Most operators running LLM workloads in production overpay by 2–5x; some high-volume users pay an order of magnitude more. That's Meryem Arik's core claim in her QCon San Francisco 2025 talk, "Producing the World's Cheapest Tokens." The Doubleword co-founder has mapped a systematic playbook for inference cost reduction. General-purpose stacks like vLLM and SGLang optimize for chatbot latency, but teams running offline agents, labeling pipelines, or synthetic data generation pay for latency they don't need.

Arik frames the design space as a three-way tradeoff: latency, cost, and quality—pick two. Latency-first workloads (coding assistants, chatbots) run at low batch sizes on expensive hardware with latency-optimized kernels; Cerebras and Groq serve this corner. Quality-with-speed workloads (guardrails, routers, on-device) trade model size and precision to stay cheap and fast. The third corner—quality at low cost, latency irrelevant—is where Doubleword has focused for the past 6 to 12 months. Use cases: overnight agent workflows, summarization, data labeling, synthetic data generation for RL or fine-tuning. As models improve and autonomous workflows gain trust, this bucket will absorb more production workload.

The three inference workload corners: latency-first, quality+speed, and quality at low cost — each demands a different hardware and scheduling strategy.
FIG. 02 The three inference workload corners: latency-first, quality+speed, and quality at low cost — each demands a different hardware and scheduling strategy. — Meryem Arik, QCon San Francisco 2025 — "Producing the World's Cheapest Tokens"

Four levers drive cost reduction. First, batch-specific optimizations: general-purpose inference servers optimize for individual requests at low latency; switching to workload-aware scheduling—including queue reordering to maximize batch fill—changes the GPU utilization profile. Second, hardware selection: decoupling inference from P99 latency SLAs opens GPU tiers uneconomical for synchronous serving but efficient for throughput jobs. Third, speculative decoding: a smaller draft model generates candidate tokens that the target model verifies in parallel, compounding throughput gains across large batch runs. Fourth, quantization: INT8 or FP8 precision cuts memory bandwidth demand by 2–4x versus FP16 baselines, enabling larger effective batch sizes on the same GPU footprint.

LeverMechanismCost / Throughput Effect
Batch schedulingQueue reordering to maximize batch fill; workload-aware schedulingHigher GPU utilization; better $/token on offline jobs
Hardware selectionDecouple from P99 latency SLAs; use cheaper GPU tiers uneconomical for synchronous servingLower hardware cost per token for throughput jobs
Speculative decodingSmaller draft model generates candidate tokens; target model verifies in parallelCompounds throughput gains across large batch runs
Quantization (INT8 / FP8)Reduced precision vs FP16 baseline cuts memory bandwidth demand2–4× memory bandwidth reduction; larger effective batch size on same GPU
FIG. 03 Four levers for inference cost reduction in batch/async workloads — Meryem Arik, QCon San Francisco 2025

The compound effect is substantial. Doubleword markets its async and batch inference service 50–90% cheaper than real-time APIs for long-running, high-volume workloads. The platform is OpenAI-compatible and targets agents, evals, and pipelines—workloads that tolerate minutes of latency for dramatically lower $/million tokens. Industry data supports this direction: the Stanford HAI 2025 AI Index reports inference cost for a GPT-3.5-level system fell 280-fold between November 2022 and October 2024, driven by roughly 30% annual hardware cost declines and 40% annual energy efficiency gains. Inference consumes roughly 80% of AI infrastructure budgets. Training is a one-time event; inference runs on every request forever.

Inference cost for a GPT-3.5-level system fell 280× between November 2022 and October 2024, driven by ~30% annual hardware cost declines and ~40% annual energy efficiency gains.
FIG. 04 Inference cost for a GPT-3.5-level system fell 280× between November 2022 and October 2024, driven by ~30% annual hardware cost declines and ~40% annual energy efficiency gains. — Stanford HAI 2025 AI Index

Most teams reach for the popular runtime first. vLLM's continuous batching produces up to 23x throughput improvement over naive serving (per Anyscale benchmarks), but it's still designed around latency-sensitive request-response patterns. Adapting it for pure-throughput, high-latency workloads requires deliberate configuration—batch size limits, scheduler priority, queue depth tuning—that aren't defaults. Speculative decoding acceptance rates vary significantly by task type; picking the wrong draft model erodes speedup. Quantization at INT4 saves more memory but needs evaluation against task-specific quality thresholds before production rollout.

Arik's earlier TitanML work flagged a related pattern: teams deploying multiple fine-tuned variants of the same base model spun up separate GPU instances per fine-tune. LoRA-based multi-adapter serving—hosting one base model and hot-swapping fine-tuned adapters on the same GPU—collapses this into a single instance, cutting idle GPU-hours without sacrificing model specificity.

LoRA multi-adapter serving collapses N separate GPU instances into one, hot-swapping fine-tuned adapters on a shared base model to eliminate idle GPU-hours.
FIG. 05 LoRA multi-adapter serving collapses N separate GPU instances into one, hot-swapping fine-tuned adapters on a shared base model to eliminate idle GPU-hours. — Meryem Arik, TitanML / QCon San Francisco 2025

Direct takeaway: if no human watches the token stream in real time—agents, evals, labeling, synthetic data—you pay for latency you don't need. Queue reordering, async scheduling, quantization, and hardware-tier selection targeted at the non-real-time bucket deliver 50–90% cost reduction.