The Uncomfortable Truth
Every microsecond of signing latency costs your fund money. Not theoretically — measurably. We call this Alpha Decay: the exponential loss of trading alpha as execution latency increases.
This isn't speculation. It's physics applied to finance.
The Math: Exponential, Not Linear
Alpha Decay follows an exponential curve. A 2019 BIS study on high-frequency FX trading found that a firm's profitability decays as:
α(t) = α₀ · e^(-λt)
Where:
α₀= maximum theoretical alpha at zero latencyλ= decay constant (market-dependent, typically 0.01–0.1 per µs)t= total execution latency in microseconds
What This Means in Practice
| Signing Latency | Alpha Retained | Annual Impact ($10M/day vol.) |
|---|---|---|
| 42 µs (Sentinel) | 99.6% | -$0 |
| 5 ms (Internal HSM) | 95.1% | -$1.2M |
| 150 ms (AWS KMS) | 22.3% | -$10.4M |
| 350 ms (Fireblocks) | 3.0% | -$15.8M |
The decay is exponential. Going from 42µs to 5ms doesn't cost you 100× more — it costs you 4.5% of your annual alpha. Going to 150ms costs you 77%.
The Jitter Tax
Raw latency is only half the story. The other half is jitter — the variance in latency between requests.
Deterministic latency (low jitter) allows your strategy to account for execution time. Non-deterministic latency (high jitter) forces your strategy to add safety margins — padding that costs you trades.
| Provider | P99 Latency | Jitter (σ) | Safety Margin Required |
|---|---|---|---|
| Sentinel | 42 µs | ±2 µs | 2 µs |
| AWS KMS | 312 ms | ±42 ms | 84 ms |
| Fireblocks | 450 ms | ±120 ms | 240 ms |
The Jitter Tax is the hidden cost of that safety margin. Your strategy is slower than it needs to be because your signer is unpredictable.
Why Cloud HSMs Are Fundamentally Broken for HFT
Cloud HSMs (AWS KMS, Azure Key Vault, GCP Cloud KMS) were designed for web applications — not microsecond-sensitive financial infrastructure.
Every signing request follows this path:
Your Server → Network → Load Balancer → HSM API → HSM Hardware → Return
That's a minimum of 3 network hops. Each hop adds:
- TCP handshake overhead
- TLS negotiation
- Queue wait time (shared infrastructure)
- Geographic latency
You cannot optimize away physics. Light travels 300m per microsecond. If your HSM is in a different availability zone (typically 1–10km away), you're paying 3–33µs just for photons.
The Sentinel Approach: Elimination, Not Optimization
We don't optimize the network path. We eliminate it.
Sentinel Prime runs inside an AWS Nitro Enclave — a hardware-isolated virtual machine on the same physical server as your trading application. The signing key never leaves the enclave. Communication happens over vsock, a direct memory channel:
Your App → vsock (in-memory) → Enclave → Sign → vsock → Your App Total: 42µs
No network. No load balancer. No queue. No jitter from shared infrastructure.
Quantify Your Jitter Tax
Use the zcp CLI to measure your current signing latency:
brew install zerocopy-systems/tap/zcp zcp audit --volume YOUR_DAILY_VOLUME
The audit produces a grade (A–F) and estimates your annual P&L impact.
Conclusion
Alpha Decay is not a marketing concept — it's a measurable physical phenomenon. Every fund that signs transactions through a cloud HSM is paying a Jitter Tax they may not even know exists.
The question isn't whether you can afford sovereign signing infrastructure. The question is whether you can afford not to have it.
See how your infrastructure compares.
Jitter Tax Calculator