Big Beat, Meh

The Growth Machine Rolls On

Nvidia did it again: sales and profits above expectations, and another huge revenue forecast for next quarter.

• Record fourth-quarter revenue at $68B, up 73% year-over-year
• Net income at $43B, up 94% from a year ago
• Record full-year revenue at $216B and record net income at $117B

For most companies, the results would send the stock flying — but nothing short of perfection is enough for the investors of the world’s most valuable firm, valued at $4.75 trillion. Instead, they got “just” another monster beat, sending shares down.

Foundry Can’t Build Fast Enough

Almost every advanced AI chip in the world starts at the same company: Taiwan’s TSMC. It is the world’s biggest chip-making foundry and a critical supplier to the likes of Nvidia.

The AI boom has created a supply crunch at TSMC. One of its main customers, Apple, has already admitted it will be hit by the lack of chips.

Nvidia says it has enough supply lined up at TSMC to keep shipping GPUs for months. But everyone wants the same cutting‑edge manufacturing slots, and while TSMC is investing tens of billions in new factories, it can’t magically create more overnight. Small delays can ripple through Nvidia’s new fiscal year.

AI’s Real Shortage Isn’t GPUs

The hottest commodity in AI right now isn’t GPUs — it’s memory. High‑bandwidth memory (HBM) is the stuff that feeds data into Nvidia’s chips, and demand is exploding. Prices are up. Supply is tight. One industry survey even called the shortages “pandemic‑like,” a throwback to the chip crunch during the Covid-19 supply chain shock.

Shares of memory makers like South Korean SK Hynix are soaring, as they capture more of the profits. While Nvidia dipped after the earnings, SK Hynix spiked 8% in Seoul.

Without memory, Nvidia’s growth hits a wall, no matter how good its chips are.

Data Centers Hit the Limits

AI needs more than chips. It needs land, power, cooling, and hundreds of billions in investment per year. “Hyperscalers” like Meta, Amazon, and Alphabet plan to spend over $600 billion on data centers and chips this year — but even they have limits. Energy shortages are slowing projects, and governments are increasingly demanding that data centers have their own power supply.

This “bring your own power” (BYOP) approach means tech companies may not only have to dish out for the data center but also build a power plant on-site. This requires more money and time.

Dependent on Two Customers

Nvidia relies heavily on just a few mega‑customers: two of them now make up 36% of its sales, up from 34% the previous year. Nvidia has not revealed the names of the mystery customers but says they’re direct customers, pretty much excluding the usual suspects like Microsoft and Meta. Instead, names like Dell, Foxconn and Quanta are often floated as good guesses.

Direct customers are the middlemen. They buy Nvidia’s chips in bulk, build them into full systems or circuit boards, and then sell those finished products to cloud providers, data center operators, and other end‑users. Cloud giants still matter massively, as they’re the ones driving the demand.