| Ticker | Name | Research | View | What's said | Source |
|---|---|---|---|---|---|
| NVDA | NVIDIA | QT · SA · STK · FA | Positive | "Can't build fast enough." Q2 FY27 revenue accelerated 106% Y/Y to $96.2B ($4.1B beat) — Data Center +117% to $89.0B, Edge Computing +27% to $7.2B — with gross margin 75% (+3pp), operating margin 66% (+5pp), non-GAAP EPS $2.22 ($0.13 beat), operating cash flow +57% to $24.1B and free cash flow +59% to $21.3B against $99.4B of cash and $33.4B of debt. Q3 guided to $108.0B (+12% Q/Q, +89% Y/Y, a $3.4B beat) at a 74% gross margin, assuming no Data Center compute revenue from China. Growth accelerated (85%→106% total, 92%→117% Data Center), adding ~$15B sequentially with another ~$12B guided; demand broadened — Hyperscale $48.7B while AI Clouds, Industrial and Enterprise grew faster to $40.3B (neoclouds, enterprises, sovereign AI alongside Big Tech). Rubin is arriving without a digestion pause: Blackwell Ultra still ramping while Rubin ships with purchase orders "from every major hyperscaler, AI cloud, and system OEM," and it raises the take per AI factory — ~$18B/GW (Hopper) → ~$25B/GW (Blackwell) → ~$40B/GW (Vera Rubin), delivering 30x throughput per megawatt and 35x lower token costs than Grace Blackwell Ultra. The costs: gross margin bends to 74% in Q3 and ~71%–72% in Q4 on higher memory costs ("supply-driven rather than a sign of weaker demand or pricing"), and supply and capacity commitments jumped from $119B to $279B in three months, largely to secure memory — "a huge vote of confidence in future demand—and a much larger commitment if the cycle eventually slows." NVIDIA as financier: nearly $50B invested in frontier AI labs, partnerships with major financial institutions helping raise >$500B of third-party capital, plus credit support and take-or-pay commitments — "critics call this circular financing. NVIDIA argues it is simply removing a capital bottleneck," with the strategic benefit of locking in deployments "before competing silicon reaches scale"; NVIDIA says labs receiving balance-sheet support could be roughly one-quarter of its business next year. Jensen Huang's answer to custom silicon: "whereas many of these XPUs are inference-specific chips for one cloud or one service, NVIDIA is a platform, an entire AI factory platform that spans the entire AI life cycle that you can use in any cloud" — and on demand: "when the world goes to agentic, fully agentic systems, you are going to have agents running all the time, working with other agents running all the time," which would decouple inference demand from human usage. Outlook: ~70% revenue growth in FY28, "far above prior Wall Street expectations," adding more than $200B of annual revenue. Valuation: up ~20% YTD, still only ~20x forward earnings, "well below the multiple it commanded earlier in the AI boom and below the rest of US Big Tech"; Q2 13Fs showed hedge funds not accumulating as they used to, many still underexposed against its 8% S&P weight. "In semiconductors, a low P/E can sometimes signal peak earnings rather than a bargain… So far, that skepticism has been repeatedly proven wrong." Bear: spending outruns the profits it generates while custom silicon takes inference share. Bull: "agentic AI keeps expanding compute demand faster than efficiency gains and competition can reduce it." "So far, NVIDIA is still winning that race." A disclosed author holding. | article ↗ |
| OpenAI | OpenAI (private) | — | Neutral | Both NVIDIA's customer and the sharpest named threat to it. OpenAI "just published the first results for Jalapeño, its custom inference chip," claiming it "delivered 1.5x–1.9x more throughput per watt and materially lower latency than the NVIDIA systems tested across several models." App Economy's read is measured: "OpenAI still plans to use NVIDIA broadly, but Jalapeño shows that NVIDIA's largest customers have a growing incentive to move specialized inference workloads onto their own silicon." Jensen Huang's rebuttal on the call is aimed squarely at it — "whereas many of these XPUs are inference-specific chips for one cloud or one service, NVIDIA is a platform, an entire AI factory platform that spans the entire AI life cycle that you can use in any cloud" — with App Economy adding that "his argument isn't that customers won't build their own chips. It's that those chips tend to optimize specific workloads, while NVIDIA's advantage is a fungible platform… The question is whether that breadth remains valuable enough to justify NVIDIA's premium economics." Listed first among "what I'm watching." Named as a competitive/strategic factor, not as a stance on OpenAI itself. | article ↗ |
| Hugging Face | Hugging Face (private — being acquired by NVIDIA) | — | Neutral | "According to The Information, NVIDIA has agreed to acquire Hugging Face for $12.9B, nearly triple its 2023 valuation." Hugging Face is "one of the main hubs for developers to discover, share, and deploy open AI models, often dubbed the 'GitHub of AI'" — and with "only about $150 million in annual revenue, NVIDIA is clearly buying strategic positioning rather than near-term profits" (roughly 86x revenue). The rationale ties to Jensen's own argument: "NVIDIA benefits whenever AI models proliferate. Open models are particularly attractive because startups and enterprises generally don't build custom chips and overwhelmingly rely on existing compute infrastructure." Owning it "would move NVIDIA one layer closer to developers and tighten the link between open-model adoption and its broader computing platform." The stated risk: "NVIDIA ownership could weaken the neutrality that helped make Hugging Face so valuable in the first place." Referenced as the acquisition target; no standalone stance. | article ↗ |
"View" is App Economy's analytical framing in this free edition — NVDA positive (accelerating growth at enormous scale, a ~70% FY28 guide, and ~20x forward earnings, with the bear/bull cases laid out explicitly rather than resolved), OpenAI neutral (customer and custom-silicon threat) and Hugging Face neutral (the reported acquisition target). App Economy Insights is financial-analysis journalism, not a buy/sell stance — BUY/SELL/HOLD ratings are shared only with App Economy Portfolio members; the author discloses owning AAPL, AMD, AMZN, GOOG, META, MSFT and NVDA. Research: QT Qualtrim · SA Seeking Alpha · STK Stock Analysis. The "Source" link opens the newsletter (no per-name timestamps — it's a written post). Named only in passing and not given rows: AAPL, AMD, AMZN, GOOG, META, MSFT (author-disclosure line only, not discussed in the body); The Information (the outlet reporting the Hugging Face deal); the S&P 500 (NVDA's ~8% weight); and NVIDIA's own product/architecture names — Hopper, Blackwell, Blackwell Ultra, Grace Blackwell Ultra, Vera Rubin.
A jargon-free summary of the read behind each name. (Plain-language companion to the table above; renders on each ticker's consolidated page.)
NVIDIA sold $96.2 billion of chips and systems in three months — more than double the same quarter last year. What makes that remarkable is not the size but the direction: growth sped up, from 85% to 106%. Big companies almost never do this. The bigger you are, the harder each extra percentage point becomes, because you have to add more absolute dollars to move the same rate. NVIDIA added roughly $15 billion of revenue in a single quarter and expects to add another $12 billion next. And management says next fiscal year should grow about another 70% — more than $200 billion of new annual revenue — while still being limited by how much it can physically build, not by how much customers want.
The most useful new framing in this issue is dollars per gigawatt. AI data centres are ultimately constrained by electricity, so the industry measures itself in gigawatts of power. NVIDIA's revenue from each gigawatt of AI capacity has gone from about $18 billion with its Hopper generation, to $25 billion with Blackwell, to about $40 billion with the new Vera Rubin systems. That matters because it means NVIDIA can grow even when the world can't build power plants any faster — it simply extracts more from each one, by selling not just the chip but the CPUs, networking and software around it.
Two complications are worth understanding properly. The first is what critics call circular financing. NVIDIA has invested about $50 billion into AI labs and helped arrange more than $500 billion of outside capital for AI infrastructure, sometimes backing customers' borrowing directly. So some of the money that comes back as NVIDIA revenue was, in a sense, put there by NVIDIA. The company's defence is that its customers' demand is growing faster than their balance sheets, and it is removing a financing bottleneck rather than fabricating demand. Both can be true. The number that keeps it honest is NVIDIA's own: labs receiving balance-sheet support could be about one-quarter of next year's business. That is the share of revenue that depends on those customers eventually making money from what they're building.
The second is customers building their own chips. OpenAI just published results for its own inference chip, Jalapeño, claiming 1.5 to 1.9 times more work per watt of electricity than the NVIDIA systems it tested. Inference — running a trained model to answer questions — is repetitive and predictable, which makes it the easiest workload to design a specialised chip for. Jensen Huang's answer is not that customers won't do this, but that their chips are narrow: built for one workload in one cloud, while NVIDIA sells a general-purpose platform that handles training, inference and networking anywhere. The real question is whether that flexibility is worth NVIDIA's much higher prices as the workloads become more standardised.
Two more things to hold in view. Profit margins are set to slip from 75% to about 71–72% by the fourth quarter — but because memory chips have become expensive and scarce, not because customers are pushing back on price. And NVIDIA has committed to $279 billion of future supply purchases, up from $119 billion three months ago, mostly to lock up that memory. That is an enormous vote of confidence — and an enormous obligation if demand ever cools.
Finally the valuation, which is the counterintuitive part. After all this, NVIDIA trades at roughly 20 times expected earnings — cheaper than Apple, Microsoft or Google. That sounds like a bargain, and the author flags exactly why it might not be: "in semiconductors, a low P/E can sometimes signal peak earnings rather than a bargain." Chip companies are cyclical; the market often prices them cheaply right before profits fall, because the "E" in the ratio is about to shrink. A low multiple can be a warning dressed as a discount. The author's honest summary is that the debate is no longer whether the stock looks cheap, but how durable these extraordinary earnings are — with the bear case (spending outruns the profits it generates while custom chips take inference) and the bull case (always-on AI agents expand demand faster than efficiency reduces it) laid side by side. "So far, NVIDIA is still winning that race." The author owns the stock. Analysis, not a recommendation.
OpenAI is both NVIDIA's biggest kind of customer and, increasingly, a competitor. It has designed its own chip, Jalapeño, built for one specific job: inference — running an already-trained AI model to produce answers. That is different from training, the enormously expensive process of building the model in the first place. Training is varied and experimental; inference is the same operation repeated billions of times, which is exactly the sort of work a purpose-built chip can do far more efficiently than a general-purpose one.
OpenAI's published results claim 1.5 to 1.9 times more throughput per watt than the NVIDIA systems it tested, with lower delay. Throughput per watt is the metric that matters at data-centre scale, because electricity — not the hardware price — is the binding constraint and the largest running cost. If you can serve the same number of answers on half the power, you can serve twice as many customers from the same building.
OpenAI says it will keep using NVIDIA broadly, and that is credible: designing chips is slow, and most workloads still benefit from flexibility. But the direction of travel is the point. As the author puts it, Jalapeño shows "NVIDIA's largest customers have a growing incentive to move specialized inference workloads onto their own silicon." Inference is also the part of AI demand expected to grow the most, so losing share there matters more over time than losing training share would.
NVIDIA's counterargument, in Jensen Huang's own words, is that these custom chips are "inference-specific chips for one cloud or one service," while NVIDIA sells "an entire AI factory platform" usable anywhere across the whole lifecycle. Whether that breadth stays worth paying a premium for is, in the author's framing, the central open question. Named here as a strategic factor in the NVIDIA read, not as a view on OpenAI itself. Analysis, not a recommendation.
Hugging Face is the place AI developers go to find, share and download open models — the community hub for anything not locked inside a company like OpenAI. The Information reports NVIDIA has agreed to buy it for $12.9 billion, nearly three times its 2023 value, against only about $150 million of annual revenue. That is roughly 86 times sales, which tells you plainly this is not a purchase for the profits.
The logic follows from NVIDIA's core interest: it makes money whenever AI models spread, whoever built them. Open models are especially good for NVIDIA because the startups and companies that use them almost never design their own chips — they rent or buy standard computing infrastructure, and the standard is NVIDIA. Owning the hub where those models are found puts NVIDIA one step closer to the developers making the choice, and tightens the link between open-model adoption and its own platform.
The risk sits in the same sentence as the rationale. Hugging Face's value comes largely from being neutral ground — a place where every model from every lab sits side by side. Being owned by the dominant chipmaker could erode exactly that neutrality, and with it the reason developers gathered there. Referenced as the acquisition target in the NVIDIA read; no standalone view. Analysis, not a recommendation.
Key points & figures extracted from the public App Economy Insights newsletter (article text in transcript.txt) for personal study. Not investment advice. © App Economy Insights for source material.