DigitalOcean

Hardcoding model selection is a scaling nightmare, but using frontier models for every simple task is a margin killer. At DigitalOcean, we built a solution that doesn't force that tradeoff. 💡 Our Inference Router routes each request to the right model in ~200ms, using live cost and latency metrics instead of static configs. Built on Envoy, WASM, and Rust. 🌩️ The result? Routing that adds milliseconds, not seconds, to every request. Stop maintaining brittle routing logic and start shipping agentic workflows that actually scale. 🌊

How fast can you set up infra, run an experiment, and get insights to the product team? 💭 These are the criteria the team at Workato's AI Research Lab used when evaluating every major cloud provider as they built AI infrastructure for distributed training, fine tuning, and inference. DigitalOcean delivered on time-to-value along with additional performance gains, and Workato’s AI Lab experienced: ✅ 67% cost savings on inference ✅ 67% better throughput ✅ 77% faster time to first token

Every team has access to the same models. The real differentiator is what's around them. 🏗️ Frontier models are becoming a commodity—which means your routing logic, data pipelines, and cost controls are where you actually compete. And yet most teams are still re-platforming at every growth milestone. At DigitalOcean, we unified serverless inference, dedicated GPUs, and intelligent routing into one continuous surface. Our Inference Router matches prompts to the right model pool automatically—cutting costs by up to 97% for standard tasks. No code rewrites. No provider lock-in. 🌩️

Day 2 of AI Council San Francisco: moving AI from prototype to production. 🌁🤖☁️ Stop by our booth tomorrow to learn how we’re removing the "complexity tax" to help you scale AI agents with simplicity and speed.

Day 1 at AI Council San Francisco. ✅ Stop by our booth tomorrow to learn more about our AI-native cloud. ☁️🤖

Frontier-scale intelligence, simplified. 🧠🆕☁️ Kimi K2.6 is now available through our AI-Native Cloud, giving developers the reasoning power of a trillion-parameter model with the efficiency required for production. Running natively via the DigitalOcean Inference Engine, Kimi K2.6 is purpose-built for autonomous workflows & multi-agent orchestration. No extra vendors, no operational overhead—just security-hardened infrastructure with predictable, usage-based billing. Ready to scale your AI workflows? Check the link in our comments. ⬇️

DigitalOcean just had its best quarter ever. 📈 • $258M Q1 Revenue, up 22% YoY • 221% YoY AI Customer ARR Growth at $170M • 179% YoY $1M+ Customer ARR Growth at $183M • $62M Record Incremental Organic ARR, at 168% YoY • 41% Adjusted EBITDA margin • 18% TTM Adjusted FCF Margin “The inference and agentic era needs its own cloud. This quarter proves it. 221% AI ARR growth. Record organic ARR. 22% revenue growth. The inference economy isn't coming, it's here, & DigitalOcean’s AI-Native Cloud is powering it." - Paddy Srinivasan, CEO DigitalOcean ☁️

Now Available: DeepSeek-V4-Pro on DigitalOcean’s AI-Native Cloud. 🌊 DeepSeek AI’s latest frontier model is ready for your complex, multi-step AI workflows. Built on a sparse Mixture-of-Experts (MoE) architecture, DeepSeek-V4-Pro is optimized for agentic reasoning & multi-step task execution. It runs natively alongside your apps & data on DigitalOcean. Why developers are choosing DeepSeek-V4-Pro on DigitalOcean: 🧠 Agentic Planning: Optimized for autonomous task execution. 📄 1M Context Window: Ingest entire codebases or massive datasets. 💰 Predictable Pricing: Usage-based billing with no additional vendors or contracts. No extra vendors, no complex contracts. Just production-grade AI at scale.

Among the fastest DeepSeek V3.2, MiniMax-M2.5, & Qwen 3.5 397B inference in the market, per Artificial Analysis benchmarks (April 2026). ⚡️ Sub-1-second TTFT. 230 tokens per second. Not from better GPUs alone—we co-designed every layer of the stack, including vLLM performance optimization, on NVIDIA HGX B300. In production, Workato has seen up to 77% faster TTFT, 79% lower latency, and 67% lower inference costs. Results may vary by configuration and usage. All three models are live on DigitalOcean Serverless Inference. Full engineering breakdown in the comments. ⬇️