# data center revenue — X 热门讨论 (2026-09-27 22:56 UTC)

## @Venu_7_ (Venu) · 09-27 21:13 · ♥56 ↻7 💬9 Viavi $VIAV - AI Data Center + Optical Testing.

Makes the testing equipment used to validate high-speed optical + data center networks.

FY26 Revenue +40% & EPS +113%.

Technically, holding the 200-day SMA + attempting to break the multi-month downtrend. https://t.co/feNFj44mto https://x.com/Venu_7_/status/2104318489586729152

## @MilkRoadAI (Milk Road AI) · 09-27 17:02 · ♥50 ↻11 💬5 The greatest bull run is just getting started.

Truist estimates that AI cloud annual recurring revenue could rise from $310 billion in 2025 to nearly $2.1 trillion by 2030, representing approximately 47% annual growth.

The bigger story is not simply that companies will keep training larger models but rather inference, which is the computing required every time someone uses AI, is becoming the primary source of infrastructure demand.

Training creates the model once but inference monetizes that model repeatedly through every prompt, generated image, recommendation, software agent, and automated workflow.

This makes AI cloud revenue look less like a one time infrastructure boom and like a recurring consumption business.

AI agents could accelerate this trend because a single request may trigger dozens of model calls as the agent reasons, searches, uses applications, evaluates the results, and corrects its own mistakes.

The declining cost of inference could actually increase total infrastructure spending because cheaper intelligence encourages developers to embed AI into more products and run it more frequently.

This is the same economic effect seen in cloud computing and internet bandwidth, where lower unit costs produced far greater overall consumption.

Truist expects implied AI capacity to rise from 24 gigawatts in 2025 to 100 gigawatts in 2030, while AI’s share of total data center capacity increases from 23% to 50%.

The industry must add 76 gigawatts of AI capacity in five years and that requires power generation, grid connections, transformers, switchgear, cooling equipment, networking, memory, land, construction, and financing.

Now here are some of the stocks that could benefits from this.

Microsoft, Amazon, Google, and Oracle should capture the largest share of cloud revenue, while specialized providers such as Nebius could benefit when demand exceeds hyperscaler capacity.

Micron benefits from rising memory requirements, Arista benefits from connecting larger clusters, and Vertiv, Eaton, GE Vernova, Powell Industries, nVent, and EMCOR benefit from the physical infrastructure supporting those clusters.

Another overlooked beneficiary could be the power industry because data center operators may increasingly sign long term contracts directly with utilities, natural gas producers, nuclear developers, and energy storage providers to secure reliable electricity.

Milk Road Pro subscribers are already positioned across the AI cloud, memory, networking, power, and data center names I think benefit most from this $2 trillion buildout.

If you want to see exactly what I hold, check out my full Milk Road Pro portfolio below.

https://t.co/thIhK9ZH4E https://x.com/MilkRoadAI/status/2104255250756743399

## @nextbigfuture (nextbigfuture) · 09-27 18:32 · ♥32 ↻5 💬1 The minimum level of truly meaningful AI Data Center in space is about 10 Gigawatts per year before the end of 2030. This needs at least 1200 Starship V3 launches with 120 tons of payload and the 70 kw per ton design. V4 Starship with 200 tons of payload per launch needs about 700 launches. One Starship launch per hour for a full year is 8680 launches. If 7000 version 4 starship launches were AI data center with 70 kwh per ton that would be 100 Gigawatts per year. Reusing the booster and upper stage just lowers the costs and makes it easier to get the high frequency launch.

The reason is that the world is adding 20 to 40 gigawatts per year of data center on earth. Creating a new category of AI data center to move the global needle needs to have a lot.

One Gigawatt per year would be meaningful to SpaceX revenue BUT even here they will have 3.2 gigawatts working by the end of 2026. SpaceX in Q4 will be showing a 500 Megawatt (0.5 Gigawatt) per month completion rate. If this carries through 2027 and even speeds up in 2028 and 2029 then SpaceX will have 8 Gigawatts added in 2027 without any Space AI. Rubin minihard 220k data centers would be 700 MW or 0.7 GW per month in 2027. SpaceX earth based AI data center additions is heading to 10-20 Gigawatts per year in 2028, 2029 and 2030.

The practical level needed for this minimum level is about 120 Starship launches per Gigawatt. This is using the 70 kilowatt hours per ton for AI1 or AI2 satellites. The AI satellites need to be about 250 kilowatts each and weight about 2-4 tons. 30 to 40 of the satellites need to fit into each Starship.

The discussion of FAA allowing 1000 launches per year would need to happen and be raised to 2000 launches or more. This would allow 1200 spaceX AI launches. SpaceX would still use hundreds of launches a year for V3 and V4 communication satellites and for NASA and other missions. > 引用 @herbertong: Starship just loaded ship and booster with liquid oxygen and liquid methane in about 30 minutes, reportedly faster than a Falcon 9!

But fueling isn't the limit. Turnaround is.

Larry @TeslaLarry breaks down the reported schedule. Flight 15 is targeting Oct. 19 and Flight 16 is targeting Oct. 30, carrying Starlink 30-1 with possibly up to 60 V3 satellites.

That's an 11-day gap between flights. Getting to one flight per day would mean compressing it by about 11x.

Pad refurbishment, booster reflight, and range availability are where that gets decided.

If Oct. 30 holds, Starship stops being a test vehicle and becomes a revenue-generating launch system.

$SPCX https://x.com/nextbigfuture/status/2104278104885916049