Arshad Dokadiya

🇮🇳 28 | Lawyer | Techie | AI | Stock Market | CoFounder & Solo VibeCoder @crazymoneyai

[Arshad Dokadiya @arshaddokadiya]

A US AI startup just moved 100% of its business away from an American AI model to a Chinese one.

The reason? It’ll save them millions of dollars.

And they are not the only ones. Here’s the quiet shift happening in AI right now 👇

The data that shocked Silicon Valley 📊

OpenRouter is a neutral platform where developers route their work across 400+ AI models from 60+ providers. It shows what companies actually use in production — not marketing claims.

Here’s what it revealed:

In June 2025, US models — Google, OpenAI, Anthropic — held around 70% of all token usage on the platform.

By June 2026, that share collapsed to roughly 30%.

Chinese models now process the majority.

Let that sink in. In twelve months, American AI’s dominance on this platform was cut in half 📉

CNBC confirmed it with harder numbers.

The share of tokens US companies route to Chinese models has stayed above 30% every single week since February 8, 2026 — peaking at 46%.

The average over the previous 12 months? Just 11%.

In the first half of 2025 it was 4.5%.

This isn’t a trend. It’s an avalanche.

The one word driving all of it: price 💵

Here’s the brutal economics.

DeepSeek’s V4 Flash model costs $0.09 per million input tokens.

OpenAI’s GPT-5.5 and Anthropic’s Claude Opus 4.8? $5 per million.

That’s a gap of roughly 55 to 1.

Chinese open models run 60% to 90% cheaper than the leading American models — for performance that’s now “good enough” for most tasks.

The real-world examples 🏢

→ AI startup Lindy moved 100% of its traffic from Claude to DeepSeek — expecting to save millions.

→ https://t.co/dE22VzYygS’s GLM-5.2, released in June, became the fastest-growing model on Vercel this year — daily usage grew 27x in its first week.

→ Uber reportedly burned its entire annual AI budget in just 4 months as American model costs exploded.

The names leading China’s charge 🇨🇳

DeepSeek — now the single largest provider on OpenRouter at 16.3% of all traffic, ahead of Google, OpenAI, and Anthropic.

Alibaba’s Qwen — has crossed 1 billion downloads, overtaking Meta’s Llama as the world’s most-downloaded open model.

https://t.co/dE22VzYygS’s GLM-5.2, Moonshot’s Kimi, MiniMax — six Chinese models now rank above Claude by usage.

Why this is bigger than a price war 💡

The American AI business model was built on one assumption: our models are so much better that companies will pay premium prices per token.

The OpenRouter data breaks that assumption.

A huge chunk of AI demand is now behaving like a commodity — routed to whichever model is cheapest and good enough.

Palantir’s CEO called the token-based pricing model “fundamentally broken” on CNBC.

The uncomfortable truth for US AI labs 🌍

China figured out something powerful.

They couldn’t win by building the single best model. So they built cheap, open, capable models — and gave them away for anyone to self-host.

Now every price-sensitive workload on earth quietly flows toward them.

The US still leads at the frontier — the hardest, highest-stakes reasoning.

But the everyday, high-volume, real-money work of AI? That’s increasingly running on Chinese models.

The most important chart in AI right now isn’t a benchmark score.

It’s a usage curve — and it’s pointing east 🔥

[Arshad Dokadiya @arshaddokadiya]

39 million Indians hold crypto. The RBI just told the government to ban all of it.

This isn’t a rumor. It’s in official documents reviewed by Reuters today.

Here’s what’s actually happening — and what it means for your money 👇

The news broke July 8, 2026 📌

Reuters reviewed government documents showing the Reserve Bank of India has formally recommended that private cryptocurrencies be prohibited.

Not regulated. Not taxed. Prohibited.

RBI Deputy Governor Rohit Jain laid this out before the Parliamentary Standing Committee on Finance on July 2.

The Finance Ministry confirmed it: “RBI has recommended that crypto assets should be prohibited.”

The RBI’s core argument 🏦

The RBI believes regulating crypto would legitimize it.

Their preferred strategy is “containment” — keeping crypto completely outside the banking system rather than bringing it under rules.

They want banks barred from holding, trading, or having any exposure to crypto.

The deeper fear: monetary sovereignty 💵

Here’s what actually drives this.

The RBI is most afraid of stablecoins — crypto tokens pegged to currencies like the US dollar.

Their logic: if millions of Indians hold dollar-backed stablecoins, it weakens the RBI’s control over the rupee and its monetary policy.

Even rupee-backed stablecoins worry them — they say it could reduce government revenue and create instability in a crisis.

The tax angle makes it worse 📊

In the year ending March 2023, fewer than 25% of the 645,000 people who traded crypto declared it on their taxes.

Three out of four crypto traders went completely undetected.

The CBDT has already identified ₹888 crore in undisclosed crypto income and sent notices to over 44,000 taxpayers.

The Enforcement Directorate uncovered ₹2,500 crore in unauthorized cross-border crypto transactions.

But here’s the catch 🤔

India has NOT actually banned crypto.

The RBI has recommended it. The government has not committed to it.

This exact fight has gone on for years:

→ RBI banned crypto banking in 2018
→ Supreme Court struck it down in 2020
→ A crypto policy paper has been shelved at least 5 times since 2025 — largely due to RBI’s opposition

Crypto trading remains completely legal in India right now.

What IS changing — quietly 🔐

While the ban stays stuck in debate, the noose tightens through rules instead:

→ From April 1, 2026: exchanges face ₹200/day penalty for non-reporting
→ Crypto reclassified as financial assets under global tax-sharing frameworks
→ Exchanges must report all OTC transactions over $10,000
→ 30% flat tax on gains + 1% TDS per trade, no loss offset

The government found a smarter path than a ban: make crypto so heavily monitored and taxed that it controls it without ever outlawing it.

The bigger picture 🌍

The world is splitting into two approaches.

The EU built MiCA — a licensing framework. Singapore and UAE regulate crypto openly.

India’s RBI wants to fence it off entirely.

39 million Indians holding ₹20,000 crore are caught in the middle.

The RBI wants prohibition.
The market wants clarity.
The government wants tax revenue.

Until those three forces agree, every Indian crypto holder is investing in a legal grey zone that could shift with a single monsoon session of Parliament 🔥

Not financial advice — just where things stand today.

[Arshad Dokadiya @arshaddokadiya]

Dear @IRCTCofficial @RailwaySeva @RailMinIndia,

I did not forget my password. Your system expired it under the 180-day rule.

Now to reset it, I need an OTP — which is not arriving on my registered mobile number. Not late. Not delayed. Simply not coming.

So I am locked out of my own account by a security rule, and locked out of the fix by a broken OTP system.

Search “IRCTC OTP” on this platform and you will find thousands of Indians reporting the exact same thing in just the last few days.

This is Digital India’s flagship ticketing platform. A basic OTP — the simplest piece of technology in the entire system — is what’s failing.

If the login can’t work, the crores of daily bookings behind it can’t either.

Please fix the OTP delivery and rethink a 180-day expiry that forces users into a system that doesn’t deliver the OTP needed to recover.

@AshwiniVaishnaw 🙏

[Arshad Dokadiya @arshaddokadiya]

India’s banks quietly lost control of who borrows money in this country.

Fintech apps now originate nearly 90% of all small personal loans in India.

The bank branch didn’t get disrupted. It got bypassed entirely.

Here’s what the data shows 👇

The shift in six years 📊

In FY18, fintechs disbursed 50% of all loans under ₹1 lakh.

By FY26, fintech NBFCs account for nearly 90% of all personal loan originations under ₹1 lakh.

Banks’ share of this segment has collapsed to around 10%.

The smallest, fastest loans in India are now almost entirely digital.

The scale of it 📈

In just the first half of FY26, fintechs originated 6.4 million personal loan accounts — up from 5.9 million a year earlier.

The value jumped to ₹97,381 crore from ₹78,084 crore.

The average loan size? ₹15,177.

These are small loans. Millions of them. Moving through phones, not branches.

Why banks walked away 🏦

This is the part most people miss.

Banks didn’t lose this market. They chose to leave it.

A ₹15,000 loan costs a bank almost the same to process as a ₹15 lakh loan — but earns a fraction of the return.

So banks moved upmarket, chasing high-value loans for prime borrowers.

Fintechs moved into the gap they left behind.

Who is actually borrowing 👥

Over 65% of fintech NBFC borrowers are under the age of 35.

Nearly 43% of these loans are under ₹50,000.

The growth is coming from Tier 3 cities and younger borrowers — exactly the segment traditional banks historically ignored.

For millions of young Indians, their first-ever formal loan is coming from an app, not a bank.

The risk nobody is pricing in ⚠️

Here’s what makes this fragile.

For fintech lenders, unsecured loans make up roughly 70% of their entire loan book.

Gross NPAs on unsecured retail loans sit at 1.7% — more than double the 0.7% on secured loans.

The RBI is watching this segment closely. Unsecured lending growth at this pace, concentrated in young and first-time borrowers, is exactly the kind of thing that looks fine until it suddenly doesn’t.

The bigger picture 💡

This is one of the fastest financial power shifts in India’s history — and it happened almost silently.

No bank collapsed. No dramatic headline.

Just a slow, complete handover of small-ticket lending from century-old institutions to apps that are barely a decade old.

The winners of the next decade of Indian finance won’t be the banks with the most branches.

They’ll be the platforms with the best data on who actually pays back 🔥

[Arshad Dokadiya @arshaddokadiya]

The government just dropped a 10-point "myth vs fact" on E20 petrol.

But here's what they're NOT telling you: even their OWN data admits your mileage drops.

Let me break down the full story 🧵

On July 3, the Ministry of Petroleum released a point-by-point rebuttal to the viral panic around E20 (petrol blended with 20% ethanol). It came alongside a report titled "Ethanol Blending Programme in India: A Journey."

The context matters. This wasn't random. A LocalCircles survey of 22,000+ people found 66% of pre-2023 petrol vehicle owners reporting mileage drops over 10%. Kejriwal demanded a rollback. Priyank Kharge called it "an experiment on 3.6 crore Indians." A protest hit Jantar Mantar. The government had to respond.

So what are the 10 points?

1. "1 litre ethanol wastes 10,000 litres of water" → False. That number wrongly blames ethanol for paddy's entire water footprint. Modern distilleries use just 3-5 litres per litre under Zero Liquid Discharge systems. Maize now supplies 40%+ of ethanol.

2. "E20 is a risky experiment" → The US runs E10/E15, Brazil mandates E27, Canada/Thailand/Japan use blends. It's globally tested.

3. "E20 drastically cuts mileage" → ARAI trials over 40,000 km in cars found "only marginal changes." Ethanol's higher octane (108 vs petrol's 84) can improve acceleration.

4. "It melts engine parts" → No adverse effect found on metal/plastic. Cuts CO emissions ~30% in cars, ~50% in two-wheelers.

5. "It voids warranty & insurance" → SIAM and insurers confirm both stay valid for E20-approved vehicles.

6. "Sugar in ethanol attracts ants" → Distillation removes all sugar. Denaturants repel insects.

7. "Govt called it an experiment in Supreme Court" → The AG's office clarified the case was only about ethanol procurement contracts, not the fuel.

8. "Water forms inside the tank" → Modern vehicles have seals preventing water ingress.

9. "Videos show petrol & sugarcane juice separating" → Fabricated. Fuel ethanol is made industrially, not by mixing raw juice.

10. "Ethanol plants deplete groundwater" → They need statutory clearances and ZLD compliance. Programme saved ₹1.9 lakh crore in forex, paid farmers ₹1.6 lakh crore.

Now here's the honest part the headlines skip.

The government's OWN NITI Aayog roadmap admits a 6-7% efficiency loss for older four-wheelers designed for E0, 3-4% for two-wheelers, and 1-2% for newer E20 cars.

Minister Hardeep Puri himself conceded: "Mileage, yes, it may drop a little."

And roughly 80% of petrol vehicles sold in the last 15 years (~300 million) were built for E5/E10, not E20. Maruti is now preparing retrofit kits priced ₹4,000-6,000.

The real grievance? E20 isn't sold cheaper despite lower energy content. NITI Aayog recommended pricing higher blends lower back in 2021. That was never implemented.

So where does this leave you?

→ Post-2023 car: relax, 1-2% loss, warranty intact.
→ Pre-2023 car: expect a real 3-7% dip. Maintain well. Consider a retrofit kit if you're keeping it long-term.

Bottom line: E20 is broadly safe. The engine-destruction panic is overblown. But the mileage penalty is real, older vehicles bear it, and you get no cheaper fuel and no choice at the pump.

Safe ≠ free. Both things are true.

[Arshad Dokadiya @arshaddokadiya]

China just declared war on American AI.

Not with weapons.

With $295 billion and a plan that will make the next decade irrelevant for the US.

Here’s what just happened — and why it should scare every person who understands what AI is about to do to the world 👇

On June 9, 2026, Bloomberg broke a story that shook every AI company in Silicon Valley.

China’s National Development and Reform Commission is drafting a blueprint to spend 2 trillion yuan — $295 billion — building a nationwide network of AI data centers over the next five years.

This is not a startup funding round.
This is not a tech company’s quarterly capex.
This is not a government subsidy program.

This is an entire nation deciding that AI is the most important technology in human history — and backing that belief with sovereign debt, state companies, and a five year plan that cannot be cancelled by a board meeting or a bad earnings call.

Here is exactly what the plan looks like 🏗️

State telecom giants China Mobile and China Telecom will build and operate hundreds of interconnected data centers across the entire country.

The goal is to connect every single one into a unified national computing grid by 2028.

One network. One country. One purpose.

Healthcare powered by AI. Transportation managed by AI. City infrastructure run by AI. Banking monitored by AI. Education delivered by AI.

A nation of 1.4 billion people running on a single AI grid built, owned, and controlled by the government.

And here is the detail that crashed American chip stocks the same day 😰

The plan mandates that at least 80% of ALL technology — including every AI chip — must come from domestic Chinese suppliers.

Not 50%. Not 60%. 80%.

Nvidia? Locked out by law.
AMD? Locked out by law.
Every US semiconductor company? Legislated out of the largest computing procurement in history.

The day Bloomberg published this:
Nvidia stock fell 2.4%
AMD stock fell 4%

In a single news report. In a single day.

But this isn’t a future threat. It is already happening 📉

Nvidia reported ZERO Data Center Hopper chip shipments to China in the quarter ending April 2026.

Zero.

Not a decline. Not a slowdown. Zero.

The same quarter one year earlier — $4.6 billion in China AI chip sales.

$4.6 billion to zero in twelve months.

China’s AI chip market has been legislated completely out of existence for American companies 🚫

So who fills the void?

Huawei 🇨🇳

Huawei’s Ascend AI chips — built entirely on their own proprietary architecture with zero dependence on American technology — are already running at full scale inside Baidu and ByteDance’s data centers.

In May 2026, the Chinese government approved 9 categories of domestic AI chips for all government and security sectors.

The approved suppliers: Huawei. Alibaba. Shanghai Biren Technology. Moore Threads.

Huawei alone shipped 812,000 Ascend chips in 2025 and is projecting $12 billion in AI processor revenue for 2026 — a 60% jump from last year.

When the government guarantees 80% of a $295 billion buildout goes to domestic vendors — those vendors get a captive market no private company could ever create 🏭

Now the number everyone is getting wrong 🔢

The $295 billion is ONLY the data center infrastructure.

It does not include:
→ Alibaba’s AI spending
→ Tencent’s AI spending
→ Baidu’s AI spending
→ ByteDance’s AI spending
→ Power grid upgrades

When power grid costs are included — analysts estimate the full requirement exceeds 5 TRILLION yuan.

China’s power demand from AI data centers alone will rise 300 to 500 billion kilowatt-hours by 2030.

That is one-fifth of China’s ENTIRE national electricity demand growth.

For AI alone 💡

Here is the comparison that changes everything 🤔

US companies — Meta, Microsoft, Google — are spending $725 billion on AI in 2026 alone.

China is spending $295 billion over five years.

Sounds like America is winning?

Chinese data centers cost a fraction of American ones — cheaper labor, cheaper materials, cheaper construction, full government subsidies.

$295 billion in China builds far more compute than $295 billion in the US.

But the deeper difference is not about money.

It is about the nature of the commitment.

A private company can slow down or cancel AI spending if the stock drops or the board panics.

A government plan backed by national security does not get cancelled by a board meeting.

It does not slow down because of a bad quarter.

It accelerates regardless of market conditions 💨

Now here is what nobody is saying clearly enough 🔐

China’s National Intelligence Law — Article 7 — requires every organization operating in China to cooperate with government intelligence requests on demand.

China Mobile. China Telecom. Huawei. Alibaba.

Every company building this grid is legally required to hand over any data the government requests.

No warrant needed.
No court order needed.
No notification to users.

Every piece of data on China’s national AI grid is accessible to the Chinese state the moment they want it.

The world’s largest AI infrastructure is simultaneously the world’s largest surveillance system.

Built with government money.
Operated by state companies.
Running on government chips.
Reporting to Beijing by law.

This is not a side effect of the plan. This is a feature 📌

The market understood immediately 📊

GDS Holdings rose 12% in pre-market trading the day Bloomberg published.
Vnet Group climbed 17%.
Huawei’s position became unassailable overnight.

Here is the bigger picture 🌍

The AI race in 2026 is not a competition between companies.

It is a race to build the infrastructure that will run the world’s economies, governments, and militaries for the next 50 years.

Whoever builds the most compute wins.

Not just the AI war. The entire next century.

America is betting on private companies chasing profit.
China is betting on state power chasing dominance.

One stops if profits disappoint.
The other doesn’t 🤷

And the question every Indian should sit with 🇮🇳

India has no $295 billion AI plan.
India has no national compute grid.
India has no domestic AI chip manufacturer.

The two most powerful countries in the world are building infrastructure that will define who controls global technology by 2035.

The countries that build the compute own the future.
The countries that don’t will rent their intelligence from whoever did.

Which side India ends up on is being decided right now.

In budgets that don’t mention AI.
In policies that haven’t been written.
In investments that haven’t been announced.

The question is not whether India can build this.

The question is whether anyone in power understands that the clock is already running 🔥

[Arshad Dokadiya @arshaddokadiya]

OpenAI just filed to go public at $852 billion.

The CEO owns zero shares of his own company.

And they’re losing money on every single dollar they make.

Here’s the IPO story Wall Street doesn’t want you to read 👇

The headline number everyone is celebrating: OpenAI is targeting a $1 trillion valuation at IPO. Goldman Sachs, Morgan Stanley, and JPMorgan are running the deal. 800 million people use ChatGPT every week. Revenue went from $6 billion in 2024 to $25 billion annualized today.

Sounds incredible. Here’s what’s underneath it 📉

OpenAI lost $1.22 for every single dollar it earned in Q1 2026.

That is not a typo.

They are projected to lose $14 billion in 2026 alone. They won’t be profitable until 2029 or 2030 at the earliest. And HSBC analysts say OpenAI needs $207 billion MORE in funding between now and 2030 just to keep operating — even accounting for all the revenue growth.

The company committed $1.4 trillion in infrastructure spending over the next 8 years.

They currently earn $25 billion a year.

The math doesn’t work 🧮

Now here’s where it gets genuinely strange.

Sam Altman — the man who built OpenAI, runs OpenAI, and represents OpenAI everywhere — owns zero equity in the company.

Not a small amount. Zero.

The leaked cap table shows his position listed as “None/Pending.”

Every major tech IPO in history has had a CEO with significant ownership. When your CEO doesn’t own the company he’s leading into the world’s most high-profile listing, that is a risk factor unlike anything Wall Street has ever seen.

Who DOES own it? 🏦

Microsoft: 27%
SoftBank: $34.6 billion invested
Amazon: $50 billion committed
Nvidia: signaled up to $100 billion

And here’s the circular money problem analysts are quietly terrified about.

Nvidia invests in OpenAI → OpenAI pays Nvidia for chips → Nvidia’s revenue goes up → Nvidia’s stock goes up → Nvidia invests more in OpenAI.

Critics are calling this the most dangerous money loop in tech history. Multiple companies inflating each other’s valuations without real end-user cash flow underneath.

But wait — there’s more 🔥

While everyone focuses on OpenAI’s IPO, their biggest competitor just overtook them.

Anthropic — the company behind Claude — is now valued at $965 billion privately.

That’s MORE than OpenAI’s $852 billion.

Anthropic filed for its OWN IPO on June 1, 2026 — one week BEFORE OpenAI.

Anthropic is now profitable. OpenAI isn’t.

ChatGPT’s market share dropped from 86.7% to 64.5% in 12 months.

Google Gemini’s share went from 5.7% to 21.5%.

OpenAI is losing the AI war while preparing to ask the public for $1 trillion 📌

And then Bloomberg dropped this 2 days ago.

OpenAI is now considering delaying the IPO to 2027.

The same day that news broke, SoftBank’s stock crashed — its worst day since August 2024.

SoftBank invested $34.6 billion in OpenAI. They NEED this IPO to show returns to their own investors.

So here’s where we actually are right now.

The most hyped IPO in tech history is:

→ Losing money on every dollar it makes
→ Led by a CEO with no equity
→ Being overtaken by its main competitor
→ Potentially delaying to 2027
→ Dependent on $207 billion in future fundraising just to survive

The question is not whether OpenAI is an incredible company.

It is.

The question is whether it’s worth $1 trillion when it loses $14 billion a year, has no path to profit until 2030, and its CEO owns nothing.

Public markets are about to answer that question.

And the answer will define the entire AI industry’s future 🤷

[Arshad Dokadiya @arshaddokadiya]

Apple just raised prices on every MacBook and iPad YESTERDAY.

Their stock crashed 6% the same day.

Tim Cook said “price increases are unavoidable.”

Here’s the $270 story nobody is breaking down for you 👇

RAM prices have risen 98% in a single quarter in 2026. Not 10%. Not 20%. 98%.

Experts are officially calling it “RAMageddon.”

A 16GB chip that cost $6.84 just 9 months ago now costs over $27. That’s a 4x jump. And it’s still rising 📈

Why is this happening?

AI is consuming the world’s entire RAM supply.

Every ChatGPT response. Every AI image. Every data center Nvidia is building — they all need a special type of RAM called HBM (High Bandwidth Memory).

And here’s the brutal math 🧮

Making 1 bit of AI RAM requires 3x more factory space than making regular RAM for your phone. So every time Nvidia buys AI memory — 3x less RAM exists for you.

Your MacBook is now literally paying the price for ChatGPT.

Here’s what the world’s biggest CEOs are actually saying 🎙️

Tim Cook (Apple CEO): “The situation has become unsustainable. Price increases are unavoidable.”

Intel CEO Lip-Bu Tan: “There will be no relief until 2028.”

Elon Musk (Tesla CEO): “We’ve hit a chip wall. We either accept it or build our own factory.”

Dell COO Jefferey Clarke: “We’ve never seen costs escalate at this pace. I don’t see how this will not make its way to customers.”

Samsung CEO: “In 2026, semiconductor supply issues are going to affect everyone.”

Here’s exactly what Apple raised YESTERDAY 💻

→ MacBook Neo: $599 → $699
→ All MacBooks: price hike
→ All iPads: price hike
→ Apple TV: price hike
→ HomePod mini: +$30
→ Mac Studio: up to +$1,300

The one thing they DIDN’T raise yet? iPhones.

But TechInsights already warned — passing full RAM costs onto the next iPhone Pro could add $270 to the price 📱

Now here’s the part that should make you angry 🏭

Just 3 companies control 95% of all global RAM production.

Samsung. SK Hynix. Micron.

All 3 stopped prioritizing your phone and laptop to serve AI companies.

Micron literally shut down its entire consumer brand to focus only on AI data centers.

SK Hynix’s complete 2026 production was sold out before the year even began.

The RAM crisis timeline 📊

Q3 2025 → prices up 30%
Q4 2025 → prices up another 40-50%
Q1 2026 → prices up 80-98%
Q2 2026 → set to rise another 58-63%

Intel CEO confirmed: no relief until 2028.

New factories being built today won’t produce results until 2028.

Two more years of this.

And it’s not just Apple 🌍

Dell, HP, Lenovo, Acer, ASUS → 15-20% price hikes confirmed
Samsung Galaxy S26 → already 10% more expensive than S25
Xiaomi, OPPO, Vivo → cutting RAM from budget phones just to survive
Global smartphone sales → projected to fall 13% in 2026

Here’s the brutal truth nobody is saying out loud 💡

AI companies needed RAM.
They paid record prices for it.
Memory makers chased that profit.
Stopped supplying consumers.
Your MacBook costs more.
Your phone costs more.
And the companies doing this just posted record profits.

Micron’s revenue quadrupled in a single quarter.

You are funding the AI revolution whether you chose to or not.

Your next device will cost more.
Your current device is already worth less.
And it won’t stop until 2028.

Now you know 🤷