... of infrastructure software.
The low cost of training and inference of the new model could trigger a new wave of AI adoption by improving the economics of AI projects and thus enabling wider use.
Increasing demand through lower costs:
DeepSeek's model offers superior inference performance at a disruptively low cost, which is expected to increase demand for AI solutions.
Bernstein points to previous examples, such as OpenAI's 4o-mini model, which promised lower prices with higher quality and increased usage by 50% within a quarter. Such effects could also occur again with DeepSeek.
Winners at the software infrastructure level:
Despite lower computational intensity, higher demand leads to opportunities for software vendors that provide the infrastructure for AI projects. This includes areas such as:
Data infrastructure (e.g. Confluent $CFLT (+1,06%) )
Identity management (e.g. Okta $OKTA (+0,55%) )
Observability tools (e.g. Datadog $DDOG (-0,5%) )
Communication technology (e.g. Twilio $TWLO (-1,2%) )
Browser integration and delivery (e.g. Cloudflare $NET (+0,39%) )
Platforms for service desk AI projects (e.g. ServiceNow $NOW (+1%) ) could also benefit, especially if they introduce a consumption-based pricing model that covers inference costs.