OpenAI targets enterprise growth with cheaper AI models

OpenAI cut pricing for its budget-tier Luna model by 80%, driving roughly 10x usage growth, while enterprise revenue rose 32% between June and July versus 20% growth in overall annualized revenue. CFO Sarah Friar said the company is targeting specialized AI applications in chip design, life sciences and financial services, including outcome-based pricing models to demonstrate enterprise ROI. OpenAI also reported 25 million Codex users and said its Jalapeno chip reached tape-out in nine months using its AI models, while it competes with lower-cost Chinese open-weight models and Anthropic.

Analysis

The relevant market signal is not demand growth but AI price deflation moving faster than enterprise budget expansion. If lower-cost models become functionally adequate for routine coding, document processing and customer-service workflows, model providers face a volume-versus-gross-margin trade-off while enterprise software vendors lose some pricing power on AI add-ons. This favors owners of proprietary workflow data and distribution—MSFT, NOW, CRM, RELX and VEEV—over horizontal application vendors whose features can be replicated through a lower-cost model layer.

For cloud infrastructure, token-volume growth is not automatically bullish: lower model pricing can increase inference demand, but customers may shift workloads toward self-hosted/open-weight alternatives or optimize usage aggressively. MSFT has the strongest distribution and enterprise channel optionality, while ORCL and GOOGL are more exposed to price competition for compute workloads; the key near-term read-through is whether AI consumption revenue rises faster than price-per-token declines. The chip-design productivity claim is directionally supportive of AI-assisted engineering, but tape-out is not evidence of commercial silicon economics, yield, or a durable reduction in design-cycle costs; it should not yet be used to underwrite a semiconductor supplier trade.

Consensus remains focused on AI demand as a cloud-capex accelerant. The underappreciated second-order risk over the next 6-18 months is that model commoditization shifts value from foundation-model access to proprietary data, implementation and liability-bearing vertical workflows, while forcing model vendors to fund more inference volume at lower realized revenue per unit. The thesis is falsified if public cloud disclosures show sustained AI consumption growth with stable or expanding AI-related gross margins, rather than token growth alone.

Key Decisions for Investors

• No action in GS: conference-host exposure is immaterial, and the information does not alter Goldman’s earnings path or valuation.
• Initiate a 1-3 month pair trade: long MSFT / short ORCL in equal beta-adjusted dollars. MSFT has superior enterprise distribution and application-layer monetization if lower-cost AI expands adoption; ORCL is more dependent on winning price-sensitive incremental cloud workloads. Target 8-12% relative return; exit if Oracle reports AI cloud backlog/conversion materially above consensus or Microsoft reports Azure AI pricing pressure without offsetting usage growth.
• Build a 6-18 month basket long RELX and VEEV versus an equal-weight short basket of lower-moat horizontal SaaS names with limited proprietary data exposure, using IGV as the hedge if single-name shorts are unavailable. Vertical incumbents can embed lower-cost models into regulated workflows and retain pricing through data, compliance and distribution; the basket should be cut if net retention and AI attach rates fail to improve over two reporting cycles.
• Set an earnings watch on MSFT, GOOGL and ORCL for AI revenue per workload, cloud gross margin and customer self-hosting commentary. Do not add semiconductor exposure on the reported chip-design milestone unless independently verified production partners, volume commitments and economics are disclosed.