Everyone Is Asking If Nvidia Is 'Cheap' Right Now. The Question Itself Is the Problem.

Everyone Is Asking If Nvidia Is 'Cheap' Right Now. The Question Itself Is the Problem.

I've been running hosting infrastructure for over a decade, and in all that time I've learned one thing about markets: when the stock that built an entire industry suddenly gets called "cheap," it's never just about the stock. It's about what everybody else is being asked to believe.

Here's where we are. Nvidia closed Tuesday at $219.74, a stone's throw from the $236.54 record it set in May. Six days before it reports earnings — August 26, after the bell — Bank of America put out the word that the AI chip giant is undervalued. Not slightly. BofA has a $350 price target, calls Nvidia its top semiconductor pick, and says the stock trades at only about 16 times forward earnings — the lowest forward P/E in nearly a decade. On the free-cash-flow basis BofA prefers, Nvidia trades 34% to 50% below its AI-compute peers.

Let me tell you something: when I read that, I didn't feel reassured. I felt like I was watching two different markets describe the same company.

The News — A Stock Pushing Against Its Record High, and a Bank Saying It's Undervalued

The setup is straightforward. BofA expects Nvidia to report second-quarter revenue of $94 billion to $95 billion — above the company's own guidance of $91 billion — and guide third-quarter sales to $107 billion to $108 billion, comfortably above the roughly $104 billion the Street is modeling. Stifel is on the same side with a $282 target and a beat-and-raise call. The bull case is not complicated: the biggest buyer of AI compute on earth is still growing faster than the company told you it would, and the market is refusing to pay a growth multiple for it.

That refusal is the whole story. A company that prints over $20 billion of free cash flow a quarter — BofA's math puts its projected free cash flow at about $470 billion over the next two years — trading at a decade-low forward P/E is, on its face, the kind of setup that makes fund managers call their grandmothers.

And it's not just the number-crunchers. The GPU spot market is screaming demand: B200 rental rates are near historical highs at $5.66 an hour. If somebody is still paying $5.66 an hour to rent compute, the demand curve hasn't rolled over yet. BofA thinks Nvidia keeps 65% to 70% of the AI accelerator market through 2030 and earns more than $25 a share by then.

The Bull Case — Why the "Cheap" Crowd Has Real Numbers on Its Side

I want to be fair to the bulls, because they're not wrong about the arithmetic. Nvidia's free cash flow is genuinely enormous. It just authorized an $80 billion buyback — the largest in company history — and raised its dividend. BofA's model basically says: imagine Nvidia generates $100 of free cash flow, $50 of it comes back to shareholders through buybacks and dividends, and the market values that return stream at a peer multiple. Under that math, the stock is a bargain.

The reason the P/E got low in the first place is worth spelling out. Nvidia hasn't been hoarding cash like a normal mega-cap. Between 2022 and 2025, it returned only about 47% of its free cash flow to shareholders — the industry norm is closer to 80%. The difference went into the ecosystem: a roughly $30 billion stake in OpenAI, investments in Anthropic, and a widening web of vendor financing that now includes the $500 billion compute-financing platforms it announced with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR on August 10. BofA's retort to the worrywarts: those direct equity commitments are only about 15% of that projected $470 billion of free cash flow. The balance sheet can take it.

So the bull case is coherent. It's not dumb. It's just — from where I sit — incomplete.

The Two Readings — A Market Finally Catching Up, or a Market Being Paid to Look Away

Reading one: the market is finally seeing Nvidia clearly. The earnings keep compounding, the free cash flow keeps printing, the buyback is enormous, the demand is real, and the stock — after all the AI-bubble noise — is objectively cheaper than it's been in ten years. Under this reading, "cheap" is a gift. You buy the pick-and-shovel company of the largest infrastructure buildout in history at a discount, and you hold on.

Reading two: the market is being paid to look away. The same Nvidia that looks cheap on forward earnings is the Nvidia backstopping its own customers — guaranteeing loans, taking equity in OpenAI and Anthropic, running $500 billion of compute-financing platforms, and making its chips the collateral for a credit machine I wrote about two weeks ago (Post 8202) and that the Bank of England felt the need to publicly warn about. The "vendor financing risk" that investors are supposedly pricing into the shares, per TheStreet's coverage of BofA's note — that's not a hypothetical. That's the actual structure of this cycle. A low P/E on a company that has turned itself into the financial system of its own industry isn't necessarily a bargain. Sometimes it's the market telling you the earnings are real, the risks are somewhere else, and the somewhere else is where the pain will land.

Both readings are true at the same time. That's what makes this setup so dangerous.

The Secondary Bottleneck Nobody's Talking About — Peak Earnings Make the P/E a Lie

Here's the part that keeps me up at night, and it's not the stock — it's what the stock is priced against. A forward P/E of 16 times means the market believes the earnings Nvidia will produce next year are roughly the earnings it will keep producing. That's a statement about the future, and the future has a cost problem hiding in it.

DRAM now runs 40% to 50% of manufacturing costs, according to the same analysis BofA is circulating — historically it's been a fraction of that. Component costs are climbing. The Vera Rubin transition supposedly only shaves about 60 basis points off margins, which is the kind of precise reassurance that analysts give you when they want you to ignore the direction of travel. And the demand that justifies $107 billion quarters is being financed — not funded, financed — by a credit stack that I've spent the last month documenting: private credit checks to Anthropic (Post 8645), hyperscaler bonds with no real insurance behind them (Post 8722, this morning), and a data-center IPO wave arriving exactly as the private money needs to exit (Post 8199).

That's the secondary bottleneck: not chips, not power, not cooling — information. A "cheap" P/E on peak earnings is the oldest trap in finance, because the P/E is only cheap if the peak holds. And the peak now depends on a credit market that is simultaneously telling you it's fine ($350 targets) and quietly re-pricing risk (weak bond order books, rising spreads, regulators publicly nervous).

What This Means for Independent Hosting Providers

First — stop reading the P/E, start reading the guidance. The August 26 print is a market-clearing event for hardware pricing. If Nvidia guides to $107 billion-plus, expect another round of capacity hoarding and worsening lead times; if guidance disappoints, expect the secondary GPU market to move before your suppliers do. Watch the number, not the multiple.

Second — watch the spot market like a canary. That $5.66-an-hour B200 rate is real-time truth. When spot pricing rolls over, the buildout narrative rolls over with it, and colo pricing follows within two quarters. Rent before you buy when the signal is ambiguous.

Third — price against the credit stack, not the stock chart. The buildout is financed by debt, vendor financing, and now equity markets. Every dollar of that is a claim on future revenue that has to be paid before your invoices get paid. If the bond market re-prices AI paper, your cost of capital moves whether or not NVDA goes up.

Fourth — don't confuse a cheap stock with a cheap cycle. A 16-times forward P/E on a hyper-growth cyclical at the top of a capex boom is not the same thing as a bargain. It's a company whose margin story is being subsidized by a credit machine that hasn't been tested yet.

Fifth — and this is the founder's truth — the asymmetry has flipped. When the market can't decide whether a $94 billion quarter is cheap or expensive, the only people who can still price compute rationally are the ones who own it outright. That's your edge. Keep your balance sheet boring.

The Bottom Line

Bank of America is asking whether Nvidia is cheap. The honest answer is that it's both cheap and terrifying at the same time, because the price-to-earnings ratio stopped being the relevant metric about eighteen months ago. What matters now is whether the earnings hold up while the whole industry leans on them — and whether the credit market keeps funding the lean.

I'll be watching August 26 like it's my own payroll. If you run infrastructure, you should be too. The stock is not the signal. The guidance is. And the guidance is only as good as the credit underneath it.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: Yahoo Finance video (August 19, 2026), InsiderFinance (August 19, 2026), TMGM/Bank of America analysis (August 10, 2026), 24/7 Wall St. (August 19, 2026), TheStreet (August 19, 2026), Benzinga (August 18, 2026).