Zoho’s Vembu Warns AI Tech Bubble Is ‘Bigger Than 1999’

Sridhar Vembu, founder and chief scientist of Zoho Corporation, has called the AI-driven rally in big tech an “insane bubble, even bigger than 1999,” pointing to price-to-sales ratios of roughly 20 times at Nvidia and close to 19 at Micron. Critics fired back within hours, noting that today’s largest tech firms convert more than half their revenue into profit, unlike the cash-burning startups that defined the dot-com crash.

What makes the warning sting is the number Vembu leaned on. He skipped the usual price-to-earnings comparison and reached for price-to-sales, the same blunt yardstick a famous Silicon Valley executive used in 2002 to explain why his own company’s stock had been absurd. That borrowed math is the real argument, and whether it still applies in a market full of high-margin giants is the question the whole debate now turns on.

The Six Multiples Vembu Flagged

Vembu’s post on X listed six companies by price-to-sales (P/S, a company’s market value divided by its annual revenue rather than its profit). His figures: Nvidia at 20 times, Apple at 10, Alphabet at 11, Microsoft at 10, Meta at 7.5, and Micron at 19. The choice of metric was deliberate. Price-to-earnings flatters profitable companies; price-to-sales strips that comfort away and asks what investors are paying for each dollar of top-line revenue.

At a P/S near 20 times revenue, Nvidia sits where chip and memory names like Micron also cluster, which is striking given Micron’s far thinner margins. Nvidia’s revenue base has ballooned past $130 billion in its latest fiscal year, per its most recent quarterly results filed with the SEC, yet the multiple has not cooled. The table below shows what each name does and the ratio Vembu attached to it.

The 2002 Warning He Borrowed

The line Vembu quoted came from Scott McNealy, the co-founder and former chief executive of Sun Microsystems, the workstation and server maker that was one of the dot-com era’s biggest winners and then one of its biggest casualties. In a 2002 interview, after Sun’s stock had collapsed, McNealy walked through what a 10-times-revenue valuation actually demands of a buyer.

At 10 times revenues, to give you a 10-year payback, I have to pay you 100% of revenues for 10 straight years in dividends. That assumes zero cost of goods sold, which is hard for a computer company. That assumes zero expenses, zero R&D, and that I pay no taxes. Do you realize how ridiculous those basic assumptions are? What were you thinking?

McNealy was describing his own shareholders. Sun’s stock had run from single digits to about $64 at the peak, then cratered roughly 90% as orders dried up after 2000. The company never recovered its independence. Oracle eventually bought it for $9.50 a share in a deal worth about $7.4 billion, announced in 2009 and completed in early 2010, according to Oracle’s acquisition disclosure on EDGAR. By invoking that exact quote, Vembu is not just saying valuations are high. He is saying the market is repeating a specific arithmetic mistake that wiped out a Silicon Valley icon.

Where the Dot-Com Parallel Still Holds

Some of the echoes are real. Market concentration in a handful of mega-cap names today rivals or exceeds the levels seen around the 2000 top, and a single theme, artificial intelligence, is driving the bulk of index gains. When one story carries a market, sentiment can reverse fast.

The picks-and-shovels comparison is also fair. Cisco Systems was the dot-com era’s essential supplier, the company selling the gear everyone else needed, much as Nvidia supplies the chips for the AI build-out. Cisco reported net sales of about $18.9 billion and pro forma net income near $3.9 billion in fiscal 2000, per its year-end results filed with the SEC. It was profitable and dominant, and it still lost the better part of its value when the cycle turned and took more than two decades to reclaim its old high.

That last point is the one bulls tend to skip. Being a real business with real earnings did not save Cisco’s share price from a brutal de-rating. The same risk attaches to Nvidia, which is why some investors read its recent push toward software and recurring revenue as a hedge against exactly this kind of hardware-cycle hangover.

Where the 1999 Comparison Breaks Down

Then the parallel hits a wall, and it is a thick one. In 1999, the market’s leaders were promising profits they did not have. Pets.com and Webvan burned cash and folded. Today’s leaders are printing it. That difference is not cosmetic; it changes what the multiples mean.

One of Vembu’s critics put it bluntly on X: price-to-sales is built for low-margin manufacturers selling undifferentiated products, and it breaks down for a company like Nvidia that turns more than half of every sales dollar into net income. A second user, Kislay Parashar of Cosmic Labs, added that 1999 companies were burning cash while these companies are printing it, and called whether valuations are stretched “a different argument entirely.”

• Above 50% of Nvidia’s revenue converted to net profit in its latest fiscal year, per its fourth-quarter results on EDGAR.
• 36% to 38% net margins at Microsoft and Meta on revenue bases in the hundreds of billions.
• Roughly $350 billion in combined free cash flow from the five largest US tech firms in their most recent fiscal years, by one industry tally.
• Below 1x capital spending relative to free cash flow today, against nearly 4x at the 2000 peak.

The Case That Price-to-Sales Still Bites

Here is what the bull case glosses over, and why Vembu’s metric is not as easy to dismiss as the critics suggest. McNealy’s whole point was that he was talking about a profitable company too. Sun made money. The math still did not work, because a 10-times-revenue price bakes in years of flawless growth, fat margins, and no competition. Pay that price and you have already priced perfection; any stumble comes straight out of your return.

The bears’ reply rests on three uncomfortable points:

• High margins today are not guaranteed tomorrow. AI chip pricing depends on Nvidia’s lead holding, and rivals plus custom silicon are circling.
• A P/S of 20 needs revenue to keep compounding for years just to grow into the price, which leaves no cushion for a demand air pocket.
• Concentration cuts both ways. The same names that lifted the index can drag it down together if the AI capital-spending cycle pauses.

None of that proves a crash is coming. It does mean the “they’re profitable, so it’s fine” rebuttal answers a question Vembu did not ask. Profitability tells you the businesses are healthy. The multiple tells you how much optimism is already in the price.

Why the Zoho Founder Is Saying It Now

Vembu is an unusual messenger, which is part of why the post traveled. He built Zoho into a global software firm without venture capital and without an IPO, and he has spent years warning that easy money distorts how technology companies are valued and run. He recently used Zoho’s messaging app Arattai to argue that real products rest on years of patient in-house engineering rather than hype cycles.

His timing tracks a market where capital is rushing toward a single trade. The pull of US AI names has grown so strong that investors elsewhere are starving local startups of funding, a dynamic that has already drained money from Europe’s smaller tech hubs. When one theme soaks up that much capital, the people who lived through 2000 start checking the exits.

For now, the numbers can be read either way. The same revenue multiple that looks insane against McNealy’s arithmetic looks defensible against margins no dot-com company ever came close to posting. If AI demand keeps compounding at the pace the chip orders imply, the high multiples grow into themselves and the bubble talk fades. If that demand cools even for a couple of quarters, the market finds out the hard way which half of the argument was right.