$40 billion. That's what Google is currently investing in Anthropic. The shape of the deal matters more than the headline number, so let's take it apart.
The numbers.
- $10 billion in cash now, at a $350 billion valuation.
- $30 billion on top if Anthropic meets its performance targets.
- 5 GW of TPU compute committed over five years.
This is not a strategic dabble. This is a bet on the entire AI infrastructure of the next decade — compute, capital, and a long-running commercial relationship welded together.
Google Cloud is reportedly earning more from the Anthropic API than from Gemini. Google is hosting its direct competitor — and generating more revenue from it than from its own model.
What the deal actually says.
Antigravity and the current Gemini 3.1 models are, by most working benchmarks, among the worst in market right now. Perhaps that says more about the deal than the $40 billion does. When the cloud arm of the world's second-largest tech company is making more money reselling someone else's frontier model than its own, the rational move is to lock that someone else in. Which is exactly what this deal does.
And as if that weren't enough: SoftBank Group Corp. is reportedly planning to invest up to another $40 billion in Anthropic separately. $80 billion, on a single AI company, from two of the world's largest investors.
When two of the largest pools of capital on earth converge on the same model family, the question for delivery teams stops being "which AI?" and becomes "how fast can we ship with it?"
— prodct [Labs] · April 2026
Why this validates what we do.
One more reason we feel fully validated in our strategy. The stack we use isn't a thesis — it's the operating layer behind every shipping engagement at prodct:
- We rely on Google Cloud as the basis and foundation for our projects and rapid delivery.
- We use Anthropic to deliver even more efficiently and to build the agents and models that fit the engagement.
- We're ready. The market seems to see it that way too.
What this means for European retailers.
Three things change in the next 12 months for retail and commerce teams already shipping with Claude:
1. Model access tightens, then loosens.
Capacity will be uneven for the next two quarters as the TPU build-out catches up to demand. By Q4 2026, expect noticeably better latency on EU regions and clearer commercial terms for production workloads.
2. Contractual ground gets firmer.
Enterprise contracts via Google Cloud Marketplace become the cleanest path for teams that need EU data residency, GDPR-compliant processing, and a single procurement line. We expect this to be the default for European buyers by year-end.
3. The "which model?" question gets less interesting.
For 80%+ of retail use cases — merchandising assistants, content generation at SKU scale, internal copilots, customer-service routing — the model differences are now within noise. The real differentiation moves to the data layer underneath, and to the operating model wrapped around the agents. That's where we focus.
The bet we'd make this quarter.
If you're a head of commerce or data at a European retailer with a budget signal in 2026, the bet is straightforward: do not wait for the model market to settle. Ship a working agent against one real operational problem, on Google Cloud, against Claude, in the next eight weeks. The infrastructure question is now answered for you. The delivery question is not.
If that's a useful lens for your team, talk to us. We've shipped this pattern eleven times in the last four quarters and we're getting faster at it.
