PIX has 40 seconds. From the user's tap to settlement in SPI, the Central Bank rejects the transaction if it exceeds this limit, according to the PIX Times Manual, BCB 2025. Within those 40 seconds, everything happens. And every millisecond is budgeted. We at Tech86 see PIX not as a payment product, but as a low-latency architecture problem with embedded compliance — where engineering and regulation are the same thing.
The anatomy of 40 seconds
Within the 40-second budget, each component consumes part of the time. The DICT query, with cache regulated by a 60-second TTL for CPF and EVP, according to BCB 2026. Balance validation in memory, never on disk — disk I/O kills the budget. ML fraud scoring, which must execute before SPI submission. ISO 20022 signing with an ICP-Brasil certificate in an HSM (FIPS 140-2 Level 3). Transmission via RSFN, with no public internet, where dedicated connectivity is mandatory. And settlement in SPI.
Every millisecond is budgeted. Each component consumes part of the budget. The problem lives on the outskirts of that budget — that is where latency accumulates invisibly, without distributed tracing that follows the transaction through all microservices. A single slow hop in the DICT cache lookup, a few extra milliseconds in HSM signing, or a retry on the RSFN path can quietly consume the margin that separates a settled transaction from a rejected one.
The SPI has no business hours
The SPI operates 24/7/365. No maintenance window. According to Res. BCB 195, 99.9% SLA equals 43 minutes of downtime per month. Every deploy is a production deploy. There is no business hours for instant payment.
This changes engineering. There is no safe window for rollout. No quiet early morning for rollback. Every change goes live into a system that cannot stop — and that rejects transactions exceeding 40 seconds. The architecture must treat every deploy as production, because that is exactly what it is. Blue-green deploys, canary releases, and automated rollback are not nice-to-haves; they are the only way to ship into a system that never sleeps.
The fraud paradox: preventive, not reactive defense
According to BCB, there were R$4.9 billion in PIX fraud losses in 2024, +70% versus 2023. MED recovery is 8.0%. With recovery so low, defense must be preventive. Scoring must run before SPI submission. After money leaves, it scatters to multiple accounts in seconds.
MED 2.0, in 2026, traces up to 11 days. But 11 days is not real time. The defense window is the 40-second budget — not the 11 days after. That is why fraud scoring must execute within the budget, before submission, without choosing between security and the 40-second window. WAF with Bot Management on payment endpoints is part of that preventive defense.
Volume makes the budget critical
According to BCB, there were 79.8 billion transactions in 2025. R$35.36 trillion moved. And according to Deloitte/Febraban, R$50.4 billion in banking tech investment in 2026, +58% over 5 years.
When the volume is 79.8 billion, every cent per operation is architecture. FinOps focused on cost per transaction stops being an optimization exercise and becomes a design decision. Distributed tracing that identifies where latency accumulates stops being observability and becomes a budget instrument. Every millisecond saved across 79.8 billion transactions is architecture translating into scale.
Regulatory isolation: BCB Resolution 538/2025
BCB Resolution 538/2025, in force since March 1, 2026, requires physical and logical isolation of the PIX environment. Dedicated cloud instance. Public multi-tenant may not be sufficient. According to Deloitte/Febraban 2025, private cloud investment grew 574% in 2025 — the market already responded to the regulatory requirement.
Isolation is not optional. It is regulation. And it reorders the infrastructure: dedicated cloud, dedicated RSFN connectivity, dedicated HSM, payment environment separated from the corporate environment. The architecture must be designed isolated from the start, not isolated afterward.
How Tech86 architects within the budget
We map every millisecond of this budget. Distributed tracing that follows the PIX transaction through all microservices and identifies where latency accumulates. FinOps focused on cost per transaction, because when the volume is 79.8 billion, every cent per operation is architecture. WAF with Bot Management on payment endpoints, so fraud scoring does not have to choose between security and the 40-second window. IaC with Terraform for multi-region DR in minutes, because 43 minutes of downtime per month is the ceiling, not the target. Managed cloud with a São Paulo datacenter, 99.99% SLA, minimal latency to DICT. Elite DeepTech with Hard Tech and Hard Law for environments where engineering and compliance are the same thing.
Conclusion
The 40 seconds are the budget. The rest is architecture. PIX is not a payment product — it is a low-latency system with embedded compliance, where every millisecond is budgeted, each component consumes part of the budget, and every deploy is production. We help banks and fintechs architect within this budget, with regulatory isolation, preventive fraud defense, and per-transaction FinOps. Before latency accumulates where nobody is looking.