Cerebras Systems vs infrai

Cerebras and Compute Nordic Finland Announce New 165 MW AI Data Centre in Mikkeli, Finland

Cerebras Systems and Compute Nordic Finland announced a new 165 MW AI data centre in Mikkeli, Finland. The company also unveiled the CS-4, claiming up to 30 times faster than GPU-based solutions, and announced a partnership with Callosum for ultra-low-latency heterogeneous agentic inference.

Cerebras shares fall 14% after Q2 earnings

Cerebras Systems shares dropped about 14% in after-hours trading after the company reported second-quarter results following its May IPO. The chipmaker reported core revenue of $180 million (below LSEG consensus of $194 million) and an adjusted loss per share of $0.05 versus $0.17 expected, while total revenue was $210 million including pass-through items. Cerebras raised full-year core revenue guidance to $880–$890 million and said it expects core revenue of $214–$216 million this quarter. The company recorded a net loss of $450.5 million, driven largely by $386.6 million of stock-compensation costs. CEO Andrew Feldman said AI demand is "through the roof," and the company cited partnerships with AMD and usage of its chips by OpenAI for GPT 5.6 Sol.

• Cerebras reported core revenue of $180 million for Q2, below LSEG consensus of $194 million.
• Adjusted loss per share was $0.05 versus $0.17 expected; total revenue was $210 million including pass-through revenue.
• Cerebras raised full-year core revenue guidance to $880 million–$890 million (previously $855 million–$865 million) and expects current-quarter core revenue of $214 million–$216 million.

Lovable and Cerebras Announce Inference Partnership

Lovable and Cerebras Systems announced a partnership to run Lovable’s latency-sensitive inference workloads on Cerebras’ high-performance inference infrastructure. Lovable — which says more than 50 million projects have been built on its platform since its November 2024 launch — will use dedicated Cerebras capacity to reduce response times and enable more interactive, multi-step software creation workflows. Cerebras’ Wafer-Scale Engine is highlighted as keeping an entire model’s weights on a single wafer to deliver greater memory bandwidth and faster token generation than GPU-based systems. Both companies said they will jointly explore new product experiences enabled by faster inference; technical results and availability details will be shared later.

• Lovable and Cerebras Systems announced a partnership to power Lovable’s software creation platform with Cerebras’ inference platform.
• Lovable will run selected latency-sensitive workloads on dedicated Cerebras capacity.
• Lovable reports more than 50 million projects have been built on its platform since its November 2024 launch.

SMS Delivery Status Polling for Waitlist Outage Alerts

The article advises that teams should only rely on an SMS API for critical outage alerts if their backend can poll delivery status and own retry, escalation, cancellation, and timing logic. Delivery reliability and timing constraints drive the design: define service-level objectives, record four reliability invariants (application-owned send IDs, bounded/idempotent retries, defined next actions per delivery state, and incident recovery that suppresses obsolete alerts), and treat providers as transport adapters. The author shortlists Twilio, Vonage, Sinch, and Infrai for evaluation, provides load-testing guidance, and includes a runnable Python example that polls SMS status, honors Retry-After, and applies backoff. The recommended architecture keeps durable incident state in the application and makes provider polling a replaceable adapter.

• Choose an SMS API for critical outage alerts only if the backend can poll delivery status and implement retry, escalation, cancellation, and timing logic.
• Four reliability invariants: application-owned identifier per send; bounded and idempotent retries; every delivery state must map to a defined next action; incident recovery must stop obsolete alerts.
• Article shortlists Twilio, Vonage, Sinch, and Infrai as candidate SMS providers to validate against the same decision record.

Backend-Owned SMS OTP: Cooldowns and Attempt Caps

This technical blog post explains best practices for implementing passwordless phone logins using SMS OTPs in an Express/Node.js backend. It argues that the backend must own resend cooldowns, verification attempt counters, and anti-abuse policies (not the client), model the authentication state machine (ready → code_sent → verified/expired/locked), persist minimal authoritative state, use atomic database transitions, emit single transition events for observability, and use idempotency keys and retry/backoff handling when calling providers. Provider choices (Twilio, Firebase, Auth0, Amazon SNS, Infrai) are discussed with trade-offs between managed verification and owning template/state-machine responsibilities.

• The article recommends the Express/Node.js backend should own SMS OTP resend cooldowns, maximum verification attempts, and anti-abuse counters rather than trusting the client.
• Designs should expose explicit states: send-code, verify-code, resend-code, and lockout; persist minimal authoritative state (challenge ID, phone identity, expiry, next-send time, counters, lockout).
• Use atomic database transitions and idempotency keys tied to admitted transitions to prevent race conditions and duplicate sends.

Bulk LLM Text Classification with Tenant Chargeback

The article recommends treating tenant accounting as the primary artifact when performing bulk CSV moderation with LLMs: create a tenant-owned job with stable row IDs, estimate costs before submission, submit asynchronous batch classification (preferably chat classification with a closed label set), and attach returned results and export references to the same tenant ledger for reconciliation. The author provides an example TypeScript batch submission pattern (idempotency derived from the validated request, bounded retries, handling 429), argues for allocating costs at the job boundary and reconciling at the row level, and discusses when to call providers directly (Infrai, OpenAI, Anthropic, Google Gemini) versus renting batch execution.

• Author recommends asynchronous chat classification with a closed label set and using a tenant ledger as the primary artifact for billing and reconciliation.
• Pattern: create a tenant-owned job with a stable ID per accepted CSV row, show an estimate before submission, persist the provider batch identifier, then reconcile results and costs back to the job and rows.
• TypeScript example demonstrates deriving an idempotency key from the validated batch-request.json, honoring Retry-After for HTTP 429, and using bounded exponential backoff.