GM. This is Milk Road AI, the newsletter hunting down AI's next bottleneck.
The first AI bottleneck was compute. Then it shifted to memory. We believe connectivity is next and today's edition breaks down our latest trade in that space.
Speaking of building the future, today’s partner is Arc by Circle.
Your agent can code, browse, and find the best deals — but it’s flat broke. Circle Agent Stack is a programmable money layer for the agentic economy that lets agents hold USDC, discover and pay for services, and get paid.
THE NEXT BOTTLENECK IN THE AI BUILDOUT 🚨
As an investor, your job is simple: find the biggest bottleneck in the AI buildout.
That's where the best opportunities tend to emerge. The CEO of Intel breaks this thesis down very well here.
Broadly, there have been three core bottlenecks that have defined the AI buildout:
Bottleneck #1: Compute
- The first bottleneck was raw processing power.
- The industry needed dramatically more compute to run modern AI.
- NVDA led that revolution and became the world's first $5T market cap company.
Bottleneck #2: Memory
- Larger models required enormous amounts of memory and bandwidth.
- Memory companies scaled aggressively to meet that demand.
- Three new $1T market cap companies (including MU) emerged from this cycle.
Bottleneck #3: Connectivity
- We think the bottleneck is shifting again.
- Now it's connectivity that will define the limits of the infrastructure.
- Just like compute and memory before it, the industry will rally to meet this challenge with Marvell (MRVL).
Here’s our simple thesis on Marvell:
The world's biggest cloud companies don't want to pay Nvidia's margins forever. Instead, they're designing their own custom AI chips.
Marvell is the company that's helping to build them.
Right now, Marvell has 18 active custom chip programs with 12 of them tied to the four largest hyperscalers on the planet. Management just guided fiscal 2027 revenue to $11.5B, up 40% year over year.
And by fiscal 2029, its custom silicon business alone is expected to generate $10B+ in annual revenue.
Here is Marvell’s biggest USP:
When a hyperscaler hires Marvell to build their AI accelerator, Marvell also sells the switches, the optical interconnects, the memory fabric and the retimers that go around it.
One customer turns into four or five revenue streams.
This is exactly the kind of setup our analysts look for before the crowd catches on. Try Milk Road PRO for just a buck and see what they're buying next.
But as there is with every investment, Marvell comes with its risks:
Risk #1: Amazon makes up 30-40% of Marvell's revenue so any disruption to that relationship would hurt.
Risk #2: The stock isn't cheap. You're paying a real premium for this growth.
Risk #3: Broadcom is chasing the exact same market.
But a company with locked-in multi-year design wins at every major hyperscaler is a hard setup to pass on and that’s why Melvin just bought MRVL yesterday.
We caught the memory bottleneck early with MU (up 150%+). Now we think the next bottleneck is connectivity and we're betting on MRVL.
WE MADE 12 TRADES LAST WEEK 💰️
Yes, really. Our analyst spent the week buying healthcare, software, and even MORE bottleneck stocks. All this after buying Micron, Nebius, and Bloom Energy back in March.
And they aren't done. See what they're buying next, for just a buck at Milk Road PRO:
THE FIRST FINANCIAL STACK BUILT FOR AGENTS
Your AI agent can browse the web, write code, and schedule your meetings.
Hand it a bill, though? Uh-oh spaghetti-o.
Until now, money was the missing piece. Circle Agent Stack fixes that.
Circle Agent Stack is the financial-infrastructure layer built specifically for autonomous agents:
- Agent Wallets
- Agent Marketplace
- Nanopayments powered by Gateway
- Circle CLI
- Circle Skills
All powered by USDC.
Agents are evolving from assistants into economic actors. The agents that run on purpose-built financial infrastructure will be the ones that actually work.
