Nvidia reported second-quarter results on 26 August. Going into the release the company was worth about USD 5.2tn, roughly 7.5% of the S&P 500 and a larger index weight than Apple. Consensus revenue for the quarter had risen from USD 78bn in January to USD 92bn. The same eight months included a semiconductor bear market in July, when more than USD 1tn of chip market value was lost in two weeks, and a Korean market that fell by USD 2.5tn before entering a new bull market on 13 August. Estimates of the gap between global spending on artificial intelligence and revenue earned from it now run to USD 1.5tn a year.
Four largest US cloud builders spent about USD 172bn on capital projects in the second quarter, 77% more than a year earlier, and capital expenditure at the five big hyperscalers should pass USD 690bn this fiscal year. TSMC reported second-quarter revenue of USD 40.2bn at a 67.7% gross margin, with high-performance computing at 66% of the mix, and raised full-year growth guidance above 40%. Revenue at the model developers has also become material: OpenAI was running at about USD 27bn annualised in the second quarter and Anthropic reported a run rate above USD 65bn in late July.
Incremental debt rose from 9% of hyperscaler capital expenditure in FY2024 to 32% by mid-2026. Alphabet raised USD 84.75bn of equity in June and S&P downgraded Oracle to BBB- in July, citing negative free cash flow. Much of the commitment sits outside the balance sheet. Footnote disclosures at nine large technology companies put AI-related off-balance-sheet obligations at about USD 3tn, against roughly USD 600bn of reported capital expenditure and about three times their combined lease and debt liabilities. Alphabet’s contractual obligations were USD 811bn at the end of June, against USD 332bn three months earlier.
Select commitments, on-balance sheet vs. off (Most Recent Quarter, USD bn)
Source: Company data, InterCapital Research
Nvidia’s role has widened along with this. The company now guarantees data-centre leases, promises minimum revenues to smaller cloud providers, and holds equity in businesses it also sells to. Lucent Technologies did something similar in the last telecom cycle, lending to customers who could not otherwise buy its equipment, and was cut to a third of its workforce when the credit went bad. Nvidia is far better capitalised and its order book still runs ahead of supply. The open question is whether vendor backstops become the standard way large data centres get financed.
Revenue was USD 96.2bn, up 106% year on year and above both the USD 91bn company guide and the USD 92bn consensus. Data-centre revenue of USD 89.0bn rose 117%. Gross margin was 75.0% against 71.1% for fiscal 2026 as a whole, operating margin was 66.2%, and diluted earnings per share of USD 2.46 compared with USD 1.08. Guidance for the third quarter is USD 108bn plus or minus 2%, above the USD 104bn consensus, and Vera Rubin entered production shipment this quarter. Nvidia is up 14.2% in 2026 while AMD and Intel have more than doubled.
Nvidia Key Financials (FY 2023 – H1 2027, USD bn)
Source: Company data, InterCapital Research
The commitments note carries the new information. Supply and capacity commitments rose from USD 119bn at the previous quarter end to USD 279bn at 26 July, mostly for memory and manufacturing capacity. With cloud service agreements, leases not yet commenced, committed equity investments and capital expenditure, total forward commitments were USD 366bn. A further USD 56bn covers AI cloud agreements and leases Nvidia intends to reassign, and guarantees add USD 108.5bn, taking total forward obligation to about USD 530bn against USD 320bn of total assets. The largest item is a USD 105bn guarantee entered in August over SB Energy’s build for an OpenAI affiliate at Pike County, Ohio, covering 4.25 GW of IT load in nine phases from fiscal 2029 on 20-year leases. It falls away if OpenAI obtains a satisfactory credit rating, and Nvidia holds an option over a further 3.8 GW. The USD 500bn financing programme reported earlier in August is described in the filing as memoranda of understanding that may not lead to definitive agreements.
Nvidia Supply Chain Commitments (Q1 2026 – Q2 2027, USD bn)
Source: Company data, InterCapital Research
Investment gains have become a visible part of earnings. Gains from equity securities were USD 7.8bn of the USD 7.8bn reported as other income in the quarter and USD 23.7bn for the half, nearly all unrealised, or 17% of first-half pre-tax profit. Equity and non-marketable holdings rose from USD 35.1bn in January to USD 93.9bn in July, and the company bought USD 42.4bn of equity securities during the half against USD 1.2bn a year earlier. Part of the funding came from USD 25bn of senior notes issued in June across seven tranches, the first large debt raise Nvidia has done. Accounts receivable rose from USD 38.5bn to USD 63.1bn, which the company attributes to extended payment terms on large multi-quarter agreements.
Nvidia’s Share of Pre-tax Profit from Gains on Equity Securities (Q1 2025 – Q2 2027, USD bn, %)
Source: Company data, InterCapital Research
Customer concentration increased. Three direct customers accounted for 16%, 15% and 13% of first-half revenue, against two at 20% and 15% a year earlier. Within data centre, hyperscale revenue grew 101% while the AI cloud, industrial and enterprise line grew 138%, so the faster growth is coming from the customers Nvidia is helping to finance. China contributed less than 1% of data-centre revenue. Nvidia took a USD 0.4bn charge in the half on excess H200 inventory and purchase obligations, pays a 25% US import tariff on H200 units that it cannot pass on, and has assumed no China data-centre revenue in guidance.
Nvidia is reported to have agreed to buy Hugging Face for USD 12.9bn. The price is close to three times the USD 4.5bn valuation of the 2023 funding round in which Nvidia participated, on annual revenue of about USD 150m. It follows the USD 20bn Groq transaction, of which USD 2.9bn was paid in cash during the first half, and a USD 6bn agreement with Poolside. Hugging Face is the main distribution point for open-weight models, which compete with the closed models built by Nvidia’s largest customers, and it has always hosted models that run on rival silicon.
On these numbers the binding constraint is capital rather than silicon. Memory supply is already contracted inside the USD 279bn, power sits inside the USD 105bn guarantee, and China has been written down to nothing in the outlook. What remains open is whether USD 530bn of forward obligation can be matched by demand that Nvidia is now partly underwriting itself. The revenue estimated as necessary to justify the sector’s investment was USD 200bn in 2023; the current figure is USD 1.5tn. A large part of that spending is now disclosed in Nvidia’s own commitments note.