# "GPU demand" — X 热门讨论 (2026-09-10 19:18 UTC)
## @VSR9000 (VSR) · 09-10 08:46 · ♥33 ↻6 💬3 #ESDS #E2ENetworks #TCS #HCLTECH
The biggest mistake is treating all AI beneficiaries as comparable.
They are starting from very different places.
1️⃣ E2ENetworks owns more of the compute layer. That gives direct upside when GPU demand is strong, but it also brings higher capex, utilization and obsolescence risk.
ESDS is different. It already has cloud, managed services and regulated enterprise relationships, but part of its AI economics depends on contracted compute. So the key question is not AI demand alone, but how much value it can add above that contracted compute and whether it can defend margins.
Many are comparing only PE between E2E and ESDS, and retail investors can easily get misled by these surface-level comparisons.
In my view, ESDS is expensive at current valuations.
2️⃣And margins themselves can mislead.
ESDS/E2E-type infrastructure businesses can show much higher EBITDA margins, while TCS/HCLTech operate at lower service margins.
But the divergence comes from the business model.
Compute businesses carry GPU capex, depreciation, utilization and obsolescence risk.
Services businesses carry people cost, but much lower hardware intensity.
So a 40–50% EBITDA margin in compute is not automatically superior to an 18–25% EBIT margin in services.
3️⃣TCS and HCLTECH start from the opposite end.
Their core strength is enterprise relationships, integration, engineering and IP.
Now they are moving downward into AI infrastructure.
That is why pricing power, capital risk and economics visibility cannot be judged with the same lens across all four.
🚩And if hyperscalers keep expanding capacity, improving efficiency and moving further down the stack, can they eventually make smaller undifferentiated AI-infrastructure players obsolete?
Same AI tailwind. Different starting points. Different risks. Different economics.
DYOR https://x.com/VSR9000/status/2097969954838745457