Morgan Stanley: 10 Key Convictions on AI Investment
0xBroomberg
Morgan Stanley Investment Management laid out ten core judgments on AI, noting $2.3 trillion in committed AI capex since 2017 and projecting compute capacity at 250× current levels by 2028 — a compounding pace with no historical template, demanding an entirely new investment framework.
$2.3 trillion committed — where is the money going and how fast?
Since 2017, roughly $2.3 trillion has been committed to AI capital expenditure globally, and the pace is still accelerating.
In 2025 alone, token consumption — the basic unit of work an AI model processes — grew more than 10×.
At the current rate, MSIM projects AI compute will reach 250× today's level by 2028. This means → the growth is exponential, not linear, and MSIM itself concedes "there is no reliable historical template for this compounding speed."
The bottleneck keeps moving — where it lands next is the opportunity?
MSIM argues the AI bottleneck is not fixed in one layer but migrates continuously: chips → power → memory → networking → cooling, one after another.
Each migration turns an ordinary commodity into a scarce asset. In plain terms = whoever spots the next chokepoint first captures the investment edge.
MSIM's own framing: "The semiconductor story is no longer about who builds the best chip — it is about which layer of the supply chain becomes the next indispensable node."
Is the software industry's billing model about to be upended?
MSIM compares data centers to modern factories, with tokens as the output. "A $2 trillion software industry was built on licensed seats; the next chapter will be built on token consumption."
This means → software revenue is shifting from fixed subscriptions (per-seat pricing) to usage-based billing (pay for what you consume).
For existing SaaS companies, this creates re-rating pressure — the high P/E multiples built on "stable subscription revenue" may need a new justification.
AI moves from "answering questions" to "doing the work" — what does that change?
MSIM flags that AI is shifting from passive response to autonomous execution — what it calls the "agentic transition" (AI no longer just answers your question; it independently completes an entire workflow).
In vertical SaaS, future winners must control three elements at once: data, domain expertise, and distribution.
Reasoning, orchestration, applications, and workflows will generate more durable recurring revenue. This reflects a value migration — from "the model itself" to "who can put the model to work."
AI enters the physical economy and geopolitics — what should investors watch?
AI is entering the real economy through robotics, autonomous driving, drones, and industrial automation. MSIM's call is blunt: "AI is no longer just analyzing the economy — it is starting to run it."
At the same time, AI is increasingly treated as a national-security issue; the US-China contest has moved beyond commerce. In plain terms = this is no longer market competition between companies — it is strategic rivalry between states.
Regulation is also lagging: AI capability is advancing visibly faster than governance frameworks can form, and private companies are making decisions with broad economic and geopolitical consequences while oversight has yet to catch up.
The biggest winner may not exist yet — what does that mean for today's investor?
MSIM's final call may be the most consequential for the long run: AI's biggest winner may not yet have been founded.
The firm draws an analogy to past technology revolutions — infrastructure is always built first, but the most important applications and investment returns tend to emerge in later stages.
This means → for investors currently concentrated in the infrastructure layer, the call is both validation (the infrastructure phase is still accelerating) and warning (the real outsized winner may come from an application-layer company you have not yet seen).
Content is for reference only, not financial advice.