AI revenue has crossed an important financial threshold
Artificial intelligence is beginning to generate enough revenue to cover the estimated depreciation cost of the chips and data centres supporting it.
Global AI sales outside China reached approximately $25 billion in the first quarter of 2026, according to research firm Exponential View. Estimated depreciation tied to AI computing infrastructure totalled about $21 billion during the same period.
Revenue has now exceeded depreciation for two consecutive quarters.
The result offers early support for the argument that the enormous capital investment behind generative AI is being met by real customer demand. It does not establish that the industry is broadly profitable, but it moves the debate beyond a market supported entirely by infrastructure spending and investor expectations.
The economics are improving, but the cushion remains small
Depreciation alone consumed more than two-thirds of the quarter’s estimated AI revenue.
The remaining revenue must still cover electricity, data-centre operations, engineering talent, model training, networking, interest expense and the cost of continually improving AI products.
A modest change in pricing, demand or infrastructure utilization could therefore move parts of the industry back below the economic threshold.
Large technology companies can absorb that volatility through profitable cloud, advertising and software businesses. Smaller infrastructure providers and heavily financed AI cloud companies have less room for error.
Many of those companies are funding expansion through equipment leases, secured debt and equity issuance. If demand slows or computing prices decline faster than expected, losses could increasingly shift from corporate balance sheets into capital markets.
The spending commitment remains enormous
Meta, Alphabet, Microsoft and Amazon are expected to spend as much as $725 billion on capital expenditures this year, with a substantial portion directed toward AI chips, servers, networking systems, power infrastructure and new data centres.
That represents one of the largest corporate investment cycles in history.
The amount being spent does not need to translate into immediate operating profit to be economically rational. Infrastructure is normally built ahead of demand, particularly when companies expect a technology platform to support years of future growth.
The latest revenue figures suggest utilization is developing quickly enough to support the early stages of that investment. They do not yet prove that every planned data centre or AI accelerator will earn an acceptable return.
AI adoption appears to be scaling faster than previous technology cycles
Exponential View estimates that generative AI produced approximately $110 billion in revenue outside China during the past 12 months.
According to the report, the market is scaling roughly three times faster than previous major information-technology waves, including the internet, mobile applications and cloud computing.
The estimate is based on spending data collected across more than 1,000 companies. Sources included corporate filings, executive statements, cloud-provider disclosures and press reports, with adjustments intended to prevent the same revenue from being counted at multiple points in the supply chain.
That methodology is important because AI spending moves through several layers. A customer may pay an application provider, which then pays a model developer, which in turn purchases cloud computing backed by Nvidia chips. Counting every payment as separate end demand would overstate the industry’s true economic scale.
Depreciation remains the most contested assumption
The analysis assumes a six-year useful life for information-technology equipment, including GPUs.
That assumption has become one of the most debated accounting issues in the AI investment cycle.
New chips are arriving quickly, with each generation offering better performance and energy efficiency. Critics argue that older accelerators could lose economic value well before their accounting schedules end, forcing companies to replace equipment sooner or recognize writedowns.
A shorter useful life would increase annual depreciation expenses and make current AI economics look less attractive. It could also pressure earnings at hyperscalers that have extended the depreciation schedules used for servers and networking equipment.
The risk is not that older chips suddenly stop working. The concern is that newer processors may become sufficiently cheaper or more efficient that earlier generations are no longer competitive for the most demanding workloads.
Older GPUs are retaining more value than expected
Current market data does not show an immediate collapse in the value of older AI processors.
Hourly rental prices for Nvidia’s H100 remain close to 80% of their original level, even as the chip enters its fourth year in the market. In some cases, prices have increased as demand for computing capacity exceeded the available supply of newer Blackwell systems.
Amazon Web Services has also said it continues to operate its six-year-old A100 servers because customers still want the capacity.
Those examples support the case for longer depreciation schedules. Newer chips may handle the most advanced training and inference workloads, while older accelerators remain useful for less intensive models, scientific computing, fine-tuning and routine enterprise applications.
A growing AI market can absorb several generations of hardware simultaneously, particularly when demand for computing continues to exceed supply.
More revenue does not guarantee strong returns on every data centre
The aggregate figures describe an industry moving toward sustainability. They do not show how evenly the economics are distributed.
Nvidia and other infrastructure suppliers recognize revenue when chips and systems are sold. Cloud providers earn revenue as customers rent computing capacity. Model developers and application companies must then generate enough income to support those infrastructure costs while funding their own operations.
Some parts of that chain may produce attractive margins while others struggle.
The greatest risk sits with companies that build capacity based on aggressive demand assumptions and finance it with large fixed obligations. If computing prices decline, customers migrate to more efficient models or utilization falls short of forecasts, those operators may face pressure even as the wider AI market continues growing.
Cheaper models are changing how customers buy AI
Developers are increasingly distributing workloads across a wider range of models.
OpenRouter data cited by the report shows that the share of tokens processed by Google, OpenAI and Anthropic models fell to 33% in June 2026, down from 72% a year earlier. Open-weight and Chinese models, including DeepSeek, gained a larger share of activity.
The shift does not necessarily mean the leading model developers are losing their most valuable customers. Token volume and revenue are not the same. Premium models can still command higher prices for complex coding, research and reasoning tasks.
Routine work is becoming more price-sensitive. Companies do not need the most advanced and expensive model to extract information from a receipt, categorize an email or complete a simple support request.
Lower-cost models can handle those tasks while premium systems remain reserved for work where accuracy and reasoning justify higher prices.
Premium model providers will need stronger differentiation
As capable models become cheaper and more widely available, technical performance alone may not support premium pricing.
Leading providers will need to compete through enterprise security, proprietary data connections, workflow integration, reliability and broader software ecosystems. Customer retention may depend increasingly on how deeply a model is embedded in a company’s operations rather than its position on a public benchmark.
That shift could strengthen companies able to combine advanced models with cloud infrastructure, productivity software and industry-specific applications.
It could also place pressure on standalone model developers whose expenses remain high while lower-cost alternatives narrow the performance gap.
The AI investment debate is moving into a new phase
The first stage of the AI boom was defined by supply: GPU orders, data-centre construction, power agreements and record capital budgets.
The next stage will be judged by utilization and revenue.
Investors will need to determine whether AI workloads remain large enough to keep new infrastructure productive, whether prices support acceptable returns and whether the useful lives assigned to chips reflect economic reality.
The first-quarter figures provide encouraging evidence. Customer spending is now large enough to cover estimated depreciation at the industry level.
The margin remains too narrow to declare the investment cycle fully validated.
WSA Take
AI demand is beginning to catch up with the infrastructure built to serve it.
Crossing the depreciation threshold for a second quarter is an important milestone because it shows that revenue is no longer negligible relative to the capital committed. Continued demand for older GPUs also weakens the most aggressive argument that AI equipment becomes economically obsolete within only a few years.
The industry still faces a demanding financial test. Depreciation absorbs most current revenue, capital spending continues to accelerate and smaller infrastructure providers are taking on significant financing obligations.
For now, the economics are improving rather than breaking. The next question is whether AI revenue can continue scaling fast enough to create a meaningful return on the hundreds of billions of dollars still being invested.
Disclaimer
WallStAccess is a financial media platform providing market commentary and analysis for informational and educational purposes only. This content does not constitute investment advice, a recommendation, or an offer to buy or sell any securities. Readers should conduct their own research or consult a licensed financial professional before making investment decisions.
