Yesterday’s press release OpenAI and NVIDIA announce strategic partnerships to deploy 10 gigawatts of NVIDIA systems, is a real eye-opener, with NIVIDIA intending to invest up to $100 billion in OpenAI.

Following on from OpenAI signing with Oracle to purchase $300 billion in computing power over 5 years, one of the largest cloud contracts in history, it lays bare the growth ambitions in GenAI.

In fact The Information (subscription required) estimates that OpenAI plans to spend a whopping $450 billion renting servers through 2030.

Some commentators have pointed out that all the stocks involved, NVIDIA, ORCL and OpenAI have jumped disproportionality on this money going from one to the other. NVIDIA gaining $170b after yesterday’s announcement, ORCL $250b higher then before its announcement and OpenAI, which is a privately held presumably also meaningfully higher.

Individually for each stock this makes sense, even if collectively it seems wrong.

However this correlation in stock gains for all three, holds not only on the way up, but should there be a hiccup in OpenAI’s growth, it will also hold on the way down, impacting all 3 disproportionality.

The key to all this, is OpenAI’s revenue growth.

Reuters reporting on July 30th, that OpenAI reached $12 billion annualised revenue in a little under 3 years and CNBC reporting that OpenAI is targeting $125 billion in annualised revenue by 2029, a 10X increase. So phenomenal numbers indeed.

The key question now where is this revenue going to come from?

B2C – Consumer

Let’s look first at the business to consumer market, so individuals using ChatGPT for personal uses.

It is estimated that ChatGPT has between 500 million to 700 million weekly active users, but while that is impressive, I expect only a small fraction of these subscribe to the $20 per month plan. For most the free plan is sufficient and there is no compelling reason to pay.

Still, lets do a comparison with the entertainment industry and Netflix, which has 300 million subscribers, the US standard plan is $18 per month and it’s 2024 revenue was $39 billion.

Do we think ChatGPT has a path to 300 million consumers individually subscribing at $20 per month?

I think that is unlikely. More significant is likely to be indirect revenue from consumers, so a business paying when a consumer purchases a product with the lead generated from ChatGPT.

Still B2C markets are unlikely to get OpenAI to $125 billion in 2029.

B2B – Business

The larger opportunity is with businesses licensing GenAI for their employees.

The goal being to achieve productivity gains, either revenue generating or cost reducing, which lead to P&L improvement.

The above CNBC article stated that OpenAI now has 3 million paying business customers, up from 2 million in February. Given that each of these businesses will have licenses for many employees, we can envisage very large revenues, e.g.

• 2,000,000 businesses with 5 users licenses each, paying $1,200 per annum, is $2.4 billion.
• 100,000 businesses, each paying $50,000 per annum is $5 billion
• 5,000 business, each paying $1 million is $5 billion
• 100 businesses each paying $50 million is $5 billion

So while the above is entirely hypothetical guesswork, you can envisage how the revenue accelerates into the tens of billions and beyond.

Productivity Gains

However that vast B2B revenue is contingent on productivity gains being realised. A recent MIT report, State of AI in Business 2025 highlighted that:

• Despite the $30-40 billion in enterprise investment into GenAI, 95% of organisations are getting zero return.
• Over 80% of organisations have explored or piloted GenAI and nearly 40% report deployment
• But just 5% of projects are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact
• Tools primarily enhance individual productivity, not P&L performance.

I won’t cover anymore of the report content, but it certainly is worth reading when you have a spare 10-15 mins to read the 26 pages.

Private Data

What I will note is that at a minimum business users need their organisations private data, both unstructured documents and structured data available to them in the context of their GenAI tool, as well as relevant industry or public documents and data.

While not sufficient by itself, it is critical to have this integration.

OpenAI, Anthropic and other GenAI vendors understand this, hence the increasing number of connectors and integrations available, many provided by incumbent business solution vendors or by new application layer start-ups.

On top of that, business users will need workflows that can lead to productivity gains with P&L improvements.

Workflows that they will either build themselves or license from specialist vendors.

If not, the stickiness of subscription revenue growth will take a downward hit, putting immense pressure on growth rates.

Success

While the jury may be out for many business use cases, at least for Software engineering, the numbers show huge uptake; Co-pilot and Cursor are both becoming ubiquitous in software engineering teams and contributing material revenue growth for GenAI firms.

We need to see that success for other use cases, functions and domains.

Be it in Business or Education, Health and Science.

To justify the capital spending in AI infrastructure.

The next two to three years will be interesting indeed.

An AI boom or AI bubble?

Only time will tell.