One of the most diabolically clever things that EMC, venerable storage maker and darling of the Dot Com Boom, ever did was swoop in and buy VMware back in December 2003 as the server virtualization transition in the datacenter and the resulting cloud boom were just set to explode.

VMware was preparing to go public, with much fanfare, in the wake of the Dot Com bust, which was nearly three years old at that point, to raise heaven only knows how much money. EMC swooped in with $635 million in cash – an offer that co-founders and spouses Diane Greene and Mendel Rosenblum could not refuse. This may seem like a pittance for a company that only had two rounds of funding for $25 million and was largely funded by its founders and employees, btu that was a lot of money way back then. Three years later, as EMC was prepping to take VMware public, it sold a $150 million chunk of VMware, and when EMC did finally take VMware public in the summer of 2007, it ended up with a market capitalization of $26 billion. Dell, of course, bought EMC (and thus a majority stake in VMware) in the fall of 2015 for $67 billion, and even when Dell went private two years later, VMware remained public. In 2021, Dell spun out the huge chunk of VMware it owned, and it wasn’t long before Hock Tan, the chief executive officer at Broadcom, snapped it up and took it private again. VMware has probably brought in on the order of $160 billion in revenues in its history thus far, and is a profitable and slowly growing part of the Broadcom empire.

I have always contended that there was no way for EMC or Dell to sell VMware to Wall Street because investors already owned VMware by virtue of holding stock in EMC and then Dell. But clearly, I am the crazy one being all literal and such. With more years under my belt and lower expectations in the Era of Open Grift, I have learned in this modern attention economy of ours – and don’t think for a second that I am not happy about this –the world, meaning our collective action driven by very human forces, does not have to make sense. It does not have to make sense. The world, as defined by the various levels of influencers starting at the top of governments and industry and working on down from there, merely has to make us react, act, or not act, as the case may be, and profit by it.

And so, as I contemplate the financials of the massive Amazon conglomerate, which has businesses nested inside of businesses and which has attained a market capitalization in excess of $3 trillion in the wake of announcing its second quarter financial results, I inadvertently started a thought experiment about how Bezos, Jassey, Garmin, & Co could unlock a potentially huge amount of “value” by spinning out Annapurna Labs, its chip design business, free of Amazon Web Services and also spin out AWS free of e-tailing and retailing parent Amazon. And what the heck, you could also spin out the media and advertising businesses free while you are at it.

Don’t think I have lost my mind here – I am not talking about complete spinouts. But maybe 20 percent or 30 percent stakes, with the Amazon mothership holding majority stakes.

The days when Amazon was a retailer making hardly any money that happened to have a profitable and fast-growing cloud computing businesses are well behind us now. AWS is riding the GenAI boom high, and its own chips as well as those it buys from Nvidia, AMD, and Intel are in high demand for both AI workloads and traditional back office and data analytics stuff that makes the world go round. The company’s Trainium 2 and Trainium 3 capacity for the next two years is largely booked out, chief executive officer Andy Jassy said in his letter to shareholders in April, which is why AWS has confidence in its AI capex. We presume the same is true for the GPU fleet at AWS, but you can’t expect for corporate PR to let Jassy say that if it is true.

And it is not just GPUs and Trainiums that are selling. Jassy Not one, but two hyperscalers offered to buy out the entire year’s worth of all Graviton capacity for all of 2026. That’s how crazy it is out there.

That Annapurna Labs chips business has been transformative for AWS and for Amazon, which its one of the largest users of AWS services in the world.

In that letter to shareholders as well as in the financial reports for the first quarter of 2026, Amazon said that the chip unit at Amazon, which includes the Graviton CPU, the Trainium XPU, and the Nitro DPU processors, grew by 40 percent compared to Q4 2025 and had an annualized run rate of over $20 billion as the first quarter of 2026 ended. Jassy also said that if AWS sold its homegrown chips to the outside world rather than renting them as EC2 capacity, it would have an annualized run rate of around $50 billion.

This time around in Q2 2026, the PR folks at AWS tell us that that the Annapurna Labs custom silicon business has an annualized run rate of more than $25 billion and is still experiencing triple digit growth. “As best as we can tell, our custom silicon business is now one of the top three datacenter chip businesses in the world,” Jassy pronounced in an email sent to The Next Platform. “And the speed at which we have gotten here is extraordinary."

This implies that the theoretical external revenue for AWS silicon would be on the order of $60 billion.

The Graviton CPU is now being used by more than 130,000 customers as of the end of Q2 2026, an increase of 45 percent year on year.

It is hard to know if the value of Annapurna Labs is baked into the Amazon stock price or not. But clearly there is an advantage to Amazon’s margins from this custom silicon business – and one that a spin out might be able to capitalize on. Ditto for AWS itself, which is an IT behemoth in its own right and quite a profitable business at that. If Annapurna was spun out, it would have to show its payments to AWS for chip development as well as how much money it gets by selling chips to AWS and soon Anthropic and OpenAI, which have commissioned gigawatts of combined Trainium datacenter capacity to drive their model building.

In the second quarter, the combined groups and divisions of Amazon had revenues of $200.61 billion, up 19.6 percent. Operating income rose by 43.2 percent to $27.46 billion, and thanks to a $53.4 billion gain on its investment in OpenAI, parent Amazon posted net income of $62.65 billion, up 2.45X year on year.

The company drew down just a tad over $22 billion in is cash reserves after adding about more than that from the operating businesses independent of that taxable gain on the OpenAI investment. Most of that cash was blown on capital expenses, which came to $47.3 billion in Q2 F2026. The company raised its capex guidance by 10 percent for 2026, to $220 billion, mainly driven by system component price increases.

The AWS revenue backlog stands at a staggering $496 billion.

In the quarter, the AWS business had $42.23 billion in revenues, up 36.8 percent year on year and up 12.4 percent sequentially from Q1. AWS operating income came to $16.62 billion, up and even more impressive 63.6 percent year on year but up only 17.4 percent sequentially.

Some of this increase in operating income has to do with keeping iron in the field longer and depreciating it slower. Jassy explained the capex situation to Wall Street thus:

“There are two major parts of the investment, the datacenters and the servers and networking equipment that go into them,” Jassy said. “These have different capital cycles. Datacenter capital is spent starting two years before we can put servers into them to start monetizing. Once a datacenter opens with servers plugged in, we start generating significant revenue right away and then get to monetize these data centers for 30-plus years without having to spend that start-up capital again.”

“Servers and networking equipment operate on a shorter cycle. We typically purchase these a few months before putting them into service, so we have strong visibility into customer demand before we trigger the spend. If the demand isn't there, we won't spend the capital. For servers and networking equipment, on average, it takes a little less than three years to break even on that investment. The servers currently have a useful life of at least five to six years, and most of our AI capacity these days is being contracted for at least five-year terms. That means that we're driving significant free cash flow on the servers and networking equipment in the two to three years after we break even.”

I think Jassy is being modest. AWS generates a hell of a lot of money from keeping that stuff in the field.

The interesting bit about this GenAI boom is how it has revived growth for AWS overall and for compute in particular. It is not anywhere near the growth that AWS had when it was a startup, of course, but it is the fastest growth that AWS has seen in four and a half years. And given the price increases on DRAM, flash, CPUs, and GPUs, the revenue growth rate will only accelerate in the coming quarters. But remember: This A portion of this growth is not driven by new business, it is opportunistic shakedowns by the component makers who have clouds, hyperscalers, and OEMs over a barrel.

Every quarter, I take my best stab at trying to figure out how much of the AWS revenue stream comes from compute, storage, networking, and software. This is admittedly a total fabrication based on hunches along, since Amazon has never given any hints about such a breakdown. Take a gander at this:

This is meant to be a conversation starter, so if you have your own way of breaking down AWS revenues into useful categories, reach out.