2 minute read

The $39B AI Market

Source: Menlo Ventures

AI spending is growing fast, but the value equation remains uncertain

Enterprise Gen AI spending hit $37B in 2025, capturing 6% of the SaaS market, according to Menlo Ventures. While $19B went to apps and $18B to infrastructure, architectures remain simple. Prompt engineering and RAG dominate implementation, with only 16% of enterprise deployments qualifying as true autonomous agents. link

While Gen AI spending has grown 3.2x this year, it represents less than 1% of the $4 trillion in 2024 U.S. corporate profits reported by the St. Louis Fed. Even stratospheric returns on AI spend won’t materially impact these aggregate profits for the foreseeable future. link

Microsoft Copilot leverages corporate data in Teams and SharePoint, but users face a steep learning curve, according to The Information. Concerned that customers aren’t seeing enough value to renew subscriptions, Microsoft is now funding employee training, while OpenAI similarly offers resources to bridge the skills gap. link

Companies need “domain owners” who have built a “second muscle” in technology, according to McKinsey. With only 5% of senior leaders having held technical roles, these executives are vital for aligning data, platforms, and engineering talent to deliver successful AI transformations. link

Expected Economic Impact

Source: Anthropic

Who’s capturing that spend? Anthropic and open-weight models surge ahead

Menlo Ventures also reports Anthropic has unseated OpenAI as the top enterprise model provider, capturing 40% of spend versus OpenAI’s 27%. Google also grew to 21%. Anthropic’s surge is powered by a dominant 54% share of the coding market and the success of Claude Code. link

Anthropic also reports its engineers primarily use Claude for debugging and code understanding. Staff use the tool in 60% of their work, estimating a 50% productivity boost. Additionally, 27% of Claude-assisted work involves “nice-to-have” tasks that otherwise wouldn’t be completed. link

A new analysis of the Hugging Face ecosystem details a “fundamental rebalancing” of AI power. As US corporate dominance wanes, Chinese developers have surged ahead, with DeepSeek and Qwen driving China’s download share to 17.1%, surpassing the collective output of American model builders for the first time. link

Exemplifying the above trend, Pinterest and Airbnb have fine-tuned open-weight models from Qwen and DeepSeek to match frontier performance. According to Applied AI, these models handle multi-modal jobs and customer support queries at less than 10% of the cost. link link

Despite the hype, most implementations remain focused on the basics

OpenAI reports that technology, healthcare, and manufacturing are the fastest-growing AI adopters, with the tech sector surging 11x year-over-year. While professional services and finance continue to lead in absolute scale, a widening divide shows “frontier” firms now send twice as many messages per seat as the median enterprise. link

AI Implementation by Sector

Source: Anthropic

While most usage remains in “predictable areas,” Anthropic highlights N26 and Thomson Reuters leveraging Amazon Bedrock for compliance and security. Anthropic calculated a 1% increase in input context length correlates with a 0.38% increase in output quality, demonstrating that robust, selective data aggregation, not just model power, is critical for success. link

Come together, right now

AWS and Google announced a jointly developed service providing high-speed networking links between their competing clouds. This allows enterprises to host data in their preferred cloud while utilizing models from other providers, reducing connection setup times from weeks to minutes. link

The Linux Foundation is organizing the Agentic Artificial Intelligence Foundation to develop shared open-source standards, according to The Information. Initial technical contributions include OpenAI’s Agents.md, Block’s Goose, and Anthropic’s MCP. OpenAI, Google, Microsoft, and Anthropic have joined the group to ensure agent ecosystems can effectively connect. link