Nvidia is the cheapest it has been relative to its own growth since the AI trade began, and that is the least interesting thing about it. At $216.85 the stock trades on roughly 21.7 times forward earnings while guiding to 85% revenue growth. The interesting thing is what appeared in the accounts three months ago: in the January-to-April quarter, Nvidia\u2019s net income exceeded its operating income \u2014 $58.3bn against $53.5bn. That had never happened before.
\nIt means non-operating gains more than covered Nvidia\u2019s entire tax charge. The largest component is mark-to-market on an investment portfolio increasingly full of the companies that buy Nvidia\u2019s chips. GAAP earnings of $2.39 a share came in 28% above the non-GAAP $1.87 \u2014 the reverse of the normal relationship, because non-GAAP strips those gains out. Add the disclosure Nvidia filed on 17 August, in which it took on residual value guaranties capped at $105bn covering roughly 4.25 gigawatts of Ohio data centre capacity leased to an OpenAI affiliate, and a pattern is visible: Nvidia is progressively underwriting its own demand. None of that is hidden and none of it is illegal. It does change what you are buying.
\nOur base case: a $340 bull case against a $135 bear case, with the stock at $216.85 as of the 20 August close. One date dominates everything below \u2014 Nvidia reports second-quarter fiscal 2027 results on 26 August 2026, five days from publication. Nothing in this analysis reflects that print.
\nKey facts<\/p>\n
NVDA closed at $216.85 on 20 August 2026, against a 52-week range of $164.07 to $236.54 \u2014 stockanalysis.com, 20 Aug 2026
\nQ1 FY2027 revenue was a record $81.6bn, up 85%, with Data Center revenue of $75.2bn, up 92% \u2014 NVIDIA 8-K, 20 May 2026
\nGAAP net income of $58.3bn exceeded GAAP operating income of $53.5bn \u2014 the first time on record \u2014 NVIDIA 8-K, 20 May 2026
\nQ2 FY2027 guidance is $91.0bn \u00b12%, and assumes zero Data Center compute revenue from China \u2014 NVIDIA outlook, 20 May 2026
\nOn 17 August Nvidia entered residual value guaranties capped at $105bn on ~4.25GW at the Portsmouth, Ohio site, with an OpenAI affiliate as tenant \u2014 NVIDIA 8-K, 17 Aug 2026
\nGross margin was 74.9% GAAP; the company added $80bn to its buyback and raised the dividend from $0.01 to $0.25 \u2014 NVIDIA 8-K, 20 May 2026
\n62 analysts polled by S&P Global rate the stock Strong Buy, average target $304.73, low $180, high $500 \u2014 S&P Global via stockanalysis.com, 20 Aug 2026<\/p>\n
NVDA has spent a year in a range while earnings nearly doubled. Bull and bear cases are FinanceFeeds estimates; Q2 FY2027 results land on 26 August.
\nThe operating business is extraordinary \u2014 and that is not in dispute
\nRevenue of $81.6bn in a single quarter, up 85% year on year and 20% sequentially, is the largest absolute quarterly revenue increase any company has produced. Data Center revenue of $75.2bn grew 92%. Within it, networking grew 199% to $14.8bn \u2014 the least discussed and arguably most defensible part of the franchise, because switching fabric locks customers into the rack architecture far more durably than any individual GPU does.
\nGross margin was 74.9%, recovered from the 60.5% of a year earlier when the China H20 write-down landed. Operating expenses of $7.6bn against $53.5bn of operating income is an operating margin of 65.6%. There is no comparable business at this scale.
\n\u201cThe buildout of AI factories \u2014 the largest infrastructure expansion in human history \u2014 is accelerating at extraordinary speed,\u201d said Jensen Huang, founder and CEO of NVIDIA, in the results release. On the operating numbers, the claim is simply accurate.
\nNvidia also behaved like a company confident in its cash: roughly $20.0bn returned in the quarter, an additional $80.0bn buyback authorisation approved on 18 May, and the quarterly dividend raised twenty-five-fold from $0.01 to $0.25.
\nWhat changed in the accounts
\nNow the part that has not been widely reported.
\nIn every quarter until Q1 FY2027, tax reduced operating income to a smaller net income. In Q1 FY2027 the relationship inverted.
\nIn each of the seven quarters before Q1 FY2027, GAAP net income came in below GAAP operating income, as it does at any company that pays tax. In Q1 FY2027 net income was $4.8bn higher than operating income. Nvidia guided its full-year tax rate to 16%\u201318%, so a normal quarter would have converted $53.5bn of operating income into roughly $44bn of net income. It reported $58.3bn. The gap between those two figures \u2014 on the order of $14bn \u2014 came from below the operating line.
\nThe clean confirmation is the GAAP-to-non-GAAP relationship. Non-GAAP earnings are normally higher than GAAP, because they exclude stock-based compensation. In Q1 FY2027 Nvidia reported GAAP EPS of $2.39 against non-GAAP EPS of $1.87. Non-GAAP net income was $45.5bn against GAAP\u2019s $58.3bn \u2014 a $12.8bn difference in the unusual direction, which is what happens when the excluded items are gains rather than costs.
\nThose gains are substantially unrealised marks on Nvidia\u2019s holdings in AI companies. Nvidia owns 9.3% of Nebius, a position it could not sell until 11 September. It holds stakes across the neocloud sector. When those shares rise, Nvidia\u2019s GAAP earnings rise with them \u2014 and those companies rise partly because they are buying Nvidia hardware. This is the same mechanism that made Intel\u2019s government-stake accounting confusing last year, running in the opposite direction.
\nThe practical guidance: use the non-GAAP number for Nvidia and ignore the GAAP headline. $1.87, not $2.39, is the earnings power of selling chips.
\nThe $105bn guarantee
\nOn 17 August 2026 Nvidia disclosed a multi-year partnership with SB Energy for the PORTS Technology Campus in Pike County, Ohio, securing land, power and shell capacity. The tenant is an affiliate of OpenAI Group PBC.
\nThe structure is what matters. Nvidia entered residual value guaranties on leases covering approximately 4.25 gigawatts of IT load, with an aggregate payment obligation \u201ccumulatively capped at $105 billion\u201d for the initial commitment, per the 8-K. It can extend credit support to a further ~3.8GW at its sole discretion. FinanceFeeds reported this as a reduction from a previously indicated $250bn, and relative to that it is one. In absolute terms Nvidia has guaranteed the residual value of infrastructure leased by its own largest emerging customer, at a scale equal to roughly 2% of its market capitalisation.
\nIt is not the only such arrangement. Nvidia has also partnered with Wall Street firms on a $500bn AI infrastructure financing push. Vendor financing is a normal feature of capital-goods industries; it becomes a problem only when demand would not exist without it. Nobody can currently prove which of those two Nvidia is in, and that uncertainty is a large part of why a company growing 85% trades at 21.7 times forward earnings.
\nChina: guided to zero, and quietly reopening
\nNvidia\u2019s $91.0bn Q2 guidance explicitly \u201cis not assuming any Data Center compute revenue from China.\u201d That is a deliberately empty bucket, and it creates asymmetry into the print.
\nThe Financial Times reported on 19 August that H200 shipments to China have resumed, with ByteDance and Tencent each receiving roughly 10,000 units in recent weeks. Read the constraints before treating that as a windfall. US licensing permits far larger volumes per approved customer across around ten cleared firms, so actual shipments are a small fraction of the ceiling. More importantly, Beijing has reportedly told those companies to keep the hardware outside the mainland \u2014 routed to Hong Kong \u2014 to avoid undermining domestic chipmakers.
\nSo China is not returning as a growth engine. But because guidance assumes zero, any China revenue at all is upside to a number the market has already accepted. That is a favourable setup, and a small one.
\nInto the 26 August print
\nNvidia has moved about 2.5% on average in the session after each of its past four reports, which is remarkably contained for a stock of this profile and tells you the market has learned to expect a beat. The stock has also underperformed the broader semiconductor complex this year despite the growth, which is why the bull argument has shifted from momentum to valuation \u2014 one widely followed sell-side analyst has argued the shares may be as much as 50% undervalued.
\nThree things matter more than the revenue number. First, the Q3 guide, and specifically whether China is still assumed at zero. Second, the gross margin trajectory: 74.9% is near the historical ceiling and the Vera Rubin transition brings new cost structures. Third, whether the non-operating gains recur. If GAAP again exceeds non-GAAP by a wide margin, the market will start discounting the headline EPS on principle \u2014 and it should.
\nThe bull case: $340
\nThe bull case does not need heroic assumptions, only continuity.
\nTake fiscal 2027 landing near $390bn of revenue, consistent with $81.6bn delivered, $91.0bn guided, and the ramp implied by the current backlog. Non-GAAP earnings of roughly $10 a share follow at current margins. Assume fiscal 2028 grows 35% to around $525bn as the Vera Rubin platform ships and ACIE \u2014 Nvidia\u2019s sovereign and enterprise AI segment \u2014 contributes, giving non-GAAP earnings near $13.50 a share.
\nAt 25 times, a discount to Nvidia\u2019s own history and to its growth rate, that is $340: 57% above the current price and above the 52-week high of $236.54. It sits between the $300 street median and the $500 street high, which is where a bull case belongs.
\nThe bear case: $135
\nThe bear case is not that AI fails. It is that the buildout digests.
\nAssume fiscal 2028 revenue flat to modestly down from fiscal 2027 as hyperscalers pause to absorb capacity already installed \u2014 the classic capital-goods pattern, and the one every prior semiconductor cycle has followed. Gross margin compresses toward 68% on a worse mix and rising memory costs, an input pressure already visible in TSMC\u2019s price increases. Non-GAAP earnings fall to roughly $7.50 a share. Investment marks reverse rather than add, so the GAAP number falls faster.
\nAt 18 times \u2014 the multiple a cyclical semiconductor company earns at the top of its cycle \u2014 that is $135, or 38% below the current price and below the 52-week low. It is also below the lowest of the 62 published analyst targets, which is $180. That is the point: if the cycle turns, price discovery happens below where the sell side has modelled, because the sell side does not model cycle turns in advance.
\nOur reading: the operating business justifies a price above $216, the accounting quality has deteriorated at the margin, and the vendor financing has grown large enough that Nvidia\u2019s demand and Nvidia\u2019s balance sheet are no longer fully independent. The forward multiple of 21.7 is not the market calling Nvidia cheap. It is the market pricing the probability that fiscal 2028 looks nothing like fiscal 2027.
\nFrequently asked questions
\nWhat is the Nvidia stock prediction for 2027?
\nOur bull case is $340 and our bear case is $135, against a spot price of $216.85 on 20 August 2026. The bull case assumes fiscal 2028 revenue near $525bn and non-GAAP earnings around $13.50 a share at 25 times; the bear case assumes a digestion year with earnings near $7.50 at 18 times. The 62-analyst consensus compiled by S&P Global is $304.73 with a Strong Buy rating.
\nWhen does Nvidia report earnings?
\nNvidia reports second-quarter fiscal 2027 results on 26 August 2026. The quarter ended in late July. Guidance issued in May was for revenue of $91.0bn plus or minus 2%, with gross margin of 74.9% GAAP and no assumed Data Center compute revenue from China.
\nWhy was Nvidia\u2019s GAAP EPS higher than its non-GAAP EPS?
\nBecause the items excluded from non-GAAP were gains rather than costs. In Q1 FY2027 Nvidia reported GAAP EPS of $2.39 against non-GAAP EPS of $1.87, and GAAP net income of $58.3bn against non-GAAP net income of $45.5bn. The difference is largely mark-to-market gains on Nvidia\u2019s investments in AI companies. For assessing the chip business, the non-GAAP figure is the more useful one.
\nIs Nvidia financing its own customers?
\nIn part, yes, and it is disclosed. Nvidia holds equity stakes across the AI cloud sector, including 9.3% of Nebius, and on 17 August 2026 it entered residual value guaranties capped at $105bn covering roughly 4.25GW of Ohio data centre capacity leased to an OpenAI affiliate. It has also joined a $500bn AI infrastructure financing effort with Wall Street firms. This is standard practice in capital-goods industries; the open question is how much of the demand would exist without it.
\nIs Nvidia stock cheap at 21.7 times forward earnings?
\nOn growth, yes \u2014 a forward multiple near 21.7 against 85% revenue growth is unusual. The market is not mispricing the current quarter; it is discounting the durability of the next several. The multiple reflects three doubts: whether hyperscaler capex plateaus, whether gross margin holds near 75%, and whether GAAP earnings quality is deteriorating as investment gains grow.
\nWill China restart Nvidia\u2019s growth?
\nNot materially on current evidence. The Financial Times reported H200 shipments resuming in August 2026, with ByteDance and Tencent each receiving roughly 10,000 units \u2014 a small fraction of licensed volumes \u2014 and Beijing has reportedly directed firms to keep the chips outside mainland China. Because Nvidia guided Q2 assuming zero China Data Center compute revenue, however, any contribution is upside against expectations.
\nThis article is analysis, not investment advice. Figures are drawn from NVIDIA\u2019s SEC filings and from S&P Global Market Intelligence data as of 21 August 2026. Nvidia reports second-quarter fiscal 2027 results on 26 August 2026; nothing here reflects that report.<\/p>\n","protected":false},"excerpt":{"rendered":"