Software incumbents are beginning to capitalize on enterprise frustration with the cost and complexity of deploying raw artificial intelligence models. At a recent customer event outside San Francisco, Palantir CEO Alex Karp sharply criticized the practice of businesses engaging directly with foundation model developers, specifically targeting the high costs and opaque returns associated with the current AI ecosystem. Karp argued that companies attempting to integrate directly with large language model providers will ultimately find the experience financially draining and strategically ineffective.
Karp told attendees that customers engaging directly with model builders will learn that the providers do not care about them, according to reports. He warned that businesses would leave these engagements feeling "poorer and less safe," paying heavily for compute tokens without a clear understanding of how the technology improves their operations. The remarks underscore a growing effort by established software intermediaries to position themselves as the necessary translation layer between raw AI capabilities and actual enterprise utility. Spokespeople for Anthropic and OpenAI did not provide a response to the comments.
The integration layer versus the model builders
The friction highlighted by Karp points to a structural tension in the enterprise AI market. Palantir, a data analytics company known for its defense and enterprise software contracts, is attempting to assert the value of its integration platforms over the raw models provided by companies like Anthropic and OpenAI. For the past two years, foundation model developers have enjoyed direct, enthusiastic engagement from corporate boards eager to experiment with generative AI. However, as pilot programs transition into production, the reality of token-based pricing and the technical difficulty of grounding models in proprietary corporate data have become apparent.
By framing direct engagement with model builders as a foolish endeavor, Palantir is articulating a sentiment that resonates with software incumbents: raw intelligence is a commodity that requires an intermediary to become a product. Foundation model companies, which have historically focused on consumer applications and API access, are now facing pushback from traditional enterprise vendors who smell an opportunity to reclaim the customer relationship. This dynamic suggests that the next phase of AI adoption will be defined not just by model performance, but by which layer of the software stack captures the enterprise budget.
Infrastructure costs and narrative warfare
As software intermediaries challenge the enterprise value proposition of foundation models, the model builders themselves are navigating immense capital requirements and intensifying competitive scrutiny. The sheer scale of compute required to train and run these systems is forcing creative financial engineering. Broadcom, a major semiconductor and infrastructure software company, is reportedly helping to finance chip deals for Anthropic and OpenAI in partnership with private equity firms Apollo and Blackstone. This reliance on alternative financing structures highlights the staggering capital intensity of the AI race, where even the most heavily funded startups must seek outside leverage to secure necessary hardware.
Simultaneously, the narrative battle among the leading AI developers is growing increasingly pointed. Microsoft AI head Mustafa Suleyman recently criticized Anthropic publicly, calling out the startup for acting as though its Claude model possesses consciousness. Microsoft, the primary backer of OpenAI, is actively competing with Anthropic for enterprise mindshare and developer loyalty. Suleyman’s critique, combined with Palantir’s warnings about token costs, illustrates a market where foundation model developers are being squeezed from multiple directions—facing skepticism from enterprise software veterans on one side and aggressive rhetorical attacks from tech giants on the other.
The converging pressures of enterprise integration challenges, massive infrastructure financing, and competitive posturing indicate a maturing and increasingly combative AI sector. As the initial novelty of large language models fades, the focus is shifting toward sustainable business models and demonstrable return on investment. Whether foundation model developers can successfully build their own enterprise workflows, or whether they will be relegated to the background by established software intermediaries, remains the central question for the industry's commercial future.
With reporting from The Information, The Verge.
Source · The Information
