The real economics of reselling LLM capacity at half price

When we tell people Grokified bills at 50% of xAI list price, the first question is always the same: how is that not just lighting money on fire? It is a fair question. A 50% discount on a commodity with thin margins should be a fast way to go out of business. The short answer is that the discount is not uniform, it is not permanent, and it is not applied to a cost base that looks like list price.

The discount is capped, on purpose

The headline 50% applies to your first $10,000 of usage, up to $5,000 in savings. After that, the rate steps down to 3% off list for life. That cap is the single most important number in the whole model. It turns an unbounded subsidy into a fixed, predictable customer-acquisition cost.

Framed that way, the math is ordinary. We spend up to $5,000 to acquire a developer who is, by definition, already spending real money on Grok. Compared to paid acquisition in developer tooling, that is competitive, and the people it attracts are exactly the ones with production workloads.

Our cost base is not list price

This is the part people miss. We do not buy capacity one request at a time at the same rate you would pay in the console. Reselling at volume means committed throughput and batched settlement, and a few structural advantages compound underneath the discount:

• Prompt caching across tenants' shared system prefixes. Cached input tokens are dramatically cheaper than fresh ones, and a proxy sees caching opportunities a single app never would.
• Batched upstream billing. We settle aggregate usage on a schedule, not per-call, which smooths the spread between what we are charged and what we charge.
• Routing. When a retired model slug maps to a cheaper current one with identical output, everyone wins and nobody changes code.

The 50% is not a price. It is a budget: a fixed amount we have decided each new builder is worth.

What it looks like in code

None of this changes your integration. The entire migration is one line, the same OpenAI-compatible SDK pointed at our base URL:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.grokified.com/v1",
    api_key=GROKIFIED_API_KEY,
)

So where does it stop being a subsidy?

The day you cross the cap. After $5,000 in savings, you keep a modest 3% off list for life, enough to stay cheaper than going direct, small enough that we are sustainably in the black on every token from that point on. The subsidy did its job: it got you onto the proxy with real production traffic, and now the relationship pays for itself.

That is the whole trick. There is no clever arbitrage and no asterisk, just a bounded, deliberate bet that the cheapest way to grow is to underwrite the first stretch of a builder's usage and earn the rest.