# Jio Doesn’t Need to Win AI to Make Money From It?

[Watch on tuber](https://tuber.guzus.xyz/video/-MMDaIY3jwg/summary) · [YouTube](https://youtu.be/-MMDaIY3jwg)

**Score: 60/100 · skim**

China's three state-owned telecom giants (China Mobile, China Telecom, China Unicom) have found a way out of the "dumb pipe" problem: they run AI data centers and sell AI tokens directly to consumers, much like data plans. Their revenue from this line of business is growing fast. Jayant Mundra argues that Reliance Jio and Airtel's large AI data center investments suggest India's telecoms may copy this model within about two years.

- **0:00-0:53** — **The "dumb pipe" problem.** For 20-25 years, telecom operators worldwide have complained that they pay for towers, underground cable and dealings with authorities, while apps built on top of their networks take the money. WhatsApp ate into call and SMS revenue, YouTube and Netflix use up most of people's data, and now OpenAI's models run on the same network. Operators argue that these services couldn't earn what they do without the network, so they deserve a share. The infrastructure owner who earns nothing from what flows through it is called a "dumb pipe."

- **1:00-1:50** — **China's recharge menus now sell AI tokens.** China's three biggest operators are all state-owned, and Mundra says they've started to solve the problem in a way the rest of the world will likely follow. When people in China recharge their phones, they can choose how many AI tokens to buy right next to how many GB of data they want. He describes tokens as "the currency of the AI world": you buy them, then spend them whenever you use an AI model.

- **1:50-2:45** — **What tokens represent and the telecom pitch.** He explains that a paid ChatGPT or Gemini subscription ($20 or $200 a month) is really an allotment of tokens. Those tokens stand for the compute and electricity used in a data center every time you send a prompt. Chinese telcos are building the AI data centers themselves, putting many models on one platform, and selling the tokens that give access to them. Customers can then use any model they like.

- **2:45-3:30** — **"Why not pay ChatGPT directly?" and China's actual offers.** He raises the obvious objection that buying tokens from a telco to use elsewhere is a hassle. His answer is mainly that it may work out a bit cheaper, which is a weak argument. The concrete offers:
  - **China Mobile:** about 400,000 tokens for 1 yuan (he says roughly ₹20) and access to more than 300 AI models from around the world.
  - **China Telecom:** around 142 models, with a trial pack of about ₹120 that lets you run each model at least once.

- **3:29-3:53** — **China Unicom bundles tokens with broadband.** Unicom has set up what he calls a "supermarket of tokens." It sells discounted AI token packages together with home broadband connections, so the AI allowance comes with the internet plan instead of being a separate purchase.

- **3:53-4:33** — **Scale, and why it matters for India.** He says daily token use in China rose from about 100 billion roughly two years ago to about 140 trillion in March of this year. He admits he couldn't find any newer figure. He then turns to India: Jio has about 500 million customers and Airtel about 360 million, plus Vodafone Idea and BSNL. Selling tokens to even 20% of them would be a large business, and these subscriber bases are the "rail" it could run on.

- **4:33-5:19** — **Telcos as the "cash register" of AI.** He asks how operators that don't build their own AI can earn from it, and why this didn't happen earlier. In his model, one company builds the AI model, another sells the tokens and another runs the data center. For example, if OpenAI wanted more reach in India, it would need a lot of data center capacity. It could build that itself or partner with a telco, which would host the compute and sell tokens that work on OpenAI's platform.

- **5:19-5:51** — **Revenue growth in China.** He cites three examples:
  - **China Telecom:** revenue from this business up about 96% in six months.
  - **China Mobile:** up about 130% in a year.
  - **China Unicom:** joined later, but its earnings are already rising quickly.

  He doesn't give the absolute size of these revenues, so the percentages may be growth from a small base.

- **5:51-6:56** — **Why now: DeepSeek and falling token costs.** This didn't work 20 years ago because the AI didn't exist, and it didn't work two years ago because running AI was too expensive. DeepSeek brought the cost per token far below American companies' prices and set off a race among Alibaba, Tencent, ByteDance, Moonshot and Zhipu (the transcript says "Zippo"), which build LLMs, video models and more. His argument for why demand grew so fast: cheaper tokens lead to much more use, and much more use creates a new economy that the telcos moved in to serve. He predicts India will see real activity within about two years.

- **6:56-7:51** — **Signs in India: Reliance and Airtel's data center spending.** Reliance has announced one of the largest AI data center complexes in India, Asia and possibly the world, in Jamnagar, and later a second one of similar size in Andhra Pradesh. He puts these at $10-15 billion each. Airtel already runs a data center business called Nxtra, is pushing into AI data centers and recently raised major funding. Because China's three operators are among the biggest in the world, he expects operators everywhere to watch the results and follow if the model makes money.

- **7:51-8:57** — **China is short of capacity and building in Inner Mongolia.** China Telecom has said publicly that it doesn't have enough capacity to meet direct token demand, so it is spending heavily on more data centers. For comparison, Elon Musk's Colossus has more than 200,000 Nvidia GPUs and is being expanded about fivefold, toward roughly a million. The Inner Mongolia project is reportedly expected to be about 1,000 times larger. He gives no source for that multiple, which seems implausible. The reasoning for the location is cheap, open land and low-cost solar and wind power, which lets tokens be priced among the cheapest in the world.

- **8:57-9:59** — **What Ambani and Mittal may be planning, then outro.** He answers the common question of where Reliance will find customers for such huge data centers. Hyperscalers like Meta and Microsoft will take some capacity, but on the China model a large share could go to selling tokens directly to Indian consumers. He claims no Indian media outlet has explained this before. The last 15 seconds or so are a call to like, subscribe and share (sign-off: Groww Edge, Jayant Mundra).

**Practical takeaways:**
- **For investors:** Treat Jio and Airtel's AI data center spending as a possible consumer token-retail business, not just capacity rented to hyperscalers. That could change how you value the telecom sector.
- **The thing to watch:** Look for AI token packs appearing in Jio or Airtel recharge menus, or bundled with broadband. That would be the clearest sign India is copying China.
- **Treat the numbers with caution:** The growth percentages have no absolute revenue figures, and the "1,000× Colossus" claim has no source. Check these before relying on them.
- **The weak point:** The case depends on telco tokens being cheaper or easier to use than paying an AI company directly. Whether OpenAI, Google and others agree to accept telco-sold tokens is unclear and not addressed in the video.
- **Big picture:** Falling token costs drive more AI use. Whoever has the cheapest power and data center capacity close to many customers is well placed to collect the money.

## Audience signal

- No high-signal audience themes available.