Vela

Smart Route

The spot market for AI compute

2026 | © Vela, Inc.

Hi! We are

Alex
Rivera

Built developer tooling and inference platforms for a decade. Shipped APIs used by thousands of teams.

Background across the full stack, from kernels to billing.

Sam
Chen

Software engineer and product lead. Specializes in user experience and turning messy infra into simple products.

M.Sc. in distributed systems.

We worked with

2026 | © Vela, Inc.

We offer: Developer Tooling In The Age of AI

We test the market and improve the offering. So far, all organic growth.

Zero ad spend.

2026 | © Vela, Inc.

$765 billion in GPUs. ½ sit idle.

40-80%

Inference workloads dip to 15–30% off-peak. A rack of GPUs costs $4M. At 40% utilization, half that investment earns nothing.

10x

Same GPU, same hour: $1.49 to $10.00. Up to 14.5× on spot.

60+

Tracked and growing. No aggregated spot market exists yet.

2026 | © Vela, Inc.

Route the tokens. Or rent the GPU.

Smart inference routing

Drop-in OpenAI-compatible API. Change one line of code. Every request scored and routed to the cheapest healthy provider, with auto-failover.

GPU spot brokering

We find and provision the cheapest spot GPU for your model. Same scoring, same reliability layer, automatic migration. Up to 80%+ discount.

GPU	Pool	$/hr	Trend
B300 · 2100 GB	bolt	$2.99	▲ 20%
B200 · 1440 GB	bolt	$2.44	▲ 13%
H200 · 1128 GB	drift	$0.60	▼ 5%
A100 · 640 GB	bolt	$0.71	▲ 10%
H100 · 640 GB	bolt	$1.30	▲ 23%
L40S · 384 GB	bolt	$0.55	▲ 37%

How Smart Route works

Drop-in. Change one line of code. Every request is scored on price and health, then routed to the cheapest healthy provider.

Resilient. Automatic failover and live migration across 60+ providers — no lock-in, no babysitting.

Spot price by GPU ($/hr)

$765B in racks — half of it idle.

The world is racing to build compute. We make the idle half earn its keep.

Photo by Taylor Vick on Unsplash

Vela, Inc. · [email protected] · vela.dev