Earnings Quality
Current Assessment — June 2026
The Money Companies Are Reporting Is Real. Whether It’s Actually Paying Off Is the Real Question.
Big tech and AI companies just reported some of their best profits in years. The natural question is: is that real, solid growth, or a bubble waiting to pop, as some past tech booms did?
This piece looks at two separate questions that often get blurred together: is the way this AI spending boom is financed actually risky (the short answer here: less than people think), and is all that spending actually turning into real, paying demand fast enough to justify it (the more honest, open question).
A few terms worth knowing:
Earnings beat means a company reported more profit than analysts (people paid to predict this stuff) expected.
Capex (capital expenditures) is money a company spends on big, long-term stuff, buildings, equipment, computer chips, rather than everyday running costs.
Circular financing describes a situation where companies invest in each other in a way that can make growth look more solid than it really is, like two people pretending to get richer by writing each other increasingly large checks.
ROI stands for “return on investment,” basically asking: did the money you spent actually make you more money (or save you more money) than it cost?
Here’s where things stand right now:
First-quarter 2026 earnings beat: 20.7% above what analysts expected
Planned AI spending for 2026: $650 to $700 billion
Companies that have actually scaled AI past a small test phase: under 40%
A penalty banks are already charging: companies seen as spending heavily on AI without proof it’s paying off get charged 0.3 percentage points more interest than companies actually selling AI tools and services
Signal: ELEVATED. Not a clean bull story, not a clean bear story. The actual picture is “bimodal,” meaning it splits clearly into two very different outcomes depending on the company, not one uniform story for everyone.
What Wall Street Is Actually Reporting
The profit numbers are real, and these aren’t shell companies
In the first quarter of 2026, companies in the S&P 500 (a list of 500 major U.S. companies) reported profits 20.7% higher than analysts expected, the biggest positive surprise since early 2021. Eighty-four percent of companies beat expectations entirely. None of this is fake or manufactured.
Here’s an important historical comparison: during the dot-com bubble around 2000, a lot of companies were valued on pure hype and promises about the future, with little actual cash coming in. That’s not what’s happening with today’s leading AI companies. Nvidia alone reported tens of billions of dollars in actual operating cash flow in a single quarter, real money coming in the door, not just promises. Even at today’s prices, the biggest AI-related companies trade at roughly half the valuation multiple that Cisco (a major networking company) carried at its 2000 peak, while generating far more actual cash than Cisco did back then.
About that “circular financing” worry you may have heard about: there’s a real concern that companies like Nvidia, OpenAI, Oracle, and various cloud providers are investing in each other in a way that could be propping up the whole picture artificially. This concern is real, but smaller than headlines suggest. For example, the specific deal between OpenAI and Nvidia represents an estimated 13% of Nvidia’s total 2026 revenue, a meaningful slice, but nowhere close to the majority of it. This looks much more like companies building expensive, real physical infrastructure (data centers, chips, power systems) than a marketing-driven hype bubble, and that distinction matters a lot for how this eventually plays out.
What this means for your money
Don’t treat the “companies investing in each other” worry as the biggest risk here. The way this spending boom is financed, real cash coming in, clearly disclosed deals, money spent on actual infrastructure rather than marketing, is meaningfully more careful and disciplined than the dot-com era was. There’s something here worth watching, but it’s probably not the thing most likely to cause real trouble.
Separate the companies actually making money from the ones just spending heavily. NVIDIA, Microsoft, and Alphabet (Google’s parent company) are funding their AI spending mostly from cash they’re already bringing in. Other companies, like CoreWeave, are funding their AI buildout mostly by borrowing money instead. Both get labeled “AI stocks,” but they carry very different levels of risk.
Pay attention to how the market reacts to spending plans, not just to the spending itself. When Alphabet recently announced its 2026 spending plans, the stock market reacted negatively right away because investors weren’t fully convinced the spending would pay off fast enough. That kind of immediate, skeptical reaction is actually a good sign; it means investors are scrutinizing whether this spending converts into real profit, not just blindly cheering every AI headline.
Watch market reactions to specific guidance raises, not just the dollar totals. When Meta raised its full-year 2026 capex guidance to $125-145 billion (up from a prior $115-135 billion range) and framed it as funding “personal superintelligence,” its stock fell over 9% the same day, the clearest real-world instance yet of the market punishing a spending raise it didn’t see matched by a clear revenue case. That’s a sharper, more current example of the same discipline the Alphabet reaction showed earlier in the year.
Is All This Spending Actually Paying Off Yet?
The money being spent is definitely real. Whether it’s converting into real value fast enough is the honest, open question.
Hyperscalers are now tracking toward $700-725 billion in AI infrastructure spending for 2026; the figure has moved up from the $650-700 billion estimated earlier in the year, as confirmed in Q1 2026 earnings reports. Microsoft alone disclosed an $80 billion backlog of Azure orders it cannot yet fulfill due to power constraints, evidence that at least part of this spending is chasing existing demand rather than getting ahead of it.
Here’s the honest, complicated truth about whether that spending is paying off: it depends enormously on which company or organization you’re looking at. A respected consulting firm, McKinsey, found that organizations that roll out AI tools correctly and at scale are seeing an average 5.8 times return on their investment within just 14 months, genuinely impressive results. Meanwhile, a separate, widely cited finding from MIT suggests that as many as 95% of company AI projects are showing essentially zero measurable financial return so far.
How can both of these findings possibly be true at once? Because they’re describing two very different groups. A minority of companies are implementing AI thoughtfully and getting real, substantial value from it, while the majority remain stuck running small test programs that haven’t been rolled out at full scale yet, sometimes called “pilot purgatory.”
Banks and bond markets are already pricing in this exact split, rather than waiting around to see how it resolves. A major bank, Citi, has identified something concrete: companies seen as “adopters” (spending heavily on AI without clearly proving it pays off yet) are now being charged an extra 0.3 percentage points of interest compared to companies seen as “enablers” (companies actually selling AI tools and infrastructure to others, where the demand for their products is already proven and real). That’s a more current, more precise warning signal than the broader “companies investing in each other” worry described earlier.
The numbers behind this story
2026 AI spending by major tech companies: $650 to $700 billion, up about 63% from the prior year. Citi has actually raised its longer-term forecast for total global AI spending (2026 through 2030) to $8.9 trillion, up from a prior estimate of $8 trillion.
Companies that have moved past small test programs: under 40%. The vast majority of companies experimenting with AI in some way haven’t yet scaled it to full, real production use.
Return for companies that do roll it out well: 5.8 times their investment within about 14 months, according to McKinsey’s research. The upside case here is genuinely real for the companies getting it right, not just a hopeful myth.
Citi’s AI revenue forecast for 2026 through 2030: raised to $3.3 trillion, up from a previous $2.8 trillion estimate, specifically because enterprise demand and adoption are moving faster than expected. Worth noting: the optimistic case is gathering more support as 2026 goes on, not less.
The interest rate penalty mentioned above: 0.3 percentage points (30 basis points) extra interest charged to companies the market sees as spending heavily on AI without clear proof it’s paying off, compared to companies actually selling the AI tools and infrastructure themselves.
What this means for your money, in plain terms
Tell the difference between companies selling AI tools and companies just spending heavily on AI. Companies that actually sell AI infrastructure and software (the “enablers”) already have demand for their products proven in the real world. Companies spending heavily on AI without clearly showing it’s paying off yet (the “adopters”) are exactly the group that bond markets are now pricing as the riskier bet.
Watch whether companies renew and expand their AI contracts, not just whether they announce new pilot programs. Announcing a new small test program is basically just a press release. A company choosing to renew or expand an AI contract it already has is actual proof that the tool delivered real value. That gap between “announced a test” and “renewed for more” is exactly where the very different 95% failure and 5.8x success findings diverge.
Don’t assume rising AI spending numbers are automatically a good sign or automatically a bad sign on their own. The fact that Citi raised both its spending forecast and its revenue forecast at the same time shows that even serious, credible analysts aren’t all agreeing on one single direction here. Rising spending alongside rising real revenue is healthy. Rising spending alongside flat, stalled-out enterprise adoption is the real warning sign to watch for.
Watch for AI spending plans slowing down gradually, not a sudden, dramatic stop. If this whole situation resolves as “the spending got ahead of actual demand” rather than “the entire financing structure collapses,” the most likely path forward is for large tech companies’ spending plans to level off sometime in 2027, rather than a sudden 2000-style crash. That kind of gradual slowdown tends to show up first in what company executives say about future spending plans, well before it ever shows up directly in quarterly earnings reports.
The Genuinely Disputed Question Here
It’s worth being honest: this is a real, ongoing disagreement among serious, credible analysts, not a settled question with one obvious right answer.
The original “is this a bubble” question has actually mostly been resolved already, on the financing side specifically: most serious analysts, including a major bank (UBS) and academic research comparing this AI boom directly to the dot-com era, generally agree that the way this spending boom is financed is meaningfully more solid and disciplined than it was back then.
The disagreement that’s still alive has narrowed to one specific, more focused question: Will companies start using AI quickly and effectively enough to catch up with the $650 to $700 billion being spent on it every year? Citi’s recent forecast increase argues yes, citing faster enterprise demand as evidence. The MIT research, along with the fact that fewer than 40% of companies have scaled beyond small test programs, suggests that the gap remains wide.
Both sides here are working from real, current data, not just speculation or gut feeling. The most honest, balanced position is that this question gets answered gradually over the next several quarters as companies renew (or not renew) their AI contracts, not in a single, dramatic moment everyone will point to later.
RBC Wealth Management’s own research captures this tension well: in their words, current concerns are “yellow warning signs rather than a full-blown bubble,” and they note Big Tech’s forward P/E multiple has actually fallen to roughly 1.2 times the broader market as of January 2026, below its 10-year average of 1.4 times, even as the dollar amount of spending has grown. RBC’s own view is conditional: AI capital spending needs to keep showing up in earnings, and productivity gains need to spread beyond the tech sector itself for current valuations to hold up.
Why Smart People Disagree: The Bull Case vs. The Bear Case
The optimistic case: real cash, and real proof that demand exists
The leading AI companies are generating enormous amounts of real cash flow right now, today, not just making promises about some distant future. Cloud computing backlogs (already-signed deals waiting to be fulfilled) are large and growing, and the new computing capacity being built is getting used about as fast as it’s actually built. Citi’s own forecast increase reflects faster enterprise adoption, not slower. This whole situation looks a lot more like the early days of cloud computing, a buildout that generated strong returns for roughly fifteen years, than it resembles the early-2000s telecom fiber-optic cable boom, which famously destroyed a huge amount of investor value.
This isn’t just an analyst-research view. RIA strategist Brian Vendig has pointed to strong projected earnings as the reason he sees no near-term problem here, putting a name and a current voice behind the same case the data supports: a market can be expensive and still be backed by real, growing profit, and those are two different questions worth holding apart rather than collapsing into one “bubble or not” verdict.
Major brokerages are largely making the same case with their own research. Fidelity’s mid-2026 outlook points out that Q1 2026 earnings growth has been strong enough that P/E valuations look “reasonable” even among the Magnificent 7, calling out 10% revenue growth and operating margins at an all-time high of roughly 16%. Schwab’s own mid-year update goes further, explicitly contrasting this cycle with the dot-com era: “Unlike the dot-com period in the late 1990s, where valuations expanded on little revenue growth, the earnings potential of AI is already showing up in strong earnings growth.”
The pessimistic case: spending is outrunning actual results
Company spending on AI infrastructure is growing significantly faster than the actual cloud computing revenue coming in to justify it. As a direct result, Amazon’s free cash flow (cash left over after covering expenses) is projected to actually turn negative in 2026. The vast majority of companies experimenting with AI haven’t scaled it up into full, real production use yet. If it ends up taking businesses longer to actually adopt and use AI than today’s massive spending plans currently assume, the companies funding their AI buildout with borrowed money, rather than with cash they’re already generating, are the ones most exposed if things slow down.
The Bottom Line
The worry about companies financing each other’s AI spending in a circular way is the weaker argument here, based on the actual evidence, and probably shouldn’t be the main reason to get defensive about AI investments on its own. The real, stronger argument worth paying attention to is the gap between spending and actual usage, and that question gets answered gradually over time, through company earnings calls and contract renewal data, not through one single, sudden, dramatic event.
The most useful thing you can do with this information: when looking at any company connected to the AI boom, ask whether it’s actually selling AI tools and infrastructure that people already clearly want and are paying for (an “enabler”), or whether it’s a company spending heavily on AI without yet proving that spending pays off (an “adopter,” funded more by debt than by its own cash). That distinction matters more right now than worrying about which companies are investing in each other.