If AI companies continue losing billions of dollars, what factors will determine whether they raise prices for consumers

The short version

Whether AI companies can raise consumer prices while losing money comes down to one idea: pricing power, which is the ability to raise prices without losing a meaningful amount of demand [1]. Companies with strong pricing power see only a small drop in customers after a price increase, while companies with weak pricing power see a big drop [2][3]. If a company has real pricing power, it can raise prices without a big wave of customers leaving [4].

What gives a company the ability to raise prices

Pricing power is strongest when a product is differentiated, technically complex, essential to customers, and has a unique value proposition with few close competitors or substitutes [5][7][8][14][15]. Patents, copyrights, and other legal protections also reduce competition and add pricing power [13]. Brand trust and reputation matter too: customers who see a brand as reliable and higher quality are more willing to accept higher prices [9]. High switching costs help as well, because customers stay if leaving is costly or inconvenient [10]. Efficient production and economies of scale give companies more room to adjust prices [11], and demand that is not very price-sensitive makes higher prices possible [12]. Strong pricing power also shows up as higher profit margins and as the ability to protect margins when costs rise or the economy weakens [16][17][18]. Market leaders can pass higher costs from inflation on to customers by raising prices [19].

The evidence also warns that pricing power is not permanent. It shifts with brand strength, competitive alternatives, and customer sentiment [46]. Customer pushback can hurt even strong brands when price hikes go too far [49]. Retailers or distributors can resist price increases and pull products [50], and commoditization pressures can weaken an entire industry's ability to raise prices [51].

What the evidence says about AI specifically

Several sources argue that most AI layers have almost no pricing power, and that this is a structural feature of the industry rather than a temporary condition [23]. One analysis says pricing power in the AI stack is inversely distributed: the companies carrying the biggest capital risk have the least ability to set prices [24]. The layer that made enormous compute commitments is described as sitting in the weakest pricing-power position in the whole stack [25]. The formula given is: an undifferentiated product plus zero switching cost plus a free substitute about four months behind equals prices collapsing toward the cost of the cheapest competent producer [26]. The price of a fixed AI capability is said to deflate by about 10 times per year, meaning any closed-lab price can be undercut by a free-to-self-host model within one to two quarters [27][28].

For consumer subscriptions specifically, the evidence says the price has been anchored at about $20 per month for three years [29]. Any provider that raises that sticker price hands market share to a free competitor, so the price cannot move up easily [30]. Instead, providers are introducing ads, pushing customers to cheaper tiers, and managing the absence of pricing power rather than exercising it [31][32][35]. OpenAI's own internal forecast reportedly projects ChatGPT Plus subscribers falling by about 80 percent as users shift to a cheaper tier, with blended consumer average revenue per user falling from about $23 to under $12 [33][34].

There is one place the evidence says genuine AI pricing power shows up: enterprise and agentic work, where the ability to charge more depends on a capability lead of about four months rather than brand strength [40][41]. The same analysis concludes that there is no AI layer where "we'll just charge more later" is available [38].

Cost pressure that pushes prices upward

The main reason AI companies feel forced to raise prices is that serving customers is expensive. The average cost of computing is expected to climb by about 89 percent from 2023 to 2025, and 70 percent of executives cite generative AI as a critical driver [79][80]. Newer reasoning models can use 10 to 100 times more compute per query than a standard response [66]. Unlike traditional software, every AI query costs real money, so AI lacks the low-cost scalability of SaaS [126][149]. AI companies also see lower gross margins, roughly 50 to 60 percent versus 80 to 90 percent for SaaS [150]. Flat-rate consumer subscriptions only work at low usage: estimates suggest OpenAI loses money on Plus once a user exceeds about 11.4 percent utilization, and loses money on top tiers above about 5.7 percent utilization [36]. A single autonomous agent can burn more than $1,000 in a day, which inverts the economics of a flat-rate plan [37]. Because of this, vendors keep passing costs along to customers even as efficiency improves [54]. Some are already charging more by bundling features into higher-tier plans [56], charging premium prices for AI features in traditional software [60], and baking compliance costs into pricing [59].

Investor and funding pressure

Many sources say current consumer AI prices are subsidized rather than profitable. About $1.4 trillion in compute obligations are described as underwritten by revenue from layers that have no pricing power, which one analysis calls a subsidy from capital markets routed through companies that cannot mark up what they sell [39]. Another source says AI subscriptions are heavily subsidized by venture capital, and that the familiar $20 price is a market-building exercise funded by billions of dollars rather than a reflection of real costs [62][63]. OpenAI is reported to have had roughly $5 billion in operating losses in 2024 against about $3.7 billion in revenue, a math that only works while investors keep writing large checks [65].

The triggers that would force prices upward include IPOs, slowing venture capital rounds, investor pressure to show profits, higher interest rates, limited partners asking harder questions, and OpenAI's shift from a nonprofit to a for-profit structure, which brings shareholder return obligations [64][68][69]. For Big Tech, rising AI capital spending is eating into free cash flow: Microsoft, Alphabet, Amazon, Meta, and Oracle are expected to spend more on capital expenditures than they generate in free cash flow by 2027 [129]. One projection shows operating cash flow growing by about $340 billion by 2027 while capex rises by roughly $534 billion [130]. Investors are looking for AI to drive incremental revenue, expand margins, and improve cash flow over the next two to three years, and if those benefits do not appear, they will question whether the investment cycle has gone too far [131][136][137]. Oracle's free cash flow has turned negative and its shares have fallen sharply, a concrete example of investor alarm [134]. Buybacks could be at risk if spending stays high and monetization takes longer than expected [135].

There is also a counterweight: leading AI firms still have access to fresh funding and are generally more profitable and carry less debt than telecom companies during the dot-com era, which gives them some buffer before prices must rise [84][146]. The scale of investment does pressure AI companies to deliver major profits within a limited timeframe because current financing cannot continue forever [123]. But critics say there is little evidence that businesses or everyday users will get enough value to justify paying large amounts [124], and most organizations report little to no profit-and-loss impact from AI, with gains concentrated among major tech firms [142].

Consumer reaction and price elasticity

Demand elasticity, or how sensitive customers are to price changes, is a key factor in whether a price increase works. Elasticity measures the percentage change in quantity demanded for a one percent change in price [100][168]. When demand is elastic, customers are highly price-sensitive and a price increase causes a larger percentage drop in quantity sold [101][176]. When demand is inelastic, customers are less sensitive and a price increase causes only a small drop in quantity sold [102][176]. The practical rule from the evidence: inelastic or low-elasticity products have pricing power and can support price increases, while elastic products need to compete on price and volume [91][103][107][117]. Software subscriptions generally run about -0.9 to -1.4 in elasticity, with switching costs creating some stickiness [111].

Elasticity is also dynamic. It shifts 20 to 40 percent per year due to competitive pressure, economic conditions, and changing consumer behavior [104]. Promotions, bundling, and competitor price moves all change demand sensitivity [90][195], and elasticity depends heavily on what competitors do with their prices [89][196]. More substitutes make demand more elastic because customers can switch away [88][192], and easier price comparison can make demand more elastic [95]. Competitive alternatives in AI can emerge quickly, so elasticity calculations are constantly changing [178][182].

Segment differences matter too. Enterprise segments are often less elastic and can be premium-priced, mid-market segments are served with value-based pricing, and SMB and individual segments are often more elastic and get only limited features [183][184][185]. Value perception varies across customer segments [177], and the original price level matters: lower original prices can be more inelastic [193]. As AI technology matures, price sensitivity often decreases, with one study finding elasticity for AI services falls by 15 to 20 percent annually as the technology becomes mainstream [180]. Premium, differentiated AI features typically show less elasticity than commodity features [181], and low-elasticity AI services should be priced on value delivered rather than production cost [186].

Companies also have to watch the consequences. A price increase can boost revenue but hurt profits if it triggers expensive marketing to offset volume declines [114]. So providers would monitor real-time customer feedback, customer acquisition costs, purchase frequency, average order value, churn rates, and revenue versus baseline projections to see whether a price increase can be sustained [118][171][172][173]. Management can also set safeguards and price limits so AI-driven pricing never exceeds internal bounds [119], and the ultimate measure is whether margins actually improve [122].

What kinds of price increases are most likely

Several sources point to non-obvious price increases rather than a simple sticker hike on the familiar $20 plan. The most likely near-term scenario is flagship consumer plans moving to $30 to $50 per month [71], with meaningful usage caps on lower tiers and pay-as-you-go overages for heavy users [72]. The $20 subscription is already narrowing in scope as features are gated behind higher tiers or metered usage [73][78]. As base model prices rise, AI features embedded in other products will rise too [77]. Enterprise plans already cost significantly more, often $30 or more per user per month as a starting point, which shows how underpriced consumer plans are by comparison [74][75]. One source says increases are already happening through model tiering and usage caps, and that OpenAI's for-profit shift, investor pressure, and rising reasoning-model compute costs point to meaningful price increases within one to three years [76].

There is also a historical argument: every tech category that launched with subsidized pricing, such as cloud services, streaming, ride-sharing, and food delivery, eventually rationalized toward sustainable pricing [67][70]. The user-acquisition phase is described as largely complete, and the monetization phase that follows usually involves prices going up [67]. At the same time, an AI price war is piling pressure on OpenAI and Anthropic as customers mix and match models to avoid premium prices [52][53]. Competition may push prices down over time, or it may create more complexity as buyers manage multiple pricing structures across tools [57].

The evidence also describes better pricing methods as an alternative to blunt price hikes. Value-based pricing and proper elasticity analysis are tied to higher margins: one source cites 10 to 15 percent higher margins versus cost-plus or competitor-based pricing, another cites 25 percent higher profit margins for value-based pricing [188][189]. Hybrid pricing models, which combine a base subscription with usage or outcome tiers, are preferred when outcomes are uncertain [158]. Outcome-based pricing only works when the provider can absorb cost variance and the outcome is clear and measurable [160]. Some AI-first companies are creating premium-pricing opportunities by redefining service-level agreements to outperform traditional vendors [159]. The evidence also warns against pricing pitfalls such as soft ROI positioning, which kills willingness to pay, and cost-plus pricing, which leaves money on the table [157][163]. Sustainable AI pricing is signaled by customers who renew without hesitation, expand usage naturally, and refer others based on clear ROI [161].

The honest bottom line

The evidence does not give a single formula for whether AI companies will raise consumer prices, because the forces pull in opposite directions. Cost pressure, investor demands, for-profit restructuring, and the end of subsidies push prices upward [64][69][76]. Competitive pressure, free open-weight substitutes, fast capability deflation, the $20 anchor, and consumer price sensitivity push prices downward or force providers to find other revenue such as ads and usage caps [27][28][30][31][57]. In the end, the deciding factors are whether a specific AI service is differentiated enough and essential enough to give it pricing power [5][15], whether customers see it as a necessity with few substitutes [8][12][47], whether switching to a competitor is costly [10], and whether demand in that customer segment is inelastic enough to survive a higher price without losing volume [103][117][183]. If those conditions hold, price increases can work; if the product is easily replaceable, raising prices will mainly hand customers to a rival [8][30].