OpenAI’s $115 Billion Cash Burn Is Just the Beginning

Managing Editor’s Note: Fed decisions can have a big impact on the market. And the upcoming September meeting promises to be a standout.

And according to Larry Benedict, if you’re ready for it, you could have one of your best trading days of the year.

The opportunity that’s building has to do with President Trump… Jerome Powell… and a multitrillion-dollar market event.

To get all the details, just go here to automatically sign up for Larry’s Fed Decision Advance Warning event on Wednesday, September 10, at 8 p.m. ET.

Last week, OpenAI shocked the tech world with a stunning announcement.

Management now expects to burn through $115 billion in cash by 2029. That’s $80 billion more than earlier projections.

It’s no wonder CEO Sam Altman calls OpenAI the “most capital-intensive startup of all time.” Very few companies in history ever earn $115 billion, let alone plan to spend that much in just a few years.

Yet OpenAI is already drawing investors like SoftBank, which is buying shares at an incredible $500 billion valuation, a valuation $200 billion higher than that of March this year. And they are fully aware of Altman’s ambition to pour unprecedented sums into compute and data centers.

And here’s the kicker: $115 billion isn’t even the whole story.

More Than an OpenAI Story

OpenAI is already generating billions in revenue today, with the potential to reach hundreds of billions annually in the not-so-distant future.

But even with that kind of growth, costs are climbing faster. Between 2025 and 2030, the company projects $150 billion in inference costs alone. That’s just the price tag to run its models in production.

Training is another massive expense. This year, OpenAI will spend $9 billion training new models. Next year, that doubles to $19 billion. And costs will only accelerate as the company pushes from artificial general intelligence (AGI) toward the frontier of artificial superintelligence (ASI).

To rein in costs, OpenAI plans to build its own AI data centers instead of renting from Microsoft, Oracle, or Amazon. In the long run, this could save them from paying extra for hyperscalers to make their margin… but it will require an additional $100 billion upfront in capital spending.

Add it all together, and OpenAI is on track to spend over $400 billion on hardware and operating costs by the end of the decade. And if we take Altman at his word, that’s just the beginning. He’s already warned that OpenAI could “spend trillions of dollars on data center construction in the not very distant future.”

This isn’t just an OpenAI story. Across the globe, AI companies are ramping up spending at a historic pace. Demand for AI infrastructure is insatiable. And this buildout is shaping up to be the single largest investment cycle of our lifetimes.

Which brings us to the question every investor should be asking…

Where Will Value Accrue?

We believe the lion’s share of returns will flow to the hyperscalers – and the semiconductor companies that supply them.

Morgan Stanley recently echoed this view in a research report on “Agentic AI,” which mapped out the AI value chain. They shared the following graphic:

Source: Morgan Stanley Research

At the top of the pyramid sit the hyperscalers and infrastructure leaders. This is exactly where we’ve focused in both The Near Future Report and Exponential Tech Investor.

Below that are the model providers. Anthropic and OpenAI are the pure plays here, but they remain private and out of reach for most investors.

Still, the market is placing extraordinary bets. Anthropic, the company behind the Claude AI models, is reportedly raising $5 billion at a staggering $170 billion valuation. That’s a nearly 300x return from its first round in 2021, when it was valued at just $550 million.

And it’s clear where that capital is headed. It’s going into data centers filled with GPUs and other AI accelerators. Altman has said OpenAI expects to have more than one million GPUs online by year-end.

Altman has long complained about not being able to get access to enough GPUs to release their latest models to the public. To ease its perpetual compute shortage, OpenAI signed a $10 billion contract with Broadcom (AVGO) to design custom application-specific integrated circuits (ASICs).

The goal: secure more processing power, lower costs, and accelerate the release of ever more advanced AI models.

Broadcom is a perfect example of what this buildout means in practice. The company just posted a 63% year-over-year surge in AI semiconductor revenue.

And Wall Street is scrambling to keep up. Goldman Sachs just raised their forecast for Broadcom’s fiscal 2026 and 2027 revenue by over 35%. Broadcom’s already high revenue numbers will triple over the next two years.

Source: Company data, Goldman Sachs Global Investment Research

These are extraordinary numbers. And the data centers to house all these chips are going to be massive.

Now, to give you a sense of scale of these data centers, take a look at the image below of OpenAI’s Stargate 1 data center in Abilene, Texas. If you peer closely, you’ll notice tiny dots at the bottom. Those are cars in a parking lot, completely dwarfed by this enormous facility.

Stargate 1 Facility | Source: OpenAI

And Stargate 1 is already coming to life. In June, Oracle began shipping the first racks of NVIDIA’s cutting-edge GB200 GPUs directly to the site. OpenAI isn’t alone – other hyperscalers are also pouring capital into data centers at a historic rate.

Amazon is on track to spend $120 billion in capital expenditures this year. Microsoft will surpass $100 billion. Google and Meta will be close this year, and both are expected to blow past $100 billion next year.

Taken together, the Big Four will invest more than $400 billion this year, with projections pushing that total to $600 billion next year. To put that in context, Norway’s entire annual GDP is $504 billion. Four companies are spending more than the economic output of some developed nations.

And here’s the critical detail: roughly 40% of that spending is earmarked for specialized chips. The GPUs from NVIDIA and AMD, and custom ASICs from companies like Broadcom.

The Importance of the AI Buildout

McKinsey estimates global investment in AI data centers could reach $5.2 trillion by 2030 under baseline assumptions.

But with OpenAI boosting its spending projections by nearly 400%, Oracle spending 33% more than expected, and Amazon, Google, and Meta all raising capex by double digits, the baseline scenario made in April is already obsolete.

The reality is that we’re on pace for McKinsey’s “accelerated demand scenario. The scenario with $7.9 trillion in AI infrastructure spending this decade.

And while this is a global phenomenon, the majority of those dollars will flow through the U.S. by funding American-built data centers powered by American-designed chips.

Industry leaders agree. As Ciena CEO Gary Smith said last week…

Globally, more than $7 trillion is projected to be spent through 2030 to accelerate investments in data centers, GPU clusters, the power grid, and AI model development.

This capital wave is already reshaping the economy. AI-driven investment has become a bigger driver of U.S. GDP growth than consumer spending. And unlike consumers, who can tighten their wallets overnight, hyperscaler budgets are locked in years ahead.

When Amazon, Microsoft, or Google commits hundreds of billions to AI, that money will be spent.

Source: Bureau of Economic Analysis/Haver Analytics

That’s why we’ve said repeatedly: once the Federal Reserve begins lowering rates – likely next week – the stage will be set for the next surge higher.

Lower borrowing costs will unleash a new wave of easy lending. That’s exactly what’s needed to fund this once-in-a-generation buildout. And it won’t just lift semiconductors. Networking firms, infrastructure providers, utilities, and electrical equipment makers all stand to benefit.

This is the heart of our thesis at Brownstone Research. Trillions in capital investment, exponential growth in compute demand, and a coming wave of cheap money.

For patient investors, that’s the roadmap to enormous gains… We’re just getting started.

Regards,

Nick Rokke
Senior Analyst