Amit Agarwal

Enterprise AI Architect | Building Agentic AI for Regulated Industries | Memory as Control Plane • Governance Gaps • Orchestration |

[Amit Agarwal @amitb_agarwal]

Here's fiscal ai's chart showing the revenue breakdown. I can only get the last 4 years, the remaining years left to your interpretation. Azure is the intelligent cloud (purple), and has the most growth, Productivity is Office 365 division and the last one is linkedin and others.

[Amit Agarwal @amitb_agarwal]

Spot on @realroseceline.

$UBER looks cheap on the numbers. $TSLA looks absurdly expensive. That doesn’t make $UBER the better stock from here.

The market isn’t pricing $UBER as if it dies tomorrow. It’s pricing a moat that’s being repriced. If AVs drop cost per mile, $TSLA and $GOOGL (Waymo) don’t need every ride. They just need the densest, most profitable cities, then compete on price. Volume can keep rising while Uber earns less per trip. That’s the actual risk.

Scaling unsupervised fleets to Uber’s volume is a brutal ops problem, not a demo problem. Elon promised unsupervised driving for years and Cybercab is just now live in Austin after months/years of mapping. There will always be cities, hours, and wait-time windows where Uber still wins on convenience - AV can be cheaper and still lose if it takes 15 minutes to show up.

Doesn’t save the equity story. Growth days the market will pay up for are over. It’s a race to the bottom on price, Chinese players are eating international, and UBER’s return profile stays meh until the autonomy endgame is actually visible. That could be years.

[Rose Celine Investments 🌹 @realroseceline]

For the 100th time, $TSLA does not need to destroy $UBER for $UBER shareholders to have a problem. It does not need every ride, every city, or even the majority of the market. It only needs to change the economics of the industry.

If autonomous vehicles lower the cost per mile, pricing pressure follows. $UBER could process more rides than ever, report record volume, and still earn less per ride.

And this is where I think the market is being way too casual. $TSLA and $GOOG are not entering autonomy so they can politely coexist and leave the best economics untouched. They are going to fight like cats and dogs for the biggest, densest, most profitable markets, and once two massively capitalized companies start competing on price, assuming $UBER’s current economics remain intact is a very aggressive assumption that feels irresponsible to me.

🌹

[Amit Agarwal @amitb_agarwal]

Hyperscalers can miss on AI products and still own distribution + the right to slow spending.

GPU landlords only work if utilization stays high and depreciation doesn’t eat the P&L.

Same theme. Different terminal value.

[Amit Agarwal @amitb_agarwal]

The AI capex boom is going to bifurcate.
LPL mapped 4 end-states through 2030.
I overlaid the companies. $AMZN and a converted Bitcoin miner are not the same trade

[Amit Agarwal @amitb_agarwal]

Neoclouds and converted miners are a different distribution.
$CRWV / $NBIS: Scenario 3 base, 4 as a real tail. ROIC 0–8%.
$IREN / $CLSK / $RIOT: earlier, more leverage, more execution risk. Negative to low-single-digit ROIC is not a stretch.

[Amit Agarwal @amitb_agarwal]

I broke down the full strategy, the llama.cpp angle, and what to watch next.

Full analysis on Substack:
https://t.co/q91GV0OjCh
What signal would make you rethink your hardware dependency?

[Amit Agarwal @amitb_agarwal]

NVIDIA didn't buy Hugging Face for $12.93B to own models.

It bought the DECISION.

Here's why this is the most important infrastructure move in AI:

[Amit Agarwal @amitb_agarwal]

This deal isn't about killing open source.

It's about making open source run best on NVIDIA — by default.

That's far more durable than closing it.

[Amit Agarwal @amitb_agarwal]

@dannycheng2022 I missed $HOOD, the remaining 3 I have added aggressively