Cheap AI Undercuts OpenAI and Anthropic IPO Prospects

CNBC reports that cheaper, near-frontier AI from Chinese labs and lower-cost Western challengers is shifting enterprise demand away from the premium offerings that underpin reported IPO valuations for OpenAI and Anthropic. Per CNBC, Artificial Analysis benchmarked inference cost-per-evaluation and found $4,811 for Claude, $3,357 for ChatGPT, $1,071 for DeepSeek and $948 for Kimi. CNBC also cites a CloudZero survey showing 45% of companies spent more than $100,000 per month on AI in 2025, up from 20% in 2024. CNBC reports that both companies have been discussed with IPO valuations "north of $800 billion," and that OpenAI could file a confidential S-1 imminently, "as soon as this week," per CNBC.

Public reporting highlights two technical drivers behind the cost gap: model efficiency improvements in smaller architectures and cheaper inference pricing from Chinese providers. Industry benchmarking firms like Artificial Analysis measure cost as a combination of latency, throughput, and hardware/instance pricing; CNBC summarizes their cross-model comparison that produced the per-query figures above. For practitioners, the implication is that lower-cost models can materially reduce inference spend for large-volume enterprise workloads, even if capability parity is partial.

Reporting places this dynamic inside a broader IPO and capital-markets cycle. The New York Times reports that the strong market debut of Cerebras helped revive appetite for large AI-related public offerings, and that investors are watching potential mega-IPOs from firms including OpenAI and Anthropic. CNBC frames the cheaper-competition story as a countervailing force to that IPO momentum: if enterprise customers can get acceptable accuracy and latency at a fraction of the cost, premium pricing assumptions embedded in high valuations become harder to justify to investors.

Companies choosing models for production should treat per-inference cost as a first-order operational variable, not an afterthought. Editorial analysis: organizations running high-throughput pipelines typically balance accuracy, latency, and cost by evaluating models on task-specific benchmarks and total cost of ownership; the figures reported by Artificial Analysis underscore why that evaluation is increasingly driven by economics as much as accuracy. Open-source and smaller commercial models, plus regional providers, are closing capability gaps while offering much lower inference bills.

• •Whether OpenAI or Anthropic publish S-1s with audited financials and detailed margin metrics, which will help investors and practitioners compare gross-margin assumptions. (CNBC reports OpenAI may file confidentially soon.)
• •Independent benchmarking updates from firms like Artificial Analysis or CloudZero that track cost-per-query across new model releases.
• •Enterprise procurement patterns, specifically share of traffic routed to lower-cost providers, which CNBC reports is already rising.

All quoted numbers and company-specific cost comparisons above are reported by CNBC, which cites Artificial Analysis and CloudZero. The New York Times coverage of the Cerebras IPO is cited for market context. Editorial labels indicate analysis and practitioner implications rather than statements of internal intent from the companies involved.