OpenAI frontier models and Codex are now available on AWS

Original Article ↗ Hacker News Discussion ↗

Pricing and Economics

  • OpenAI models on AWS Bedrock are ~10% more expensive than direct OpenAI pricing; GovCloud users report ~30% extra on top.
  • Explanations offered: US-only inference / data residency premiums, room for enterprise discounts, and the fact many enterprises will accept higher AWS costs to avoid adding a new vendor.
  • Some see this as “AWS bill padding” but still rational for companies that already live in AWS and care more about procurement simplicity than marginal cost.

Enterprise Procurement and Governance

  • Many large companies find it extremely hard to approve new vendors; they must pass infosec, legal, customer contract, and audit hurdles.
  • Using OpenAI via Bedrock avoids adding another vendor, another security review, and another data processor; it also simplifies billing and internal chargeback.
  • “Nobody got fired for buying AWS” sentiment: sticking with an existing cloud provider is seen as career-safe, even if not cheapest or ideal.
  • Bedrock is often the only realistic way to deploy LLMs in tightly controlled corporate environments.

Data Privacy, Residency, and Legal Risk

  • Bedrock runs models on AWS-controlled infrastructure; model providers do not see customer prompts, which is a major selling point for sensitive data.
  • Many commenters trust AWS’s track record, contracts, audits, and reputational incentives not to misuse data; others counter with skepticism about trusting any large US provider.
  • Some note US CLOUD Act implications: even if data is stored in a non-US region, a US company can still be compelled to hand it over, which is a blocker for certain EU/export-control use cases.

Competitive and Ecosystem Impact

  • This is viewed as huge for OpenAI and potentially worrying for Anthropic, which previously had a major advantage as the main high-end model on Bedrock.
  • Several expect this to hurt Azure, given earlier exclusivity and perceived reliability/scale issues.
  • Some criticize Bedrock’s lag in supporting newer open models, while others welcome increased model choice and competition.

Broader Observations

  • Cloud providers are seen as the new “immovable” enterprise platforms, similar to IBM/Oracle historically.
  • There’s ongoing debate about future AI progress given finite high-quality public data and increasing data “firewalls” in enterprises.