The AI Token Is the New Currency
October 19, 2026
In May 2026, China Telecom started selling AI tokens the same way it sells mobile data — as monthly subscription packs. ¥9.9 a month gets you 10 million tokens. ¥49.9 gets you 80 million. You top up through the same app you use to pay your phone bill.
China Mobile and China Unicom followed within weeks. China Mobile offered 400,000 tokens for ¥1 through a Tencent partnership. China Unicom launched plans in Shanghai starting at ¥15 for 6 million tokens. All three carriers bundled access to domestic models like DeepSeek and Zhipu GLM-5, and they're targeting 200 million consumer accounts by the end of 2026.
Meanwhile, at MWC 2026 in March, Jio's CEO Mathew Oommen stood on stage and said what everyone in telecom was already thinking: the industry is moving from minutes to data to tokens.
That's the shift. Not a metaphor. A literal restructuring of what telecom companies sell.
The playbook we've seen before
If you were in India in 2016, you watched Jio do something remarkable. They made mobile data essentially free — ₹10 per GB when the industry was charging ₹250. They bet that if you remove the cost barrier, usage explodes, and the revenue follows from scale. It worked. India went from 200 million internet users to over 800 million. Jio built the infrastructure, subsidized access, and captured the market.
The token play is the same playbook, one layer up.
In the data era, the product was bandwidth. You paid for gigabytes. The underlying infrastructure was cell towers, fiber, and spectrum. Telecoms built the pipes and charged for flow.
In the token era, the product is intelligence. You pay for compute — specifically, for the tokens that represent interactions with AI models. The underlying infrastructure is GPUs, data centers, and model serving. Telecoms are building a new set of pipes and charging for a new kind of flow.
Jio has announced a ₹10 lakh crore ($120+ billion) seven-year investment in AI infrastructure. A new subsidiary called Reliance Intelligence is building gigawatt-scale AI data centers in Jamnagar. They're developing proprietary models and a seven-layer architecture stack from chips to consumer apps, including JioAI — a voice-first, multilingual assistant designed for a country where hundreds of millions of users don't type in English.
The stated goal: deliver the lowest cost per token in the world. The same playbook. Make it cheap, make it accessible, capture the billion-user market.
What a token actually costs
To understand why telecoms see an opportunity here, you need to understand the economics.
Current API pricing from the big model providers looks something like this:
| Provider | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| GPT-5 Turbo | ~$3 | ~$12 |
| Claude Sonnet 5 | ~$2 | ~$10 |
| Gemini 3.1 Pro | ~$2 | ~$12 |
| DeepSeek (off-peak) | <$1 | <$2 |
Output tokens cost 2-4x input tokens because generation is computationally harder than comprehension. Long-context prompts carry surcharges. Caching can cut input costs by 50-80%. Batch processing and off-peak windows trade latency for savings.
The important trend: prices have dropped roughly 90% since 2023. This is the same deflationary curve that made mobile data cheap enough to sell as a mass-market product.
For a consumer, a single ChatGPT conversation might use 2,000-5,000 tokens. A heavy user might consume 500,000 tokens a day. A developer running agentic workflows — the kind where AI agents loop, reason, use tools, and generate code — can burn through millions of tokens per session.
China Telecom's ¥9.9 plan (10 million tokens) is roughly $1.45. That's cheaper than most direct API subscriptions, though developers have pointed out that the actual per-token cost works out to 5-6x what you'd pay going directly to DeepSeek. The carrier is adding a margin, but also adding something the API doesn't give you: bundled billing, local compliance, and a one-tap purchase flow that 900 million existing subscribers already know how to use.
The real product isn't tokens — it's distribution
Here's the thing most tech people miss about the telecom play: the tokens themselves aren't the competitive advantage. Distribution is.
OpenAI has roughly 400 million users. Impressive. China Mobile has 1 billion subscribers. Jio has 490 million. These carriers already have billing relationships, KYC verification, customer support infrastructure, and distribution channels that reach every tier-3 city and rural village.
When China Mobile sells you 400,000 tokens for ¥1 through your phone bill, the friction is zero. No credit card. No API key. No developer account. Just a button in the app you already use.
This matters enormously for the next billion AI users — the ones who aren't developers, who don't have credit cards linked to OpenAI, who interact with AI through voice in languages that aren't English.
Grameenphone in Bangladesh is already doing this — bundling access to ChatGPT, Claude, Grok, DeepSeek, and Gemini as telecom add-ons. e& in the UAE is bundling agentic AI into business mobile tariffs. Huawei pitched the "token economy" as the next telecom revenue model at MWC Shanghai 2026.
Western carriers haven't moved yet. They will.
From dumb pipe to intelligence broker
The strategic fear of every telecom company for the past fifteen years has been "dumb pipe" — the idea that carriers become commodity infrastructure while the value captures happen at the application layer. Google, Apple, Meta, and Netflix built empires on top of telecom pipes. The pipes got paid, but the pipes didn't get rich.
AI tokens offer telecoms something they haven't had since the SMS era: a per-unit revenue model for a digital product they can actually control.
With data, the carrier just moves bytes. The carrier doesn't know or care if those bytes are a Netflix stream or a WhatsApp message. The value is in the application, not the pipe.
With tokens, the carrier can be the broker. They can negotiate bulk model access, serve it through their own infrastructure, bundle it with connectivity, and bill it to existing subscribers. They're not just moving bytes — they're selling the output of intelligence.
This is the shift from infrastructure provider to intelligence provider. And the telecoms with the largest subscriber bases and the most aggressive infrastructure investments are positioning for it first.
What this means for builders
If you're building AI products, the telecom token economy changes a few things:
Distribution channels multiply. Today, AI access flows through API providers (OpenAI, Anthropic, Google) and app stores. Tomorrow, it also flows through carrier plans. Your AI-powered app might be bundled into a Jio plan the way Hotstar was bundled into a Jio data pack.
The cost floor keeps dropping. When Jio enters a market with the explicit goal of offering the cheapest per-token cost, prices fall for everyone. The same competitive dynamics that made Indian mobile data the cheapest in the world will play out for tokens.
Voice-first, multilingual AI becomes the default. Jio isn't building JioAI for developers who type English prompts. It's building for the hundreds of millions of Indians who interact with technology through voice in Hindi, Tamil, Telugu, Marathi, and dozens of other languages. The token economy isn't just an English-speaking developer economy.
Billing infrastructure matters. If your product charges per-use, carrier billing integration becomes a distribution advantage. The user who won't enter a credit card for a $20/month subscription might happily add ₹99 to their phone bill.
The cycle repeats
Every major infrastructure shift follows the same pattern. First, the resource is scarce and expensive (mainframe compute, long-distance minutes, mobile data, AI tokens). Then someone figures out how to make it cheap and accessible at scale. Then usage explodes in ways nobody predicted.
In 2016, nobody predicted that cheap mobile data would create a $200 billion Indian digital economy. UPI, Swiggy, Zepto, PhonePe — none of these exist without Jio making data nearly free.
In 2026, we're at the beginning of the same curve for AI tokens. China's telecoms are the first movers. Jio is making the biggest infrastructure bet. The per-token cost is in freefall.
The AI token is the new currency. The question isn't whether telecoms will sell it — they already are. The question is what gets built on top when a billion people can access intelligence as easily as they access the internet.
