crawshaw (Tailscale co-founder) launches exe.dev, a cloud built around CPU/memory pools, local NVMe, and anycast networking to fix cloud’s wrong-shaped abstractions.
Key Takeaways
Cloud VMs force fixed CPU/memory bundles; exe.dev sells raw compute you slice into as many VMs as you want.
Remote block storage fit HDDs (10ms seek, 1ms RTT tolerable) but SSDs dropped seek to 20 microseconds, making remote IOPS overhead 10x worse.
Cloud egress costs 10x datacenter rates; Kubernetes cannot fix underlying abstraction problems, it is lipstick on a pig.
exe.dev ships CPU/memory pools, local NVMe with async off-machine replication, TLS and auth proxies baked in, anycast network, and global regions.
Agents drive Jevons paradox in software output: more total programs compounds cloud abstraction tax, and bad primitives waste LLM context window directly.
Hacker News Comment Review
Tailscale irony surfaced immediately: commenters noted Tailscale itself draws battery drain and firewall-modification complaints, raising questions about whether exe.dev can hold its clean-abstraction line as it scales.
K8s debate split along org-size lines: platform teams in financial services argued complexity is org-imposed, not DevOps padding; critics said clusters start simple then inevitably balloon with software-defined networking layers.
Jevons paradox framing drew skepticism: app stores already overflow with unused software, and the agents-write-more-code demand argument may not translate to meaningful new cloud consumption.
Notable Comments
@sahil-shubham: built open-source Firecracker orchestrator (bhatti.sh) independently for the same goal: buy auctioned Hetzner hardware, carve into VMs, skip cloud provisioning overhead entirely.
@faangguyindia: Hetzner plus self-managed Postgres HA, 10 years uptime, 1/10th RDS cost; evidence the bare-metal-plus-scripts path already works at production scale.
