विमर्श Vimarsh · Arthanīti

The AI Wealth Ladder: Six Rungs, and Access Is Not One of Them

AI will not split the world into those who have it and those who do not. Everyone will have it. It will sort us by three things: the brains to make it deliver, the money to deploy it and control of the models themselves. Here is the ladder, the evidence, where I may be wrong, a second look at my own idea, where India stands and two dated wagers.

Nikhil Sharma · अर्थनीति Arthanīti · 19 November 2026 · 12 min read

Photograph: Pille Kirsi · Pexels

I run businesses that teach AI skills and back founders with capital and method, so read this knowing I want you to climb. I argue for the cause, not for any side. Every figure is sourced at the end.

I made a prediction this morning. In one to three years AI will not divide the world into those who have it and those who do not. Everyone will have it. It will divide us by what we can do with it and by who owns it.

Two things will decide where a person lands. The first is brains: the ability to make AI deliver a result someone will pay for. The second is money: the capital to put AI to work at scale. A third thing sits above both. Control of the models, the chips and the clouds that every answer runs through.

Put those together and you get a ladder. It has six rungs. Access is not one of them.

सीढ़ी · The ladder

  1. Rung 5: The ultra rich. Those who own or control the models, the chips and the clouds. What AI does for them: Collect a share of every answer anyone else gets.
  2. Rung 4: The affluent and HNI. Large capital that hires the best brains to run AI for it. What AI does for them: Rent capability and keep the upside.
  3. Rung 3: The new upper middle class. Brains and money together: the founder or professional who can deploy AI on their own terms. What AI does for them: Turn AI into owned outcomes: a firm, a product, a practice.
  4. Rung 2: The neo middle class. Brains without money: people who can make AI deliver but own nothing it produces. What AI does for them: Earn a skill premium on a salary or a fee.
  5. Rung 1: Access without prosperity. Money without brains: people who can buy AI but cannot turn it into results. What AI does for them: Little. Subscriptions do not compound.
  6. Rung 0: Left behind. Neither brains nor money for AI: most of India today. What AI does for them: Nothing yet, and the entry-level jobs it replaces may be theirs.

Rung 0: left behind. Neither the skill nor the money. This is most of India today and much of the world. AI does nothing for them yet. Worse, the first jobs it takes are the entry-level jobs they were hoping to get.

Rung 1: access without prosperity. Money without brains. They can buy every subscription and every tool and still turn it into very little. A subscription is a cost. It compounds only in the hands of someone who knows what to ask.

Rung 2: the neo middle class. Brains without money. They make AI deliver every week and they are paid well for it. But they are paid a wage or a fee. They own none of what their work produces. This is the new middle class: skilled, in demand and still renting their future.

Rung 3: the new upper middle class. Brains and money together. The founder, the professional with a practice, the small firm that can deploy AI on its own terms and keep the outcome. This is where AI turns a person into a company.

Rung 4: the affluent and HNI. Large capital that hires the best brains to run AI for it. They do not need to understand the tools. They rent the people who do and keep the upside. The rich stay rich and get leverage.

Rung 5: the ultra rich. Those who own or control the models, the chips and the clouds. Every answer anyone else gets passes through them. They collect a share of all of it.

One thing about the order. Rung 1 sits below rung 2 even though rung 1 has more money. The ladder ranks what AI adds to a life, not what a person starts with. Money that cannot use AI stands still while skill that can use it moves.

जाँच · Which rung are you on?

Three questions. Be honest with the first one. Using AI is not the same as making it deliver.

Which rung are you on? Three questions: can you make AI deliver paid results; how much money and time can you put behind it; who runs and owns the AI. Answer them on the page at nikhilsharma.com to see your rung and the next step up.

प्रमाण · The evidence

This is a forecast, so it should be tested against what is already visible. Four things are.

Skill already pays. PwC’s 2026 AI Jobs Barometer found jobs that ask for AI skills pay a 62 per cent premium, up from 57 per cent a year earlier. That is rung 2 being priced. It also found something sharper. Sectors most exposed to AI grew productivity by 34 per cent since 2018 against 24 per cent for the least exposed. The top fifth of the most exposed companies grew it by 163 per cent. Skill inside a firm that owns its output is rung 3, and it is pulling away.

Capability is getting cheap. Stanford’s AI Index found that the price of AI performing at the level of GPT-3.5 fell more than 280 times between November 2022 and October 2024. Cheap capability is why rung 0 to rung 2 is the easiest climb in history. You no longer need money to learn. You need time and intent.

Capital is concentrating. Companies put $581.7 billion into AI in 2025, up 130 per cent in a year. American private investment was 23 times China’s. Meta alone guided to as much as $145 billion of capital spending for 2026. Rung 5 is being built with numbers no nation outside two can match.

The frontier is pulling away from what anyone can download. In August 2024 the best open model trailed the best closed model by half a per cent on the Arena leaderboard. By March 2026 the gap was 3.4 per cent. Open models are excellent and getting better. But the very top is not getting more open. It is getting more owned.

  • 62%: Wage premium for jobs that ask for AI skills, up from 57% a year earlier (PwC Global AI Jobs Barometer, June 2026)
  • 163%: Productivity growth since 2018 at the top fifth of the most AI-exposed companies; most exposed sectors grew 34%, least exposed 24% (PwC, June 2026)
  • 280×: Fall in the price of GPT-3.5-level AI between November 2022 and October 2024 (Stanford AI Index 2025)
  • $581.7 bn: Corporate investment in AI in 2025, up 130% in a year (Stanford AI Index 2026)
  • 23×: US private AI investment compared with China’s in 2025 ($285.9 bn against $12.4 bn) (Stanford AI Index 2026)
  • 0.5% → 3.4%: Lead of the best closed model over the best open one, August 2024 to March 2026 (Stanford AI Index 2026)

Put a price on rung 5 and it looks like this. Seven listed companies build frontier models or the chips and clouds they run on. On 19 November 2026 they were worth about $18.8 trillion between them. That is nearly 13 per cent of the value of every listed company in the world, held by seven names. Not all of that is AI. Much of it is cloud, advertising, retail and software. But it is the market’s own price on who it thinks will own the next decade.

  • Microsoft: $3.76 trillion
  • Nvidia: $3.75 trillion
  • Alphabet: $3.41 trillion
  • Amazon: $2.79 trillion
  • TSMC: $1.86 trillion
  • Broadcom: $1.63 trillion
  • Meta: $1.58 trillion

Together 12.9% of all listed companies. Source: companiesmarketcap.com, read 19 November 2026: the seven companies together $18.8 trillion; all 11,396 listed companies tracked $145.9 trillion. Values move daily.

विरोध · Where I may be wrong

A ladder this neat deserves pushback. Here is the strongest I can make against it.

“AI is getting so cheap that money will not matter.” Partly true, and it is good news. Falling prices compress the bottom of the ladder. They make rung 2 reachable for anyone with a phone and the will to learn. But cheap tools do not make cheap businesses. Turning skill into owned output still needs capital, customers and time. Prices falling 280 times did not make 280 times as many founders.

“Open models will break rung 5.” They will discipline it. They will not break it. The gap between the best open and the best closed model widened over the last eighteen months. And control is not only the model. It is the chips, the power, the data centres and the distribution. Those are getting more concentrated, not less.

“This is just capitalism with a new name.” The shape is old. The speed is new. Capital has always hired skill. What is different is that skill can now replace a team and capital can now replace the skill. Both rungs 2 and 4 are moving faster than any ladder before them.

“The young will climb fastest.” I wish it were so. The 2026 AI Index found employment of software developers aged 22 to 25 had fallen nearly 20 per cent since 2024. The first rung AI removes is often the rung people used to learn on. Climbing from rung 0 now needs a deliberate path. It will not happen on the job, because the job may not be there.

“Inequality is not inevitable.” Agreed, and that is the point of naming it. The IMF warned in 2024 that AI will likely worsen inequality in most scenarios unless policy leans the other way. A ladder can be climbed. It can also be pulled up behind those already at the top. Which one happens is a choice.

पुनर्विचार · A second look

I asked for the strongest case against my own ladder before publishing it. Some of it lands. It changes how the ladder should be read, so here it is.

Rung 1 has an exit I left out. Money without skill can buy rung 5 on the stock market. Anyone holding an index fund owns a slice of the companies at the top. If AI profits concentrate the way I expect, passive capital will collect them without learning a single tool. So the money-without-brains rung splits in two. Money spent on subscriptions stands still. Money invested in owners rides the ladder without climbing it.

Which means the real axis is ownership, not money. Look at every rung again and the same line runs through it. Rung 2 rents out its skill. Rung 1 rents tools. Rung 3 owns its output. Rung 4 owns the people who run AI for it. Rung 5 owns the rails. The question is not how much you have. It is whether you own a share of what AI produces or are paid once for helping it produce.

The question is not how much you have. It is whether you own a share of what AI produces.

AI skill alone will not keep its premium. The tools are designed to get easier, and every easier tool turns yesterday’s skill into today’s default. PwC already splits the market into roles AI professionalises and roles it democratises. What lasts is what AI cannot copy quickly: deep knowledge of a field, judgement, trust and above all distribution. Whoever owns the customer outlasts whoever owns the prompt.

Rung 5 may not keep all the value. Railways, airlines and the fibre boom changed the world and still destroyed a great deal of the capital that built them. Prices of intelligence are falling fast and open models force price wars. Value may slide down to chips and power, or up to whoever owns distribution. My wager bets the top keeps it. History says that is not guaranteed.

India as a nation is on rung 2. This is the point I most want readers to take away. India’s great export of the last thirty years has been skill sold by the hour. It is excellent, in demand and owns little of what it builds. If AI makes skill cheaper, a rung-2 nation feels it first. The way up is the same for a country as for a person: own the output. Products, platforms and models, not only services.

The state is a player, not a spectator. Competition law, taxes on AI rents, public compute and public platforms can all reshape the top of the ladder. India has done it before. UPI went around the private gatekeepers of payments. The ladder describes the market. Policy decides how steep it stays.

And one to three years is the time for gaps to open, not for classes to settle. Classes take decades. What I expect to see by 2028 is the spread: skill premiums holding, small AI-native firms earning outsized returns, value pooling at the top. If those do not show, the ladder was a story, not a forecast.

What would prove me wrong? If the AI-skill premium falls below 20 per cent, skill has become a commodity and rung 2 has merged into rung 0. If the top seven fall below their 2026 share of the market, rung 5 is not where the money pooled. I have put both to a date below.

भारत · Where India stands

Most Indians are on rung 0 today. The IMF counts about 40 per cent of jobs in emerging economies as exposed to AI, against about 60 per cent in rich ones. Lower exposure sounds like safety. It is also distance from the gains.

As a nation India is not on rung 5. The money that builds the frontier is overwhelmingly American, with China a distant second. But India has started. The IndiaAI Mission was approved in March 2024 with about ₹10,000 crore. In February 2026 Sarvam released two Indian models of 30 billion and 105 billion parameters, and BharatGen released Param2 at 17 billion.

My view has not changed. India should build sovereign models and compute, because a country that only rents intelligence will pay rent on its own data forever. And India should do what it did with UPI: apply the best AI at population scale on rails it owns. That is how a country of rung-0 workers climbs together. Not one course at a time but one public platform at a time.

A country that only rents intelligence will pay rent on its own data forever.

चढ़ाई · How to climb

From 0 to 2. Pick one task someone already pays for: a report, a design, a translation, a sales list. Learn to deliver it with free AI tools, better and faster than before. Do it every week. That is the whole climb. It costs time, not money.

Every rung is tied by hand. The climb from nothing to a skill someone pays for now costs only time. Photograph: Khimish Sharma · Pexels

From 1 to 3. If you have money and not the skill, do not buy more tools. Learn, or partner with a rung-2 person and pay them in equity as well as fees. Skill that owns a share stays. Skill that is only hired leaves.

From 2 to 3. Own something. AI has made a small business the cheapest it has ever been to start. A share of the output beats a wage for producing it. This is the most important climb on the ladder, and the one I spend most of my working life on.

From 3 to 4. Hire other brains and let AI multiply them. Scale is the next rung, and it is a management problem more than a technology one.

To 5. Few people will get there. Nations can. That is why the sovereign model question is not a technology debate. It is a question of which rung a billion people will stand on.

दाँव · The wager

A forecast about concentration should carry a number. Here are two: one for the top of the ladder and one for the middle.

By 31 December 2028 the seven listed companies that build frontier AI models or the chips and clouds they run on (Nvidia, Microsoft, Alphabet, Amazon, Meta, Broadcom and TSMC) will together be worth more than 15 per cent of all listed companies in the world. On 19 November 2026 they were worth about $18.8 trillion of $145.9 trillion, or 12.9 per cent. The private model makers are left out, so the bet is conservative.

The scorer is companiesmarketcap.com on 31 December 2028, the same source as the baseline.

And the middle: PwC’s Global AI Jobs Barometer published in 2028 will still report a wage premium of at least 40 per cent for jobs that ask for AI skills. It was 62 per cent in June 2026. If it falls below 40 the premium is fading. If it falls below 20, skill has become a commodity and the neo middle class has merged into the rung below it.

Two markers on the way, both public. 1. The lead of the best closed model over the best open one stays above 2 per cent on the Arena leaderboard as reported in the Stanford AI Index. 2. An Indian-built foundation model enters the global top twenty on the same leaderboard. That second one is the marker I most want to be right about.

My call, first published 19 November 2026. If the seven are worth less than 15 per cent of the world at the end of 2028, the top of the ladder flattened faster than I think. If the premium falls below 40, the middle did. Either way this essay will say so.


Access is not prosperity. Capability with capital is. And control of the model is power.

Nikhil Sharma. Gurugram, 19 November 2026.


Sources, checked 19 November 2026. PwC, 2026 Global AI Jobs Barometer, 15 June 2026. Stanford HAI, AI Index 2025: State of AI in 10 Charts, 7 April 2025, for the 280-fold fall in inference cost. Stanford HAI, AI Index 2026, Economy chapter, for 2025 investment, US and China, and early-career developer employment. Stanford HAI, AI Index 2026, Technical Performance chapter, for the open and closed model gap. 24/7 Wall St. via Yahoo Finance, hyperscalers’ 2026 AI spending, 1 May 2026, for Meta’s capital spending guidance. companiesmarketcap.com, read 19 November 2026, for market values. IMF, AI will transform the global economy, 14 January 2024. PwC’s report also describes the two tracks of professionalised and democratised roles. Angel One, IndiaAI Mission model releases, 19 February 2026.