there’s additional money to be made there,” says Trinh. She adds that while the majority of data centers currently run at around a 98 percent occupancy rate, utilization rates are as low as three and five percent, with some clients – particularly in academia – not using any GPU capacity at all for months at a time because they are focusing on writing papers instead. “That means you have so much more capacity that you could sell back to the market or exchange.”

One problem, though, is that because there isn’t any standardization in the industry, there isn’t a marketplace where people can upload their resources into an exchange. “As a result, clients are simply leaving money on the table and bearing huge sunk costs,” she adds.

Importance of standardization

Clearly, what the AI industry needs are standardized prices, in much the same way as a commodity such as oil has. “What we’re lacking in the AI industry is a standardized benchmark,” explains Gabriel Ravacci, CTO and co-founder of Internet Backyard. “There’s not a Brent Crude price for GPU, which means you can’t do a forward curve, you can’t do derivatives, and you can’t hedge the price exposure when you buy these things.” He adds: “Where we are right now with compute is really where we were with oil before the advent of NYMEX (New York Mercantile Exchange).”

Without a standard metric, a compute hour in one facility may offer a vastly different performance or reliability than another, though they are often billed identically. And until there is a standardized unit to trade, the market for compute remains a series of isolated, opaque, and inefficient silos that prevent the true financialization of AI.

That said, Ravacci expects a commodity market for compute to develop soon. “Meta has already announced a trading desk, and you are going to see places where contracts can be exchanged using Blockchain for encryption very soon.”

Automating order-to-cash lifecycle

According to Internet Backyard’s Trinh, for now compute pricing remains a ‘blend of rack-rate theatre, competitor gossip and rough markup heuristics’. “That’s not a serious operating model for an industry trying to finance billions of dollars of infrastructure.”

A credible price has to reconcile three layers at once. Firstly, the physical layer of hardware, power, cooling, networking, and utilization. Secondly, the financial layer of capex recovery, debt service, and margin protection. And thirdly the market layer of spot signals, benchmark ranges and competitive context.

Bring those together and pricing starts to look less like guesswork and more like underwriting. That is a central part of gnomos, Internet Backyard’s first step towards much greater transparency in compute pricing. Aimed squarely at the builders and operators of modern-day data centers, gnomos is a financial operations agent that’s much more than a modern quote-to-cash tool.

In its fuller form, the product will detect underpriced contracts, surface revenue leakage, spot renewal and expansion opportunities, fold energy and timing into economic decisions, and compile financing-ready operating profiles. For flexible workloads, the logic can eventually extend to energy-aware orchestration: not just billing what happened, but helping decide when and where work should run so margins improve.

“It’s an autonomous commercial and financial operating system for compute providers,” explains Trinh. Its premise is not that teams need another dashboard full of alerts and recommendations, but that routine commercial work should increasingly be completed before the human even opens the screen.

“The biggest lesson we’ve learned so far is that data center operators really don’t need another dashboard. They want a system that has full context about their operations, autonomous decision making and executive agency to turn their entire operation into billable revenue,” affirms Trinh.

However, this is just ‘phase 1’ of the platform’s roll-out. Internet Backyard’s goal isn’t just better pricing, but to provide its customers with the visibility they need to forecast revenue and costs, ultimately helping them identify upsells and gaps in the market.

According to Trinh, in compute infrastructure, a quote is not merely a sales event. It is a capacity commitment, a pricing decision, a billing object, a support obligation, a deployment workflow, and sometimes the beginning of a financing case. “That is why Internet Backyard’s long-term vision matters,” says Trinh.

“Over time, gnomos is meant to become the connective tissue linking pricing, quoting, contracting, billing, internal handoffs, support, collections, utilities, tax treatment, and financing workflows,” she explains. “Put simply, it wants to turn commercial complexity into operational follow-through, and operational follow-through into reliable cash.”

Internet Backyard is currently testing gnomos with Vancouver-based AxiNorth, a technology company focused on AI and edge computing in the data center space, and is also working with all of the inference providers and data center operators. “They’re active, creative and collaboratively working on product design, features requests, testing and feedback,” says Trinh.

Providing future financial stability

Undoubtedly, the current trajectory of the AI boom is unsustainable without a corresponding revolution in financial infrastructure. As training costs soar toward the billion-dollar mark, the industry can no longer afford to operate in a manual PDF and Excel spreadsheet world.

Ultimately, by creating a transparent, tokenized unit of value, Internet Backyard is hoping to transform compute from a volatile utility into a mature, bankable asset class – one built on a foundation of absolute financial clarity, rather than administrative friction and guesswork. As Mai Trinh concludes: “The next winners in AI will not only be those who can generate or consume compute. They will also be those who can measure it cleanly, price it coherently, bill it accurately, benchmark it credibly and finance it at scale.”

For more information on its thesis, please visit Internet Backyard.