Executive Summary

For AI founders, choosing where to compete determines how much capital they need, how quickly they can reach customers, and what evidence they must produce to build a credible business. Energy, chips, infrastructure, models, and applications each have different economics. Access to data and the ability to earn customer trust influence success across all five layers.

This guide examines financing activity, customer spending, and the constraints shaping opportunity across the AI stack. It draws primarily on full-year 2025 data and research published in 2026. These sources measure different aspects of the market and do not establish how many startups operate in each layer. Their findings nevertheless reveal important differences between where founders can enter a market, where investors commit capital, and where customers are spending.

Application founders can build on existing models and infrastructure, reducing the initial cost of testing a product. Financing, however, is highly concentrated: CB Insights reports that private AI companies raised approximately $225.8 billion in 2025, with 79% going to rounds of $100 million or more and 41% to large language model developers. Customer demand is also substantial. Menlo Ventures estimates enterprise generative-AI spending at approximately $37 billion in 2025, including roughly $19 billion on applications, within the categories covered by its market model. The model is informed by a survey of U.S. enterprise decision-makers.

Growth also depends on physical capacity. The IEA estimates that global data centers consumed approximately 415 TWh of electricity in 2024 and projects roughly 945 TWh by 2030 in its base case. These figures include AI and non-AI workloads. They highlight why power availability, chips, cooling, and grid access can constrain expansion even as AI software becomes cheaper to use.

The practical implication is that founders must match their strategy to the economics of their layer. Application companies need evidence of recurring customer value, retention, distribution, and defensibility beyond model access. Companies building chips, physical infrastructure, or frontier models must also demonstrate technical performance, access to scarce resources, and a credible financing path. Across the stack, the strongest opportunity is one where a company can solve an important problem and retain a durable share of the value it creates.

1. Methodology and Definitions

This guide synthesizes published research to explain the opportunities, economics, and constraints facing AI founders. It draws primarily on full-year 2025 data and reports published in 2026. It does not include a proprietary Peachscore startup dataset or independently measure company formation, performance, or outcomes across the AI stack.

The analysis organizes the ecosystem into five primary layers: Energy, Chips, Infrastructure, Models, and Applications. Data and trust influence multiple layers and are examined throughout the guide. These categories provide an analytical framework rather than mutually exclusive company classifications; a business may operate across several layers.

Each source answers a different question:

Investment, customer spending, company counts, and electricity demand are evaluated separately. They are not combined into a single market total or used to calculate startup shares by layer. Conclusions about opportunity and defensibility represent interpretations of the cited evidence, rather than measured rankings or proof of causation.

2. The AI Value Chain

The AI economy depends on five interconnected layers. Energy powers computing systems; chips perform the computation; infrastructure makes that capacity available and usable; models provide AI capabilities; and applications apply those capabilities to customer problems. A shortage, cost increase, or performance improvement in one layer can affect the businesses that depend on it.

Data and trust influence how effectively these layers work together. Access to relevant, legally usable data affects model performance, product quality, and competitive advantage. Security, evaluation, privacy, auditability, and human oversight help determine whether customers can confidently deploy a product in their operations. These requirements can shape both the cost of serving customers and the strength of the resulting business.

For founders, the framework helps identify the assets they must build, the suppliers they depend on, and the milestones that demonstrate progress. A chip company may need to prove performance and secure manufacturing access before it can scale revenue. An application company may reach customers sooner but still need to establish retention, distribution, and attractive delivery economics. Understanding these differences is essential to choosing an appropriate product strategy, financing plan, and measure of success.

2.1 Energy and Chips: Physical Scarcity

Energy and semiconductor companies face constraints that software alone cannot resolve. Expanding power capacity depends on generation, grid connections, permitting, and construction. Developing advanced chips requires specialized engineering, access to manufacturing, packaging and memory, and customer qualification. These dependencies can extend development timelines and create substantial financing needs before revenue reaches scale.

For founders, the challenge is to demonstrate a credible path from technical capability to commercial delivery. An energy venture may need secured site rights, a viable interconnection plan, permits, and customer capacity commitments. A chip company may need validated performance benchmarks, manufacturing and packaging access, and design wins that show customers intend to incorporate its technology. The relevant milestones depend on whether the business develops physical assets, designs components, or supplies supporting technology.

Control over scarce capacity or differentiated technology can create a competitive advantage, but high barriers to entry do not guarantee attractive returns. Construction delays, supplier dependence, cost overruns, and insufficient customer demand can undermine the economics. Founders and investors should evaluate how each financing milestone reduces these risks and whether the resulting capacity or product can be delivered profitably.

2.2 Infrastructure and Models: Turning Compute Into Intelligence

Infrastructure makes computing capacity usable through cloud services, data centers, networking, and software for deploying and operating AI systems. The economics vary substantially within this layer. Owning physical capacity requires significant capital, while a software company providing monitoring, evaluation, or deployment tools may operate with a different cost structure and financing need. Founders should identify which part of the infrastructure they improve and how that improvement translates into customer value.

Useful evidence includes lower cost per workload, faster response times, higher utilization, improved reliability, or less engineering effort required to deploy and maintain a system. Those gains must be evaluated against implementation costs and the alternatives customers already use. A technically effective tool still needs a compelling reason for customers to adopt it rather than rely on their cloud provider, existing platform, or internal team.

Model companies must demonstrate differentiated capabilities and a sustainable way to deliver them. Frontier development can require substantial compute and research investment, while specialized models may compete through domain performance, efficiency, deployment flexibility, or legally usable proprietary data. Performance should be assessed on relevant customer tasks alongside inference costs, reliability, and the effort required to integrate the model.

For founders in both layers, the central test is whether technical gains translate into recurring demand and viable delivery economics. Benchmarks and developer interest can establish early credibility; sustained production usage, paying customers, and renewals provide stronger evidence of a durable business.

2.3 Applications: Accessible Entry and Commercial Execution

Application founders can test products using existing models, APIs, cloud services, and open-source tools, often with less initial capital than businesses developing physical infrastructure or frontier models. The cost of reaching production can still be substantial, particularly when customers require complex integrations, sensitive data, or high levels of accuracy and oversight.

The starting point should be a specific customer problem with a clear economic buyer. Founders need to understand how the work is performed today, what makes it costly or ineffective, and whether their product delivers enough improvement to justify its price and adoption effort. A working demonstration establishes technical feasibility; repeat usage, payment, and renewal show whether the product earns a lasting place in the customer’s workflow.

Commercial progress also depends on the cost of delivering that value. Model usage, human review, customer support, and implementation can absorb revenue even when demand is strong. Founders should establish whether each customer relationship can become profitable and whether delivery becomes more efficient as the business grows.

Defensibility develops through advantages that competitors cannot readily reproduce. These may include legally usable proprietary data, effective distribution, deep integrations, domain expertise, and trusted customer relationships. The strongest case emerges when continued customer use improves the product or strengthens its position over time.

3. AI Financing Activity and Customer Demand

Financing and customer spending reveal different aspects of the AI market. Stanford reports that 1,953 U.S. AI companies received institutional funding for the first time in 2025. This indicates the breadth of companies entering institutional financing, but does not establish when they were founded or how they are distributed across the AI stack.

Customer spending provides a separate view of commercial demand. Within the categories covered by its market model, Menlo Ventures estimates enterprise generative-AI spending at approximately $37 billion in 2025. The model combines a survey of U.S. enterprise decision-makers with market analysis. Roughly $19 billion went to applications, with approximately $18 billion allocated to model APIs, model training, and AI infrastructure. These estimates cover both startups and incumbents.

For founders, these signals answer different questions. Financing activity helps characterize the investment environment; customer spending helps identify markets where buyers are allocating budgets. Neither establishes that a particular startup category is profitable, underserved, or easy to enter. Evaluating an opportunity requires examining the specific customer, competitive alternatives, and delivery economics within it.

3.1 Where Enterprises Are Spending on AI Applications

Menlo Ventures divides its modeled 2025 enterprise generative-AI application spending into three categories: $8.4 billion for horizontal AI, $7.3 billion for departmental AI, and $3.5 billion for vertical AI. These modeled estimates include both startups and incumbents and indicate customer spending rather than the number or success rate of companies competing in each category.

The distinction matters because these markets require different approaches. Horizontal products serve needs across business functions, departmental products support particular roles or teams, and vertical products address industry-specific workflows. Each can involve different buyers, sales cycles, integration requirements, and obligations around data and compliance.

Founders should use these categories to define the market they can realistically serve. A large spending category may be difficult to access without established distribution, while a narrower workflow may offer a clearer buyer, more frequent usage, or specialized requirements that the team is well positioned to address. The relevant opportunity depends on how much customers will pay, how efficiently the company can reach and serve them, and whether those relationships can expand over time.

3.2 Infrastructure Demand Grows With AI Adoption

AI deployment creates demand for computing capacity and the systems needed to operate it. Stanford’s 2026 AI Index reports that global AI compute capacity grew approximately 3.3 times annually from 2022, reaching about 17.1 million H100-equivalents by the end of 2025. This measure expresses computing capacity relative to Nvidia’s H100 GPU; it does not represent a count of installed H100 chips.

As products enter production, customers also need tools to manage data, coordinate model calls, evaluate outputs, monitor performance, protect access, and control costs. These requirements create opportunities for infrastructure startups that address a specific operational problem. Growth in total computing capacity, however, does not establish demand for every infrastructure product.

For founders, the commercial test is whether the product delivers enough improvement to justify adoption and ongoing spending. Evidence might include lower inference costs, fewer production failures, faster deployment, or reduced engineering workload. Integration into critical systems can support retention, but customers must continue receiving measurable value as cloud providers, existing platforms, and competing tools improve.

3.3 Competing in Capital-Intensive AI Markets

For founders targeting energy, advanced chips, or frontier models, a credible market opportunity must be supported by access to the resources required to deliver it. Specialized talent, manufacturing capacity, power, and compute can become conditions for participation. A business plan should distinguish resources already secured from those still dependent on financing, supplier agreements, or regulatory approvals.

The scope of entry matters. Building a frontier model or owning physical infrastructure creates different financing requirements from developing a specialized model, designing a component, or supplying technology that improves existing systems. Founders should identify the specific advantage they can establish and the capital required to demonstrate it before committing to broader expansion.

Progress should then be assessed against milestones that reduce the business’s most consequential risks. These may include independently validated performance, manufacturing readiness, secured capacity, or customer commitments. Such achievements can provide meaningful evidence before revenue scales, but they must connect to a credible path toward commercial delivery and sustainable economics.

4. Where Capital Is Concentrating

CB Insights reports that private AI companies globally raised approximately $225.8 billion in 2025. Rounds of $100 million or more accounted for 79% of that funding. OpenAI, Anthropic, and xAI raised a combined $86.3 billion, representing approximately 38% of the total. Headline investment figures therefore reflect substantial concentration in large financings and a small number of companies.

For early-stage founders, a record funding total provides limited insight into the capital available to businesses at their stage. Assessing fundraising conditions requires examining comparable companies, round sizes, investor activity, and the milestones investors expect within the relevant category.

The funding analysis in this section uses CB Insights consistently. Stanford’s separate investment figures, discussed in the geography section, come from a different dataset and should not be combined with or used to calculate shares of CB Insights’ totals.

4.1 Model Developers and the Economics of Large Rounds

CB Insights reports that large language model developers captured 41% of total private AI funding in 2025. This concentration demonstrates the financing scale of model development, although funding raised does not establish commercial success or attractive returns.

For frontier-model companies, financing may support training, research talent, computing commitments, and the cost of serving growing usage. Investors must assess whether those commitments can produce differentiated capabilities and sufficient customer demand to justify continued investment. The amount raised reveals the resources available to compete; the business’s performance determines whether those resources are being used effectively.

Application companies can often begin with less capital, but their financing needs may grow through customer acquisition, implementation, product development, and operating costs. Comparing these businesses with model developers requires examining what each round is expected to achieve. Relevant questions include how much technical or commercial risk the funding removes, how long it supports operations, and what evidence the company must produce before seeking additional capital.

4.2 When Scarcity Can Support an Investment Case

Control over a constrained resource can strengthen an investment case when customers depend on it and alternatives are limited. Power access, manufacturing capacity, advanced packaging, or specialized compute may provide an advantage if a company can secure that resource and deliver it economically.

The investment question is how long the constraint will remain commercially relevant. New capacity, technical improvements, changes in customer demand, or alternative suppliers can reduce the value of a previously scarce resource. High development costs alone do not establish pricing power or defensibility.

Founders should therefore explain what they control, how that control is protected, and why customers will continue to value it. Evidence may include enforceable access agreements, contracted demand, a demonstrated cost advantage, or technology that materially improves resource efficiency. This connects the scarcity argument to a business that can generate returns rather than relying on market shortages to sustain its value.

5. Where Commercial Demand Is Emerging

The spending estimates discussed in Section 3 show that enterprises are allocating budgets to AI applications and the technologies supporting them. The next question is where that spending reflects a lasting customer need and whether deployment produces enough value to sustain it.

Industry-specific spending provides one signal by identifying workflows that customers are willing to fund. Enterprise adoption research provides another by examining how widely organizations deploy AI and what business impact they report. Together, these sources help distinguish market interest from commercial progress, although they do not establish the retention, profitability, or performance of individual startups.

5.1 Vertical AI Spending Highlights Demand for Specialized Workflows

Menlo Ventures estimates that enterprise spending on vertical generative-AI applications reached approximately $3.5 billion in 2025, up from $1.2 billion in 2024, within its market model. This increase indicates growing expenditure on industry-specific products, although it does not establish that vertical AI captured a larger share of overall enterprise AI spending.

Healthcare accounted for approximately $1.5 billion, or 43% of Menlo’s estimated vertical AI spending, exceeding the next four vertical categories combined. The concentration points to substantial demand within healthcare, but does not establish that every healthcare workflow offers an attractive opportunity or that startups serving the sector achieve stronger returns.

For founders, industry-level spending is a starting point for identifying a specific market. The next step is to establish which customer controls the budget, what workflow they will fund, and why existing products do not adequately meet that need. A credible opportunity connects those answers to a product the company can implement, support, and deliver economically.

5.2 Adoption Does Not Automatically Translate Into Economic Value

McKinsey’s 2025 survey illustrates the gap between widespread adoption and enterprise-wide financial impact. Eighty-eight percent of respondents reported AI use in at least one business function, while nearly two-thirds said their organizations had not begun scaling AI across the enterprise. Thirty-nine percent reported an enterprise-level impact on earnings before interest and taxes (EBIT), with most of that group attributing less than 5% of organizational EBIT to AI.

These findings do not mean that individual deployments lack value. A product can deliver meaningful improvements within a department without materially changing the earnings of an entire organization. The relevant test for a startup is whether its product produces an outcome that the customer values enough to fund and sustain.

Founders should establish a baseline before deployment and measure the change in a relevant outcome, such as processing time, error rates, operating costs, or revenue. The assessment should account for implementation, human oversight, and ongoing delivery costs. This gives customers and investors a clearer basis for evaluating whether adoption creates an economic benefit.

6. Where Value Is Captured

An AI company creates value when its product improves an outcome for a customer. It captures value through the revenue it earns and the economics of delivering that product. Strong customer benefits and rapid revenue growth can coexist with weak margins, expensive implementation, or high customer acquisition costs.

The founder’s task is to understand how those economics develop as the business scales. Lower model costs may improve margins, but competition can also push prices down. Proprietary data, effective distribution, and deep workflow integration may strengthen a company’s position, provided they translate into customer retention, pricing power, or lower costs. The following sections examine how these advantages and dependencies affect the company’s ability to build a sustainable business.

6.1 Testing the Economics of Customer Value

A customer may receive substantial value from an AI product while the provider struggles to serve that customer profitably. Founders therefore need to measure revenue alongside the direct costs of delivering the service, including model usage, cloud infrastructure, required human review, and customer support.

Consider an illustrative application charging $1,000 per customer each month. If monthly delivery costs include $200 for models and cloud services, $150 for human review, and $100 for support, the company retains $550 in gross profit, a 55% gross margin. That amount must still contribute toward sales, product development, and other operating expenses.

Now suppose increased usage doubles both model and cloud costs and human review costs, while pricing and support costs remain unchanged. Monthly delivery costs rise to $800, leaving $200 in gross profit and a 20% gross margin. Customer engagement has increased, but the provider’s economics have weakened. These figures are hypothetical and illustrate a business-model risk rather than an industry benchmark.

Founders should test how margins change with usage, implementation demands, and customer mix. Pricing, usage limits, automation, and product design should support a model in which stronger adoption can produce a financially sustainable customer relationship.

6.2 Scarcity and Bargaining Power

Scarcity can support value capture when it gives a company bargaining power with customers or suppliers. That advantage depends on controlling something customers need and cannot readily replace, such as secured capacity, differentiated technology, or legally protected access to valuable data.

The economic benefit depends on the terms of that control. A company may serve a growing market while transferring much of its revenue to suppliers that control essential inputs. Conversely, ownership rights, durable access agreements, or credible alternative suppliers may help it retain more value. Founders should distinguish assets they control from resources they can access only on another company’s terms.

These positions can change as new capacity becomes available or competing technologies improve. A durable strategy should therefore explain how the business can preserve customer value and viable margins as today’s constraints ease.

6.3 Distribution and Integration as Application Advantages

Distribution creates an advantage when a company can consistently reach suitable customers at an acquisition cost its business model can support. Existing relationships, trusted partners, specialized communities, and effective sales channels can help, but access alone is insufficient. Founders need evidence that these channels produce paying customers who stay long enough to justify the cost of acquiring them.

Workflow integration can strengthen retention when a product becomes useful within recurring customer operations. Relevant evidence includes sustained usage, adoption across teams, renewals, and expansion into additional workflows. Integration should make the product more valuable to the customer; implementation complexity or difficulty switching does not, by itself, demonstrate a strong customer relationship.

Data, domain expertise, reliability, and trust can reinforce both advantages. Founders should be able to explain how these assets improve conversion, shorten deployment, reduce support demands, or strengthen retention. This connects claims of defensibility to observable commercial performance.

6.4 Supplier Dependency and Margin Exposure

External model providers, cloud platforms, and data services can help AI companies launch quickly. They also introduce dependencies that affect costs, reliability, and the ability to adapt. A supplier’s price increase, service interruption, model change, or revised access terms can create consequences beyond the immediate cost of its service.

Founders should identify which suppliers are essential, how much delivery cost each represents, and what replacing them would require. Switching may involve more than changing an API connection: evaluation, integration, security review, and customer commitments can make migration expensive or slow. An alternative supplier provides negotiating leverage only if it can realistically support the product’s requirements.

The financial assessment should test margins under higher supplier prices, heavier customer usage, and the cost of maintaining a fallback. Lower model prices may improve economics, but those savings may be offset by increased consumption or pressure to reduce customer pricing. Understanding these exposures helps founders make informed decisions about architecture, contracts, and pricing.

6.5 Model Access and Deployment Economics

Founders should distinguish model access from deployment responsibility. An open-weight model can be operated internally or accessed through a managed provider. Using an external API reduces the infrastructure the startup must operate, while self-hosting gives the company more direct control over its deployment and creates additional engineering responsibilities.

The comparison should account for total delivery cost at the performance and reliability the product requires. Managed services involve usage charges and supplier dependencies. Self-hosting involves computing capacity, utilization, engineering, monitoring, security, and maintenance. Neither approach is inherently cheaper: the result depends on workload volume, model requirements, available expertise, and customer expectations.

Founders should test both economics and operational fit using representative customer tasks. Relevant criteria include output quality, response time, cost per completed task, data handling, customization, and the ability to accommodate demand. The preferred architecture is the one that meets customer requirements with sustainable costs and manageable operating risk.

7. Where the Bottlenecks Are

A bottleneck is a constraint that limits a company’s ability to deliver, sell, or expand its product. For an infrastructure business, that constraint may be power availability or access to computing capacity. For an application company, it may be customer integration, usable data, procurement, or the ability to demonstrate reliable results.

Founders should identify which constraint currently limits progress and what evidence supports that assessment. Improving model performance will have limited commercial effect if customers cannot complete deployment; securing more computing capacity will not resolve weak demand. Understanding the binding constraint helps determine where capital and management attention can produce the greatest improvement.

The International Energy Agency’s 2025 Energy and AI report estimates that global data centers consumed approximately 415 TWh of electricity in 2024, about 1.5% of global electricity consumption. Its base case projects approximately 945 TWh by 2030. Both figures cover total data-center demand, including AI and non-AI workloads.

AI is an important contributor to the projected increase. The IEA estimates that accelerated servers, primarily associated with AI, account for almost half of the growth in data-center electricity consumption between 2024 and 2030. The projection nevertheless depends on assumptions about adoption, efficiency, and infrastructure development.

Deployment timelines create a practical constraint. The IEA notes that a data center can become operational within two to three years, while supporting energy infrastructure often requires longer. For founders developing or procuring computing capacity, available hardware or a planned facility does not establish that sufficient power will be available on schedule. Power access, grid connections, cooling, and delivery commitments should be assessed together when evaluating capacity and expansion plans.

7.2 Falling Unit Costs and Total Infrastructure Demand

Stanford’s 2025 AI Index reports that the inference cost of achieving approximately GPT-3.5-level performance on the MMLU benchmark fell from around $20 per million tokens in November 2022 to $0.07 in October 2024, roughly a 280-fold reduction. This is a historical comparison at a specified performance level, rather than a measure of current pricing or the full cost of operating an AI product.

Lower costs can make additional uses commercially feasible, increasing the number of customers, tasks, and model calls. Total infrastructure demand therefore depends on both efficiency improvements and growth in usage. A decline in cost per unit can coexist with rising aggregate consumption, although the balance will vary by workload.

Infrastructure founders should assess demand using expected workload volumes, capacity utilization, and customer commitments. Falling unit costs alone do not establish that demand will weaken, just as growth in AI adoption does not guarantee that every provider will achieve profitable utilization.

7.3 Bottlenecks Shift Across the AI Stack

The bottleneck map summarizes common constraints across the five layers. It provides a starting point for diagnosis rather than a ranking of which markets offer the best opportunities. The significance of each constraint depends on the company’s product, customers, location, and stage of development.

Constraints can also change as a business progresses. A model company may initially struggle with technical performance and later face deployment or delivery-cost challenges. An application company may resolve product accuracy but remain unable to complete customer integration or secure renewals. Removing one obstacle can expose another.

Founders should revisit their assessment as milestones are achieved. The practical questions are what currently limits progress, what resolving it will cost, and whether doing so creates enough customer or economic value to justify the investment.

8. The Pilot-to-Production Gap

A successful pilot establishes that a product can deliver results under defined test conditions. Production deployment requires it to work reliably within the customer’s operating environment, with appropriate integration, security, support, and accountability. Moving between these stages can introduce requirements and costs that were absent from the initial demonstration.

Founders should define the path to production before a pilot begins. That includes an accountable customer sponsor, agreed success criteria, a deployment decision process, and clarity about the budget and resources needed to proceed. A pilot becomes stronger commercial evidence when both parties understand what a successful result would lead to and who can authorize the next step.

8.1 Commercial Validation Is More Than Adoption

Customer relationships provide different kinds of evidence. An unpaid pilot can reveal whether a product addresses a real problem. A paid pilot demonstrates willingness to allocate a test budget. Production usage shows that the product can operate in a live environment. None of these signals, individually, establishes a repeatable business.

Renewals, sustained usage, and account expansion provide stronger evidence of continuing customer value, particularly when they occur across multiple independent customers. Founders should distinguish revenue from one-time development work or heavily customized projects from revenue generated by a product that can be sold and supported repeatedly.

Commercial reporting should make these distinctions visible. Rather than grouping every relationship under “customers,” founders should report pilots, production deployments, recurring contracts, renewals, and expansion separately. This helps investors assess the strength of demand and helps the team identify where relationships stall.

8.2 The Commercial Validation Ladder

The Commercial Validation Ladder organizes customer evidence from an initial prototype through production use, recurring relationships, expansion, and verified outcomes. It provides a practical framework for assessing what a company has demonstrated and what remains unproven.

The stages can overlap or occur in a different order. A recurring contract may precede deployment, and measurable business impact should be assessed as early as a pilot. Enterprise standardization is relevant to some business models, but a company serving smaller customers can build a strong business without reaching that stage.

Founders should use the framework to track how customer relationships progress, how long transitions take, and why opportunities stall. Investors should examine the evidence supporting each claimed milestone, including usage records, payment history, renewals, and measured outcomes. A stage label is useful only when the underlying evidence supports it.

8.3 Match Validation to the Business Model

The evidence required to evaluate progress depends on what the company must build and how customers buy it. An application business may demonstrate demand through recurring usage, payments, and renewals. A chip or energy venture may need to establish technical feasibility, delivery capacity, and customer commitments well before substantial revenue develops.

Technical readiness and commercial demand should be assessed separately. A strong model benchmark does not establish willingness to pay, a manufacturing milestone does not establish customer demand, and a permit does not establish contracted revenue. Each achievement reduces a particular risk; its significance depends on what remains unresolved.

Founders and investors should examine the substance of customer commitments, including whether they involve payment, what conditions must be met, and when delivery is expected. This helps distinguish preliminary interest from evidence that customers are prepared to support a commercially viable business.

9. Geography and Sovereignty

An AI company’s location can influence its access to capital, talent, infrastructure, suppliers, and customers. These advantages are distributed unevenly: a region with substantial startup financing may still depend on imported chips, foreign cloud providers, or externally developed models.

Technological sovereignty concerns the ability of countries and regions to develop and operate AI with control over critical capabilities and dependencies. For founders, the practical questions include where computing and data can be hosted, which suppliers can serve the business, and whether customers require local infrastructure or deployment. Geographic comparisons are most useful when they connect these conditions to the company’s operating needs and target markets.

9.1 The United States Leads in Private AI Financing Activity

Stanford’s 2026 AI Index reports approximately $285.9 billion in U.S. private AI investment in 2025. It also records 1,953 U.S. AI companies receiving institutional funding for the first time, more than ten times the next-highest country, the United Kingdom, at 172. These figures establish the scale of financing activity within Stanford’s dataset; they do not measure newly founded companies or the availability of funding to every startup.

The United States also hosts major cloud providers, model developers, research institutions, and enterprise buyers. Stanford reports 5,427 data centers in the country, although facility counts alone do not establish their computing capacity, suitability for AI workloads, or availability to a particular company.

For founders evaluating the U.S. market, these ecosystem strengths should be connected to specific opportunities: investors whose mandates fit the business, customers with relevant budgets, suitable infrastructure, and access to specialized talent. National investment totals provide context; the resources a company can realistically access determine their practical value.

9.2 China’s AI Ecosystem Extends Beyond Private Financing

Stanford’s 2026 AI Index reports approximately $12.4 billion in Chinese private AI investment in 2025 and 161 companies receiving institutional funding for the first time. These figures describe activity captured by the private-investment dataset, rather than China’s total financial commitment to AI.

Stanford notes that government guidance funds and other state-supported financing are not fully reflected in these measures. Private-investment comparisons therefore provide an incomplete picture of the resources supporting China’s AI development. Research, model capabilities, manufacturing, and infrastructure should also be considered when evaluating the ecosystem.

For founders considering commercial relationships in China, the relevant assessment concerns the resources and customers they can access through specific suppliers, partners, or programs. Domestic model development and infrastructure can create options, but their practical value depends on product fit, deployment requirements, and the terms of access. These factors deserve separate assessment from national funding totals.

9.3 Financing and Infrastructure Across Other AI Hubs

Stanford records first-time institutional funding for 172 AI companies in the United Kingdom in 2025, alongside 108 in India, 92 in Germany, 84 in France, and 79 in Canada. These figures demonstrate financing activity across several markets, although a single year’s counts do not establish how quickly each ecosystem is growing.

Publicly backed computing infrastructure provides another perspective. Stanford reports that state-backed AI supercomputing clusters in Europe and Central Asia increased from three in 2018 to 44 in 2025. This measures the number of clusters within that regional grouping; it does not establish their combined computing capacity or the resources available to startups.

For founders, regional opportunities may involve specialized customers, research partnerships, talent, or access to infrastructure. Evaluating a particular program or computing facility requires checking eligibility, cost, available capacity, and permitted uses. These practical conditions determine whether regional investment can support the company’s development or commercial expansion.

9.4 Geography Shapes Different Layers of the AI Stack

A geographic assessment should distinguish where a company is based, where its customers operate, and where its suppliers and infrastructure are located. These may be different markets, with different implications for customer access, delivery costs, and operational dependencies.

The priorities vary by business model. Energy and chip ventures must assess physical sites, manufacturing, supply chains, and delivery timelines. Infrastructure and model companies need suitable computing capacity and technical talent. Application companies must understand customer purchasing requirements, deployment expectations, and the data needed to serve their chosen workflows.

Founders should map the geographic dependencies that could materially affect their business and assess them against actual supplier arrangements and customer requirements. This makes location a concrete operating decision: which combination of markets and resources gives the company a credible path to customers, reliable delivery, and sustainable economics?

10. Scenarios for 2027 to 2030

Four possible developments could materially change AI business economics over the next several years: cheaper and more interchangeable models, persistent compute constraints, slower enterprise scaling, and greater fragmentation across markets. The scenarios below explore their implications without assigning probabilities or predicting a single outcome.

Several could occur simultaneously, with different effects across industries and regions. Founders should use them to test the assumptions behind pricing, delivery costs, customer adoption, and financing plans. Investors can use the same framework to examine which businesses remain viable across several conditions and which depend heavily on one favorable outcome.

10.1 If Capable Models Become More Interchangeable

If multiple providers deliver comparable results on customers’ required tasks at lower prices, model access could contribute less to product differentiation. Application companies might benefit from lower delivery costs and greater supplier choice, while facing competitors able to reproduce similar capabilities.

The financial outcome would depend on whether cost savings exceed pressure on customer pricing and acquisition costs. Companies with effective distribution, specialized data, or deeply integrated workflows could retain advantages, provided those assets continue to improve customer outcomes.

Founders should test how their business would perform if a competitor could access equivalent model capabilities. Useful indicators include customer price sensitivity, switching behavior, retention, and the cost of replacing the underlying model. The strongest position would combine the flexibility to adopt better or cheaper models with customer value that remains distinctive.

10.2 If Compute Remains Constrained

If demand for suitable computing capacity grows faster than supply, companies could face limited availability, longer deployment timelines, or higher costs for particular workloads. These constraints could persist even while efficiency improvements reduce the computing required for individual tasks.

Providers that expand usable capacity or improve its utilization could benefit, including businesses addressing power, cooling, deployment, and inference efficiency. Their returns would still depend on delivery costs, customer commitments, and whether demand remains sufficient as additional capacity enters the market.

Founders should test their plans against delayed capacity, higher computing costs, and restricted access to preferred hardware. Relevant indicators include supplier lead times, available capacity, utilization, and the cost of completing customer workloads. Efficiency becomes commercially valuable when it reduces a verified constraint without compromising the performance customers require.

10.3 If Enterprise Deployment Slows

If organizations take longer than expected to move AI into production, startups could face extended sales cycles, delayed revenue, and higher implementation costs. A growing pipeline of pilots could consume resources without producing enough recurring business to support the company’s operating plan.

Products addressing integration, evaluation, security, and cost control could help resolve deployment barriers. Their opportunity would depend on whether customers are willing to fund those improvements; slower adoption could also constrain spending across supporting technologies.

Founders should monitor pilot-to-production conversion, time to deployment, implementation effort, renewals, and expansion. Financial plans should account for delays between initial customer interest and recurring revenue. In this scenario, a focused product with demonstrable customer value and manageable delivery costs could be more resilient than a business relying on rapid, widespread adoption.

10.4 If Regulation Fragments AI Markets

If markets diverge further in their requirements for data handling, model deployment, and oversight, serving customers internationally could require multiple operating arrangements. Separate hosting, product adaptations, documentation, and support could increase the cost and complexity of expansion.

Regional infrastructure, evaluation tools, and services that help companies meet market-specific requirements could see additional demand. Their commercial opportunity would depend on the requirements customers face, the budgets available, and the cost of delivering an effective solution.

Founders should test international expansion plans against the additional cost of serving each market. The assessment should connect expected revenue with deployment requirements, product changes, local expertise, and ongoing operating costs. A market can offer substantial demand while requiring an approach that is uneconomical for a company at its current stage.

11. Actions for Founders and Investors

Founders can translate this analysis into five priorities: define the customer problem, establish appropriate validation milestones, build measurable competitive advantages, align financing with delivery needs, and test whether the business can sustain its economics as conditions change.

For investors, these priorities provide a framework for examining the evidence behind a company’s claims. The assessment should connect technical progress, customer commitments, operating costs, and financing requirements to a credible path toward a sustainable business.

11.1 Define the Customer and the Workflow

Founders should identify an initial customer segment, the person who controls the budget, and the workflow the product improves. Document how that work is performed today, the outcome customers need, and the changes required to adopt the product. Customer conversations and deployments should test whether the problem is important enough to justify payment and implementation effort.

Investors should examine evidence of customer urgency, purchasing authority, and repeatable demand. Useful questions include why customers would act now, what alternatives they considered, and whether the product’s benefits justify the cost of changing their existing process.

11.2 Set Milestones That Reduce Business Risk

Founders should translate their next development stage into specific milestones, each linked to a material risk. Define the evidence required, the cost and time needed to obtain it, and the decision that follows. A milestone might establish technical performance, customer willingness to pay, delivery readiness, or the ability to serve customers economically.

Investors should assess what each milestone proves and what remains unresolved. Technical progress, signed commitments, and revenue should be evaluated in context, including any conditions, customer concentration, or unusually high delivery effort. The objective is to establish whether the company is becoming more capable of delivering a repeatable, financially viable product or service.

11.3 Demonstrate How Competitive Advantage Strengthens

Founders should identify the assets that make their product valuable and difficult to replicate, then connect those assets to measurable results. Proprietary data should improve relevant performance; distribution should support efficient customer acquisition; workflow integration should contribute to retention or expansion. Claims of an advantage should be supported by evidence that it works and that the company can continue using it.

Investors should test how that advantage might develop or erode. Relevant questions include whether customer usage improves the product, whether competitors can obtain equivalent resources, and how the business would respond if an established platform introduced a similar feature. Defensibility is stronger when the company can explain both the source of its advantage and the conditions that could weaken it.

11.4 Match Capital Strategy to the Business Model

Founders should build their funding plan around the cost and time required to reach the next meaningful milestones. Separate development expenditure, recurring operating costs, customer implementation, and capacity commitments. Allow for delays in delivery, customer payments, or subsequent financing, and identify which expenditures can be adjusted if progress takes longer than expected.

Investors should assess whether the proposed capital is sufficient to achieve the stated milestones and what additional funding may be needed afterward. Capital efficiency should reflect the company’s actual business model, including development timelines, asset requirements, margins, and working-capital needs. A credible plan explains both what the funding will accomplish and how those achievements improve the company’s ability to sustain operations or attract further capital.

11.5 Test the Durability of the Business

Founders should revisit their operating assumptions as model capabilities, supplier costs, competition, and customer requirements change. Use the scenarios in Section 10 to identify where margins, demand, or delivery could weaken, then determine which product, pricing, or financing decisions would improve resilience.

Investors should assess whether the business can adapt without losing the customer value that supports it. Across the AI stack, a credible investment case connects demonstrated demand, reliable delivery, sustainable economics, and an advantage the company can maintain. For founders, those same elements provide a practical foundation for building a business that can endure beyond the conditions that first made entry possible.

References

1. Stanford Institute for Human-Centered Artificial Intelligence. (2026). The 2026 AI Index Report, Chapter 4: Economy. Private AI investment, first-time-funded companies, and geographic financing comparisons.

2. Stanford Institute for Human-Centered Artificial Intelligence. (2026). The 2026 AI Index Report, Chapter 1: Research and Development. AI computing capacity and data-center distribution.

3. Stanford Institute for Human-Centered Artificial Intelligence. (2026). The 2026 AI Index Report, Chapter 8: Policy and Governance. State-backed AI supercomputing clusters.

4. Stanford Institute for Human-Centered Artificial Intelligence. (2025). AI Index Report 2025, Chapter 1: Research and Development. Historical inference costs at a specified model-performance level.

5. CB Insights. (2026). State of AI 2025 Report. Global private AI funding, large funding rounds, and investment concentration.

6. Menlo Ventures. (2025). 2025: The State of Generative AI in the Enterprise. Modeled enterprise spending, application categories, vertical AI, and healthcare spending.

7. International Energy Agency. (2025). Energy and AI: Energy Demand from AI. Data-center electricity consumption, demand scenarios, and infrastructure development timelines.

8. McKinsey & Company. (2025). The State of AI in 2025: Agents, Innovation, and Transformation. Survey-reported AI adoption, enterprise scaling, and financial impact.

For more information, visit:

Peachscore Website
Peachscore LinkedIn