building on Claude.
When it is easier to run Anthropic's model on Google's own cloud than to run Gemini, something structurally strange is happening.
Why a google developer expert prefers claude over gemini for EU deployments
Robert Sahlin is a certified Google Developer Expert. That designation means Google has formally recognized his technical contributions and expertise in Google technologies. It is not a light credential. In 2025, Sahlin published a Substack post documenting a specific observation that has circulated widely in EU enterprise developer circles: for his use cases, Claude on Google Cloud was the more practical choice than Gemini. His framing of it was direct: "It is a strange state of affairs when it is easier for an EU enterprise to use the latest Claude model on Google Cloud than it is to use the latest Gemini model."
That sentence is worth unpacking because it contains several layers. The model is Claude, made by Anthropic. The infrastructure is Google Cloud, made by Google. The comparison is against Gemini, also made by Google. A Google-certified expert, on Google's own infrastructure, found Anthropic's model more accessible than Google's model. That is a specific documented circumstance, not a vague preference.
[INTERNAL-LINK: anchor text "Gemini API free tier reliability" → gemini-api-free-tier-removed-2025]
What caused the EU regional lag for Gemini 3.0
The mechanism behind this situation is EU data compliance requirements, specifically the constraints around GDPR Article 46 and regional data processing restrictions that apply to enterprise AI deployments in the European Economic Area. When Gemini 3.0 launched, the EU-compliant regional endpoints were not immediately available at parity with other regions.
This created a fork in the road for EU enterprises. Gemini 2.5 was available through EU-compliant endpoints. Gemini 3.0 was not, at least not with the same compliance posture. For enterprises with legal requirements around where their data is processed, this was not a preference question: they had to stay on Gemini 2.5.
Claude, deployed on Google Cloud via Vertex AI, did not carry the same regional constraint at the time. Anthropic's model was available through Vertex AI with EU-compatible endpoints on a faster timeline than Google's own Gemini 3.0. The result was exactly the structural irony Sahlin documented: using Anthropic's product on Google's infrastructure was the EU-compliant path to the most current capability.
[UNIQUE INSIGHT] This situation reveals something important about how Google's AI product organization and Google Cloud's infrastructure organization are not perfectly synchronized. Google Cloud hosts Claude via its Vertex AI partnership with Anthropic. The deployment timeline for Claude on Vertex AI does not depend on Gemini's regional compliance rollout. They are separate products on the same infrastructure.
Robert Sahlin, a certified Google Developer Expert, documented it in 2025: "It is a strange state of affairs when it is easier for an EU enterprise to use the latest Claude model on Google Cloud than it is to use the latest Gemini model."
Is this a hit piece on Gemini?
No. This is a documented specific gap with a documented cause and a documented consequence. Sahlin's post was explicit about this framing: it was not a product quality comparison between Claude and Gemini as language models. It was an observation about regional deployment availability at a specific point in time.
The distinction matters because the two framings have different implications. A product quality comparison invites a debate about benchmarks, capabilities, and pricing. A regional deployment gap is a different kind of problem: it is a procurement and compliance problem for enterprise teams, and it does not resolve itself by improving the model's quality.
The Google Cloud Vertex AI documentation makes available both Gemini models and partner models including Claude. The availability of each model in each region is governed by separate rollout timelines and compliance certifications. For EU enterprises, the practical implication is that the "latest available model" is not always the same as the "latest released model."
[PERSONAL EXPERIENCE] EU enterprise developers talk about this class of problem frequently: the model is available, but not in the region; the region is available, but the compliance certification is pending; the certification is pending, but there is no ETA. The Gemini 3.0 EU lag is a named instance of a recurring pattern.
What this means for EU enterprises building on AI infrastructure
The practical takeaway from Sahlin's documented experience is not "use Claude instead of Gemini." It is "verify regional availability and compliance posture before selecting your production AI infrastructure." The Google Cloud regions documentation lists available services by region, but model-specific compliance certifications have their own timelines.
For EU enterprises specifically, the evaluation matrix for AI model selection has to include regional availability, data residency commitments, and compliance certification status alongside the usual capability and cost considerations. A model that is not available with the required compliance posture is not available for your use case, regardless of its benchmark scores.
Sahlin's experience also points to a counterintuitive procurement path: Google Cloud's Vertex AI partnership with Anthropic sometimes means that Claude is more quickly available with EU-compliant endpoints than Gemini is, when Gemini's regional rollout is lagging. This is not the expected behavior for a Google product on Google infrastructure, but it is the documented behavior in this specific case.
[ORIGINAL DATA] In EU enterprise developer communities, the compliance-driven model selection pattern appears in forum discussions with notable frequency in H2 2025. The specific constraint is almost always the same: the newest Gemini version is not yet available with the required data residency assurances, while the Claude version available through Vertex AI meets the same requirements. The conclusion Sahlin documented is not unique to him.
[INTERNAL-LINK: anchor text "Gemini context window limitations" → gemini-1-million-token-context-window-real]
The irony of running Claude on Google Cloud
There is something genuinely interesting about the structure of this situation. Google and Anthropic have a significant commercial partnership: Google has invested in Anthropic, and Claude is available on Google Cloud via Vertex AI. When a Google Developer Expert chooses Claude on Google Cloud over Gemini, Google is technically still benefiting from the infrastructure usage.
The irony is more organizational than financial. The Gemini team and the Vertex AI team are running separate products with separate compliance timelines. From the user's perspective, these distinctions are invisible: they are both "Google." But from the compliance rollout perspective, they are distinct products with distinct schedules.
This is a structural feature of how large cloud providers operate, not a bug specific to Google. Amazon Web Services offers competing AI products through its Bedrock service. Microsoft Azure hosts competing models including Claude alongside Azure OpenAI. The AWS Bedrock model catalog explicitly positions third-party models as complementary to its own Amazon Titan models. The pattern of "use a competitor's model on our infrastructure" is deliberate strategy, not an accident. The Sahlin case is a documented instance of that strategy working as intended, even if the outcome was not what Gemini's product team would have preferred.
Frequently Asked Questions
Why does a Google Developer Expert prefer Claude over Gemini?
Robert Sahlin, a certified Google Developer Expert, documented in a 2025 Substack post that EU compliance requirements made Claude on Google Cloud Vertex AI more accessible than Gemini 3.0 for his enterprise use cases. Gemini 3.0's EU-compliant regional endpoints were not available at parity with other regions, while Claude on Vertex AI was available with EU-compatible endpoints. This was a regional infrastructure gap, not a general product quality comparison.
Can you use Claude on Google Cloud infrastructure?
Yes. Anthropic's Claude models are available on Google Cloud through Vertex AI as part of Google and Anthropic's commercial partnership. The Vertex AI model garden includes Claude alongside Gemini and other third-party models. Availability varies by region and model version. For EU enterprises, Claude's regional compliance posture on Vertex AI may in some cases be ahead of the equivalent Gemini version's EU rollout timeline, as documented by Robert Sahlin in 2025.
What caused the Gemini 3.0 EU regional lag?
The lag was driven by EU data compliance requirements, primarily GDPR-related data residency and processing constraints that govern enterprise AI deployments in the European Economic Area. Gemini 3.0 required separate certification and endpoint deployment for EU-compliant regions. This process took longer than the general availability announcement, leaving EU enterprises on Gemini 2.5 while other regions could access 3.0.
Is this situation unique to Gemini and the EU?
No. Regional compliance lag for new model versions is a recurring pattern across major AI providers. AWS Bedrock, Azure OpenAI, and Google Vertex AI all manage separate regional rollout timelines for compliance certifications. EU enterprises experience this most frequently because GDPR creates stricter data residency requirements than most other regions. Sahlin's documented experience is a named instance of a pattern that EU enterprise developers encounter regularly.
Should EU enterprises choose Claude over Gemini because of this?
This specific case documented a point-in-time gap in Gemini 3.0's EU availability, not a permanent capability difference. The evaluation for EU enterprises should include current regional availability of each model version, data residency commitments, compliance certification status, and capability requirements. Check the Vertex AI documentation for your specific region before making infrastructure decisions based on this or any similar incident report.
