Twitter will move its offline analytics, data processing, and ML workloads to Google's Cloud after agreeing to an expanded multiyear partnership
Mary Ann Azevedo / TechCrunch :
Context & Ripple Effects
This deal lands weeks after Twitter signed a multiyear AWS agreement to power its timelines, meaning the company is now splitting its infrastructure between two hyperscalers rather than running it all in-house. The Whetlab acquisition back in 2015 signaled Twitter's early bet on in-house machine learning; moving analytics, data processing, and ML workloads to Google Cloud marks a partial retreat from that build-it-yourself posture.
The stakes of this dependency became visible two years later, when unpaid Google Cloud bills — reportedly $20M+ per month — forced CEO Linda Yaccarino to personally intervene to repair the relationship before payments resumed.
First-order effects
- Twitter's internal data science and ML teams shift offline analytics and model training onto Google Cloud, reducing the load on Twitter-owned infrastructure while timelines stay on AWS under the separate December 2020 deal.
- Google Cloud adds a marquee consumer-internet customer with steady, compute-heavy workloads at a moment when it was competing against AWS, which had just won Twitter's timeline serving business.
Second-order effects
- Twitter locks itself into a multi-cloud cost structure where two vendors each hold a critical workload — a dependency that later surfaced as monthly Google Cloud bills exceeding $20M and a public payment dispute.
- The split gives each hyperscaler leverage at renewal time: AWS owns serving latency-critical traffic while Google owns the data and ML pipeline, making either migration costly and slow.
Third-order effects
- If the pattern holds, even large consumer platforms stop operating their own analytics and training infrastructure entirely, turning hyperscalers into utility providers whose contracts outlast product cycles — a trajectory echoed by Google's later tens-of-billions TPU commitment to Anthropic.
- Cloud relationships become board-level exposures rather than procurement line items, as Twitter's payment standoff showed: a single vendor bill can escalate to CEO-level diplomacy.
The trend: Consumer internet platforms are steadily dissolving their owned compute into hyperscaler clouds for analytics and ML workloads, converting infrastructure from an asset they run into a dependency they negotiate.