CIOs Face Nuanced Enterprise AI Vendor Choices

Enterprise CIOs will need to assess AI vendors with nuance in 2027, balancing concerns about debt-financed AI infrastructure against positive reports from enterprise IT buyers. The article says enterprise AI projects have matured, executives understand the trade-offs between deterministic and probabilistic systems, and approaches such as building rather than buying, on-premises deployment and open models are popular. It argues that a troubled AI market and strong enterprise adoption can coexist.

CoreWeave has moved beyond its earlier role as a GPU provider built around Nvidia’s stack by assembling Forge, a platform focused on software and AI model management. At its Fully Connected conference, Caterpillar and Capital One were among enterprise customers presenting, although AI labs, startups and some hyperscalers still pay most of its bills. At Nvidia GTC Berlin, CoreWeave launched a Zero Trust AI Framework to control access to models, data and runtime environments, and Physical AI Launchables, tested configurations intended to help scale physical AI and autonomous vehicles. The company’s enterprise growth faces questions about whether revenue can offset risks in AI infrastructure. Interest rates on some CoreWeave debt tranches reached 15%; its latest 10-Q mentions debt 152 times and liabilities 122 times. The company regularly raises capital and is betting that infrastructure returns will help it work off debt. The article describes CoreWeave as a viable workload provider while noting Wall Street’s concern about its balance sheet and neoclouds.

Meta has announced plans for Meta Enterprise Platform after recruiting former MongoDB CEO CJ Desai, but the offering currently amounts to a few slides. The article says Meta’s language models and AI harness show potential for businesses and developers. Muse Agent, Meta Business Agent, Muse Code and Muse API are possible components of a future enterprise offering, while the author judges Meta may need a decade to become a full enterprise option. Buyers may still examine Meta and SpaceX to gain negotiating leverage with AWS, Google Cloud and Microsoft Azure. On Meta’s third-quarter earnings call, Mark Zuckerberg named the model API, Muse Code and business agents as near-term enterprise offerings, saying each was growing quickly and that people were using Muse to help run small businesses. He also said Meta plans a version of Muse for corporate environments. CFO Susan Li described models, infrastructure, global distribution and business relationships as assets for pursuing a recurring revenue stream. The article contrasts those ambitions with skepticism among executives and more optimistic expectations among traders.

Anthropic and OpenAI have enterprise customers and are available through AWS, Google Cloud and Microsoft Azure, but the article says token spending and security concerns around their AI agents have weakened perceptions. Open models, including Nvidia Nemotron, are drawing interest. Apollo Chief Economist Torsten Slok said his report found no sign of AI boosting productivity growth and that mainstream companies were treating token costs as a burden on margins, while VC-backed companies continued increasing token spending. IBM CEO Arvind Krishna argued that open models can offer performance at much lower cost, asking whether customers would choose something costing 10% as much and performing at 90% of the best. He also said larger companies are scrutinizing AI spending and its returns. SAP CEO Christian Klein said the company now checks agent use cases against model performance, outcomes and price, switches models more often, and sees a growing share of open-weight models. The article says the labs’ capital spending commitments and future financial results remain uncertain, and that enterprise adoption could continue even if the labs disappeared. It concludes that Anthropic and OpenAI benefited from runaway AI budgets, which the author says are already over.

Enterprise CIOs will need to assess AI vendors with nuance in 2027, balancing concerns about debt-financed AI infrastructure against positive reports from enterprise IT buyers. The article says enterprise AI projects have matured, executives understand the trade-offs between deterministic and probabilistic systems, and approaches such as building rather than buying, on-premises deployment and open models are popular. It argues that a troubled AI market and strong enterprise adoption can coexist.

CoreWeave has moved beyond its earlier role as a GPU provider built around Nvidia’s stack by assembling Forge, a platform focused on software and AI model management. At its Fully Connected conference, Caterpillar and Capital One were among enterprise customers presenting, although AI labs, startups and some hyperscalers still pay most of its bills. At Nvidia GTC Berlin, CoreWeave launched a Zero Trust AI Framework to control access to models, data and runtime environments, and Physical AI Launchables, tested configurations intended to help scale physical AI and autonomous vehicles. The company’s enterprise growth faces questions about whether revenue can offset risks in AI infrastructure. Interest rates on some CoreWeave debt tranches reached 15%; its latest 10-Q mentions debt 152 times and liabilities 122 times. The company regularly raises capital and is betting that infrastructure returns will help it work off debt. The article describes CoreWeave as a viable workload provider while noting Wall Street’s concern about its balance sheet and neoclouds.

Meta has announced plans for Meta Enterprise Platform after recruiting former MongoDB CEO CJ Desai, but the offering currently amounts to a few slides. The article says Meta’s language models and AI harness show potential for businesses and developers. Muse Agent, Meta Business Agent, Muse Code and Muse API are possible components of a future enterprise offering, while the author judges Meta may need a decade to become a full enterprise option. Buyers may still examine Meta and SpaceX to gain negotiating leverage with AWS, Google Cloud and Microsoft Azure. On Meta’s third-quarter earnings call, Mark Zuckerberg named the model API, Muse Code and business agents as near-term enterprise offerings, saying each was growing quickly and that people were using Muse to help run small businesses. He also said Meta plans a version of Muse for corporate environments. CFO Susan Li described models, infrastructure, global distribution and business relationships as assets for pursuing a recurring revenue stream. The article contrasts those ambitions with skepticism among executives and more optimistic expectations among traders.

Anthropic and OpenAI have enterprise customers and are available through AWS, Google Cloud and Microsoft Azure, but the article says token spending and security concerns around their AI agents have weakened perceptions. Open models, including Nvidia Nemotron, are drawing interest. Apollo Chief Economist Torsten Slok said his report found no sign of AI boosting productivity growth and that mainstream companies were treating token costs as a burden on margins, while VC-backed companies continued increasing token spending. IBM CEO Arvind Krishna argued that open models can offer performance at much lower cost, asking whether customers would choose something costing 10% as much and performing at 90% of the best. He also said larger companies are scrutinizing AI spending and its returns. SAP CEO Christian Klein said the company now checks agent use cases against model performance, outcomes and price, switches models more often, and sees a growing share of open-weight models. The article says the labs’ capital spending commitments and future financial results remain uncertain, and that enterprise adoption could continue even if the labs disappeared. It concludes that Anthropic and OpenAI benefited from runaway AI budgets, which the author says are already over.