Rising AI Costs Force Businesses to Seek Cheaper Alternatives

Businesses reliant on powerful, often expensive artificial intelligence models are now seeking cheaper alternatives. Industry leaders contend that more affordable options are necessary for broader adoption of AI tools across companies.

Credit: UNSPLASH

Top executives, including Microsoft's Satya Nadella, Nikesh Arora from Palo Alto Networks, and Brian Armstrong of Coinbase Global, have stated that smaller, less costly models can adequately address a significant proportion of corporate requirements. This stance emerges from a re-evaluation within organisations.

Companies previously encouraged extensive AI tool usage, often viewing increased consumption as a measure of productivity, a practice dubbed "tokenmaxxing". However, the associated expenditures are now becoming a financial burden. Prices for tokens, the units measuring AI usage, are falling. But the overall cost of completing tasks is rising as AI firms transition from flat subscriptions to a usage based pricing structure. This results in unpredictable and frequently higher bills for companies, as usage per task proves difficult to estimate.

Uber, for instance, depleted its entire 2026 AI budget within four months after its employees rapidly adopted AI coding tools. This prompted management to implement usage caps. Harold Byun, Chief Executive Officer of BlueRock, a startup assisting companies with secure AI system operation, confirmed that changing the licensing model surprised many. He reported that some BlueRock customers experienced a 20 per cent to 30 per cent increase in over budgeting immediately following this shift.

As companies intensify their use of AI, their expenditures are exceeding initial projections. Tasks now demand more steps, more data, and lengthier inputs, driving up costs. Gartner estimates that AI coding expenses will surpass the average salary for a developer by 2028. A survey conducted by the research firm revealed that three-quarters of executives anticipate tech budgets rising, with nearly half forecasting double digit jumps.

Businesses are adopting less expensive models and utilising routing tools such as OpenRouter, an AI marketplace. They aim to assign tasks to the most cost effective system, reserving premium models for complex work like coding. Open source tokens processed on OpenRouter increased from 34 per cent earlier in the year to 65 per cent recently, a Citi note indicated.

This development could benefit open source model developers, including China's DeepSeek. DeepSeek has gained widespread adoption among startups but has encountered difficulties penetrating large businesses due to security issues. And Palo Alto Network's Arora posted on X recently, suggesting AI labs should price tokens today at the lower rates expected in the future, if they wish to secure enterprise clients.

OpenAI, the creator of ChatGPT, appears to be adapting to this trend. Reports suggest the organisation is considering substantial price reductions, including on token usage. This move anticipates similar actions from competitor Anthropic. However, a shift towards cheaper models could negatively affect revenue growth for these firms, particularly as they prepare for potential initial public offerings. Christopher Brown, a financial adviser in private wealth management at Synovus Securities, which holds shares in several Big Tech companies, predicts a price war between OpenAI and Anthropic as they compete for earlier public market listing dates.

Tech stocks experienced a sell-off recently. Investors re assessed AI valuations amid doubts regarding returns on significant spending. These concerns were exacerbated by a weak post IPO performance from SpaceX and reports suggesting a delay in OpenAI's listing.

The increase in costs is steering more businesses towards open source models, including less expensive Chinese alternatives. The four most popular models on OpenRouter are all Chinese, with DeepSeek holding the top position. Chinese models are narrowing the capability gap with leading United States models. They charge as little as USD 0.18 per million tokens, compared to an average of USD 4 for the top models, according to the Citi note.

BlueRock's Byun stated that open source models, which were once more than a year behind leading AI models, are now estimated to be only about four months behind, and that gap is expected to continue closing. Still, some analysts caution that security concerns surrounding Chinese models will likely hinder enterprise adoption, particularly within sensitive sectors such as cybersecurity. Instead, businesses are expected to adopt strategies seen in cloud computing. This involves diversifying across multiple providers to identify the most suitable and cost effective solutions.

Val Bercovici, chief AI officer at WEKA, an organisation that helps companies run AI models more rapidly and efficiently, noted that open source models are proving to be "90 per cent as good at 10 per cent of the price." He added that premium tokens are not necessary for every level of effort.

• Businesses are moving away from expensive AI models due to rising usage costs.

• Executives advocate for smaller, cheaper models to meet corporate needs.

• Usage based pricing leads to unpredictable and higher bills for companies.

• Open source and Chinese AI models offer significant cost savings, despite some security concerns.

• A potential price war among leading AI providers could affect future market listings.

Source: Reuters