BECOME A MEMBER TO KEEP READING
(It’s free)
Is NVIDIA’s Throne at Risk—46% Says So
Welcome back. Today, we’re diving into a spicy question: Is NVIDIA, the GPU juggernaut, sleeping too comfortably on its AI chip throne? Spoiler alert—its biggest threat isn’t some scrappy startup or overseas rival. It’s the tech titans who sign its paychecks, and they’re brewing their own silicon rebellion.
From Google’s Tensor flex to Amazon’s Trainium swagger, let’s unpack why NVIDIA’s VIPs could turn into its worst nightmare.
NVIDIA’s Fab Four Problem
NVIDIA’s revenue is a high-stakes poker game, and four players hold 46% of the chips.
That’s right—almost half of NVIDIA’s $61 billion in revenue last year leans on just four customers. While the company keeps names hush-hush in SEC filings, whispers point to Google, Amazon, Meta, and Microsoft as the likely heavyweights. Tesla or XAI or X, whatever we are calling it often gets thrown into the mix, too, though it’s less certain.
This concentration is a red flag: if even one of these giants pivots, NVIDIA’s stock—already trading at lofty forward earnings—could take a hit.
Back in ‘94, Warren Buffett dropped this gem in his Berkshire letter: “We steer clear of businesses where one bad roll could tank the whole game.”
Not quite “Never invest in a business that depends on one customer,” but it’s the same vibe—don’t bet the farm on a single trick.
NVIDIA's juggling four big dogs for 46% of its cash, and that’s a tightrope walk over a shark tank (IYKYK).
Look out below…
Innovating in NVIDIA’s Backyard
The VIPs aren’t just buying NVIDIA’s GPUs—they’re building their own.
Google’s Tensor Processing Units (TPUs), now in their sixth generation (Trillium), are a direct jab at NVIDIA’s dominance.
- Google claims Trillium delivers 4x faster AI training than its predecessor, tailored for its cloud and DeepMind workloads.
- Sundar Pichai’s bet: why pay NVIDIA’s premium when you can cook in-house?
Amazon’s no slouch either.
- Its Trainium2 chip, designed by Annapurna Labs, promises up to 40% lower costs than NVIDIA’s GPUs for AI model training.
- Apple’s even jumped on board, with Benoit Dupin, Apple’s AI director, confirming at AWS re:Invent that Trainium2 offers a 50% efficiency boost for pretraining models like Apple Intelligence. That’s a mic-drop moment—when Apple, the king of vertical integration, endorses Amazon’s silicon, NVIDIA’s margins start sweating.
Meta and Microsoft are also tinkering with custom chips (like Meta’s MTIA and Microsoft’s Maia), signaling a broader revolt against the “NVIDIA tax.”
The Cisco Cautionary Tale
Yes, we’ve all heard this opinion, but let’s rewind to the ‘90s.
Cisco was the networking god, its routers and switches powering the internet’s rise. Under John Chambers, Cisco’s market cap hit $500 billion by 2000, making it the world’s most valuable company.
But its telecom clients—think AT&T and Verizon—got tired of premium prices. They turned to cheaper rivals like Juniper and even built their own gear. By 2001, Cisco’s stock had crashed roughly 80%, and it never regained its peak.
NVIDIA’s not Cisco—its GPUs are still the AI gold standard—but the vibe’s familiar. When your biggest customers start DIY-ing, loyalty fades fast. The future is never a straight line, and Jensen better hope it’s not a cliff.
Edge AI: The Decentralized Dream
While hyperscalers dominate headlines, a quieter revolution brews: edge AI.
Forget cloud data centers—startups and VCs are betting on AI that lives on your phone, car, or fridge.
- Syntiant, a California-based startup, raised $35 million in a Series B round backed by Microsoft’s M12 and others. Their tiny, low-power chips run AI on devices, cutting reliance on NVIDIA’s beefy GPUs.
Meanwhile, firms like Qualcomm and Arm are pushing edge-friendly silicon, with Arm’s designs powering 90% of mobile devices globally. This “great unbundling” could shrink demand for centralized AI compute, leaving NVIDIA’s data center empire looking like overkill.
We kept it tight this week because, well, it's been one of those weeks that feels like a decade. So much going on, but we hope this edition has piqued your interest.
To close it out, if Google, Amazon, and the edge AI rebels keep innovating, Jensen’s leather jacket might not be enough to keep the crown.
