Tae Kim on NVIDIA earnings: 85% revenue growth, buyback catalyst, and why memory stocks are still cheap

Tae Kim on Nvidia, memory stocks, and why Google should be worried

Tae Kim, founder of investment research newsletter Key Context, makes a straightforward bull case for Nvidia heading into earnings: 85% revenue growth on an $80 billion quarterly revenue base, with a stock trading at 19x forward earnings — below the S&P 500 multiple, despite growing roughly eight times faster than the index.

Kim's read is that the discount exists because AI skeptics keep pricing in a near-term growth peak that never arrives. Every year for three years, the market has called peak Nvidia; every year, Nvidia has kept compounding. The competitive threat from Google TPUs and Amazon Trainium barely registers in his math — he puts their GPU revenues at "low single digit" percentages against what he describes as a $1 trillion order book for Nvidia. AMD and Intel are similarly marginal in his framing.

Buyback as re-rating catalyst

The more interesting argument is what happens to the multiple if Nvidia starts buying back stock in size. Kim draws the parallel to Apple during the iPhone 5 and 6 cycle, when the stock traded at 7x earnings ex-cash while the market feared Android would take share. Apple's buyback program drove a re-rating from single digits to 15–30x. Nvidia has signaled it will return 50% of free cash flow over the next twelve months, though Kim notes the company was "cryptic" about whether that figure is calculated before or after prepayments to suppliers. If Nvidia puts a hard dollar number on buybacks this earnings cycle, Kim expects the multiple to move.

“NVIDIA revenue jumped 85% to $81.6B. The stock is trading at 19 times forward, below the S&P 500, which is growing at 10% — NVIDIA is growing at 80%. I think NVIDIA is going to start buying back stock in size; that's when the multiple starts going from single digits to 30 times, just like Apple did with the iPhone.”

Memory stocks

Kim is equally convicted on HBM memory. He cites a Michael Dell comment from a Wall Street conference as the core thesis: Nvidia GPUs will carry 25x more memory per chip in two years, and there will be a need for 25x more GPUs, implying a 625x revenue multiplier on memory demand. That lands against a supply backdrop where memory companies cut capacity after revenues halved in 2022, and where it takes three to four years to rebuild — leaving only three companies capable of making HBM. Kim says these stocks are trading at single-digit P/E multiples with triple-digit growth ahead, and has no sympathy for the skeptics. His Key Context portfolio, which leaned heavily into CPU and memory names after what he describes as a market bottom on March 30, has seen 13 out of 14 ideas move 50–150% higher.

China: write it off

Kim's China view is simple: stop counting on it. Despite Jensen Huang signaling in March that approvals were in hand, the deal fell apart within weeks. Gaming GPU sales have faced incremental bans. Both sides have flip-flopped enough that Kim treats any China revenue as optionality, not a base case. "Forget China," he says — it's not a meaningful number even if it partially reopens.

TSMC supply

On the supply constraint question, Kim reads TSMC's most recent earnings transcript as a meaningful shift. TSMC said its CapEx over the next three years will be "much higher" than the prior three — and Huang has publicly said supply constraints on memory, wafers, and optical components should ease materially within two to three years. Kim's view is that Nvidia's current valuation implies roughly 10% growth; he thinks 50–75% over the next three years is the more realistic scenario, and the market is simply mis-pricing the gap.

Google: a structural problem

Kim turns negative on Google after Google I/O. The specific gap is coding agents — the category he argues is "growing exponentially" and where enterprises are pouring tens of billions into Anthropic and OpenAI's Codex. Google's own coding agent has no traction. Nobody he talks to likes it.

His deeper concern is historical. He uses Yahoo and Netscape as the warning label: Yahoo powered its search with Google and handed Google the market. Google is now funding Anthropic through Google Cloud, benefiting short-term on cloud revenue, but potentially ceding the flywheel to Anthropic long-term. The more usage accrues to Claude, the smarter Anthropic's models get, the harder it becomes to pull that traffic back to Gemini. Microsoft faces the same risk with OpenAI. The companies providing the distribution are also training their future competitors.

Meta

Kim stays constructive on Meta. Their core advertising business is, in his view, more durable than Google Search — which he describes as a degraded experience increasingly crowded with ads. Meta's AI spending is large but functions like the metaverse writedown in reverse: if they pull back or break through, the stock re-rates sharply either way. He's also genuinely impressed by what Meta's infrastructure could support, having tested an OpenAI-powered WhatsApp bot that pulls his top 20 daily stories on demand. His point is that Meta has the surface area to deliver that experience at scale — and hasn't yet.

The bottom line: Kim sees Nvidia as structurally undervalued at 19x forward with 80%+ growth, memory stocks as the most mis-priced trade in semiconductors, and Google as the incumbent most at risk of repeating Yahoo's mistake.