Brief IA
›

AI in 2027: Companies Confront the Real Cost of Usage

Business & Startups·Tom Levy·

AI in 2027: Companies Confront the Real Cost of Usage

AI in 2027: Companies Confront the Real Cost of Usage
Key Takeaways
1Gartner predicts a 54.2% increase in global spending on AI platforms and models by 2027.
2According to Futurum Group, 42% of the 200 companies surveyed were using open-weight models in production as of September.
3Interact Analysis forecasts about 20% growth in cobot shipments by 2027.
Why it matters : The decline in token pricing could be overshadowed by the growth in usage, while companies also need to fund operations and oversee their agents.

Gartner anticipates $99 billion in global spending on AI platforms and models by 2027. However, unit prices are declining among several providers. For businesses, the trade-offs also involve operational resources, task distribution, and agent supervision.

Buyers Could Negotiate with an Open Alternative

Manish Jain, CEO of Strategic Horizon Research, does not foresee a radical shift towards open-weight models by 2027. Nevertheless, he assigns them a role in negotiations: a company has another option when its proprietary provider raises prices or revises terms. The extent of task transfers to these alternatives could impact the pricing flexibility of proprietary offers.

Michael Heinrich, CEO of 0G Labs, describes a distribution where open models absorb repetitive and large-scale operations, while complex reasoning remains with proprietary models. According to him, the latter can be charged 10 to 30 times more per token. He also observes architectures where a leading model establishes the plan, and then less expensive models execute numerous calls. This combination could gain traction by 2027, without eliminating infrastructure and operational expenses: Heinrich reminds us that hardware resources and engineering are paid for even when GPUs are not in use.

Jain explains the persistent preference for proprietary offerings due to their performance, ease of use, and operational support. Running open alternatives reliably, securely, and economically requires specialists, he believes. He expects their progress first in knowledge-based assistance and document processing, where data mastery, sovereignty, and budget visibility take precedence over minor performance gaps.

Commercial responses remain forecasts. Heinrich anticipates that American providers of closed models will offer smaller and cheaper models, open-weight variations, and more bundled services in enterprise contracts. Jain, on the other hand, envisions increased competition in reliability, security, tooling, and service commitments.

Gartner Forecasts a 54.2% Increase in Spending

Gartner's projections indicate a rise from $64 billion to $99 billion for AI platforms and models by 2027, representing a 54.2% increase. Nick Patience, an analyst at Futurum Group, also expects an overall budget increase despite cheaper tokens, driven by consumption. The expansion of use cases for models and agents could thus outweigh unit savings.

In 2026, models reached new cost levels. Some providers and clients had encouraged their employees to consume as many tokens as possible, a practice known as tokenmaxxing; this has since been largely abandoned. The use of cheaper open Chinese models has simultaneously fueled price competition among Anthropic, Google, and OpenAI. At OpenAI, GPT-6 costs half as much as GPT-5.6; Google has also halved the price of Gemini 3.7 compared to its predecessor.

For William McKeon-White, an analyst at Forrester, the budget does not stop at models: companies must also track their SaaS platforms. He anticipates easier management in 2027 with better tools, frameworks, and systems, as well as increased budget attention and control. This improvement does not constitute a forecast for reduced spending.

Patience expects increased attention to the cost of completed tasks rather than just the volume of tokens. His example is that of a cheap model needing to restart three or four times: it can ultimately end up being more expensive than an expensive offer that succeeds on its first attempt. Execution quality thus enters into the trade-off he describes.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Companies Will Need to Choose Who Distributes Tasks

Deepak Seth, an analyst at Gartner, anticipates a shift in competition towards platforms capable of assigning each task to the appropriate model. With more agents, companies should coordinate multiple models instead of relying on a single offer. A possible configuration would reserve Anthropic's Opus 5.5 for complex tasks and the less expensive Haiku for simple requests. The choice also concerns who operates this routing: the company itself or a provider.

Another expected trade-off in 2027 involves the distribution of computing between cloud and on-premises installations to contain costs. Nick Patience observes a growing interest in on-site deployments, even among hyperscalers. These installations are also being considered to place models closer to enterprise data.

Supervision constitutes another area of focus for 2027 alongside spending: agents are regularly operating outside the controlled frameworks intended for them, raising concerns about the dangers of these systems. Seth deems their management critical as they multiply in computing. Use cases are expected to evolve from time savings on low-risk applications to measurable outcomes or business transformations, potentially leading to user-created applications or the elimination of entire departments.

Patience predicts that labs will place more emphasis on the operational capabilities of agents than on their conversational ease. The use of tools, computer operation, and the reliability of a sequence of steps are among the priorities he anticipates.

Futurum Measures Progress of Open Models

The adoption of open models is progressing, but proprietary providers still retain a significantly higher share of measured consumption. Futurum Group published a survey in October conducted among 200 companies: 42% reported using open models in production in September, up from 31% in July. Among user companies, the proportion of tokens from these models is expected to rise from 34% to 41% over the year.

The distribution of spending remains different. Futurum's results attribute 5% of AI service purchases billed by usage to open models. Microsoft 365 Copilot accounts for 36%, and Claude from Anthropic for 29%.

The offering notably pits Cohere and Mistral against Chinese developers, including Qwen from Alibaba and DeepSeek. Mistral, based in Paris, made available on October 29 the weights of Mistral Large 4, also known as Le Chonk. This multimodal model has 1 trillion parameters, of which 52 billion are active, and aims at programming as well as cybersecurity. The company claims superior results compared to competing open offers in programming tests.

Interact Analysis Expects About 20% More Cobots

Samantha Mou, an analyst at Interact Analysis, forecasts a roughly 20% increase in cobot shipments by 2027, continuing the strong growth seen in 2026. She expects growth to outpace that of traditional industrial robots. These collaborative machines are designed to work safely alongside humans, particularly in manufacturing or assembly plants. The analyst cites drivers such as AI integration, which enhances performance and expands applications, as well as adoption by SMEs seeking greater efficiency.

The anticipated growth of robotics and physical AI in 2027 is expected to remain geographically uneven, with a concentration in the Asia-Pacific region and a predominance of factory and warehouse automation. Yueqin Shen, an analyst at Omdia, a division of Informa TechTarget, anticipates a transformation in warehousing and logistics based on a survey of industrial users. She identifies the strongest growth in projects involving autonomous guided vehicles (AGVs) and autonomous mobile robots (AMRs). Among respondents, more than half expect new installations within the next 18 months.

These robots handle the transport of goods and material handling within warehouses, distribution centers, and factories. They can also patrol around industrial facilities, including mining sites. For machine vision products aimed at logistics, Omdia projects the highest revenue growth rate at 14.5% between 2025 and 2030. The firm largely links this outlook to the proliferation of AGVs and AMRs.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.