What the TPU is going on?

How Custom Silicon is turning things sideways, and then sideways again.

Since the start of this AI boom, the narrative has been clear 👇

“NVIDIA is the undisputed king of AI infrastructure and printing with 80% gross margins while Big Tech scrambles to buy every GPU they can get their hands on.”

The picks and shovels play seemed bulletproof, and NVDA stock performance showed it. In fact, Big Tech was fighting to hand out their cash balances to NVDA to build Data Centres.

TAKE MY MONEY!

We touched on the dynamic that this AI race wasn’t about profitability or ROI — it was about survival. Simple: the world’s most profitable companies were fighting an existential battle between themselves.

And all this enabled Jensen and NVDA to achieve historically high profitability and the world’s greatest stock valuation…

But things were slowly shifting under the surface.

At Philoinvestor, we’ve been tracking this custom silicon threat for months across multiple pieces:

Weaponisation of AI from March ‘24

Downside at AI from early ‘25

AI Tech Playbook from Summer ‘25

While others were riding the NVIDIA hype train, we were asking the harder questions.

How long can one player capture this much value?

What happens when Big Tech decides they’re done paying a premium?

What’s the ROI?

And other big picture questions that the market thought to embrace several months later…

The AI Tech Playbook, June ‘25

Weaponisation of AI, March ‘24

Now the narrative is slowly shifting.

So here’s what’s actually happening: TPUs (Tensor Processing Units) and other ASICs sacrifice the flexibility of traditional GPUs for dramatic gains in efficiency and power consumption.

It’s a trade-off that makes perfect sense when you’re running the same workloads at massive scale. You just don’t need the flexibility of GPUs, so why bear all the cost?

—> Google doesn’t need infinite flexibility—they need to run their models as cheaply and efficiently as possible, billions of times per day.

What next?

Philoinvestor cuts through the noise with context other finance newsletters miss. We don’t just report what happened, we connect developments to frameworks we’ve been building for months — and by doing so obtain an analytical edge from the market.

The AI infrastructure story is entering a new chapter, and the easy money has been made.

What comes next requires understanding not just technology, but power, strategic and competitive dynamics — as well as ecosystem analysis and higher-level thinking.

We’ll continue monitoring TPU developments, custom silicon proliferation, and yes, even quantum computing’s long-term implications for this entire stack.

Subscribe to Philoinvestor for analysis that connects the dots before the market does.

What the TPU is going on?

This was a promo piece to the actual piece that will be published — for Philoinvestor subscribers only.

Philo 🦉