Google Cloud does not expect enterprises to standardise on Gemini alone. Instead, it believes the future of enterprise AI will be built on a mix of frontier, open, and third-party models, giving customers the flexibility to choose the right model for each workload.
“The world is probably going to be multi-model,” Richard Seroter, Chief Evangelist at Google Cloud, tells AIM during a media roundtable at Google I/O Connect in Bengaluru.
With governments increasingly influencing access to frontier AI models through export controls and regulations, enterprises are beginning to question whether the models they build on today will remain available tomorrow.
While expressing confidence that Google’s own models would remain available, Seroter says no company can control the decisions made by competitors or governments. He was referring to Anthropic’s Fable 5 and Mythos 5, whose access restrictions had been lifted just a few days earlier.
Despite growing competition from lower-cost Chinese AI models such as GLM 5.2 and DeepSeek V4, Seroter says Google has no intention of forcing customers to use Gemini.
Instead, customers can deploy competing models available through Google’s Model Garden on Google Cloud infrastructure.
“I want you to use Gemini. That would be great, but that is not required for us to be successful.” He adds that Google’s strategy is to provide customers with flexibility rather than lock them into a single model ecosystem.
This approach is reflected in Model Garden, its AI model catalogue on Vertex AI. Instead of restricting customers to Gemini, Model Garden offers access to hundreds of models, including Google’s Gemma family as well as models from Anthropic, Meta, Mistral AI, DeepSeek, Kimi, Z.ai and thousands of open models from Hugging Face.
The platform allows enterprises to mix and match models based on performance, cost and use case rather than committing to a single provider.
That said, Google is expected to release Gemini 3.5 Pro soon, positioning it to compete with GPT-5.6 and Fable 5.
Small Language Models
Google expects small language models to become increasingly important, particularly for organisations that want AI to run locally or be tailored to specific industries.
Seroter says Google’s strategy extends beyond frontier models, with Gemma capable of running locally on phones while Gemini Nano brings on-device AI capabilities to Chrome. “You don’t always need a trillion parameters. Sometimes you just need something that can run locally.”
He further adds that organisations can fine-tune these smaller models for sectors such as healthcare, education and legal services, while continuing to use larger frontier models only where necessary.
Token Pricing for India
Seroter acknowledges that enterprises are becoming more conscious of AI spending as token pricing becomes more transparent. While some AI providers have increased prices after initially subsidising access, he said Google's pricing has remained relatively stable.
“We’re always going to try to be best on price performance,” he says. “The cost has gotten more transparent. It’s less subsidised than before.”
That said, Google sees its Gemma family of open models playing an important role here.“You can just run Gemma on hardware, your laptop, a Kubernetes cluster, a VM. There’s no token cost. It’s a fixed infrastructure cost,” he explains.
What’s the Buzz about Forward-Deployed Engineers?
Like several frontier AI companies, Google is investing heavily in forward-deployed engineers (FDEs), teams that work directly with enterprise customers during AI implementation.
According to Seroter, the challenge isn’t that AI products are difficult to use. The real difficulty lies in organisational transformation.
Earlier this year, Google Cloud launched a $750 million fund to help partners build and deploy agentic AI. The initiative will support AI prototyping, deployment, workforce training, and provide customers with access to Google’s forward-deployed engineers (FDEs).
Seroter says that AI adoption often requires enterprises to retrain employees, rethink IT operating models and change tooling, making hands-on engineering support increasingly valuable.
- Why AI is Eating Google Search’s Ad Revenue While Boosting Meta’s
- Anthropic Makes $2.08 in Revenue for Every $1 Spent on Compute — Here’s the Full IPO Story
- OpenAI Says AGI is Here. But Not Everyone is Buying It
- Why Oracle is Bringing Post-Quantum Security to Java
- What Really Happens When You Turn Your Selfie Into a 1980s AI Pic?
- AI Agents Are Making Data Security A Bigger Enterprise Problem