Brighter with Herbert Recap: SpaceX's Second 10 Gigawatts and the 'Dark GPU' Theory

The latest full video on the official "Brighter with Herbert" YouTube channel, "EXCLUSIVE: SpaceX's Second $500B (10 Gigawatts) Explained", was posted on October 31, 2026 and runs about 57 minutes. Host Herbert is joined by two guests the description introduces as Brian Wang, the science blogger behind NextBigFuture, and Cern Basher, a chartered financial analyst who runs his own advisory firm. They try to answer two questions: how SpaceX might reach 20 gigawatts of AI computing power in 2027, and why so many AI chips that companies have already bought are reportedly sitting unused while they wait for power. All three are enthusiastic about SpaceX, so please read this as one side of the discussion.

This is an independent summary written by GNOUP. It is not affiliated with, sponsored by or endorsed by Brighter with Herbert, Herbert, the guests, SpaceX, Tesla, NVIDIA, Google, Anthropic, Microsoft, Amazon or anyone else named in the video. The views, forecasts and claims described below belong to the speakers, not to GNOUP. Nothing here is investment, financial, legal or tax advice, and nothing here is a recommendation to buy or sell any security. The channel's own disclaimer says its content is for discussion only and that the host is long Tesla and other stocks. Please consult a licensed financial professional before making financial decisions.

Key takeaways

These points come from the video and its description. GNOUP has not checked the figures or claims.

• The goal: from 10 to 20 gigawatts. As the panel describes it, Elon Musk has said SpaceX is confident about building 10 gigawatts of AI data center capacity, and that reaching 20 gigawatts in 2027 is no longer a "zero percent" possibility, though not something he is promising. Brian Wang says the company appears confident about the buildings and power, and that the open question is whether it can get enough chips and memory to fill them.
• Power is the bottleneck, and SpaceX builds its own. Brian argues that many AI companies have chips but not the finished buildings, cooling and electricity to run them, and that some electrical equipment can take years to arrive. He says SpaceX is different because it supplies much of its own power: gas turbines, a power plant under construction, Tesla Megapack batteries, and grid power that he thinks it can secure faster thanks to Tesla's experience running battery-based "virtual power plants."
• The "dark GPU" idea. Brian's boldest theory is that SpaceX could rent finished, powered buildings to big tech companies whose own data centers are delayed, so they can plug in chips they already own and start earning sooner. He suggests these could be short leases of 12 to 18 months and estimates SpaceX might earn about $30 billion per gigawatt per year from them. Cern says the idea makes strong economic sense, while Herbert says he doubts SpaceX is counting on it and sees it as a possible bonus.
• A building almost every month. Brian's model has SpaceX finishing a new standardized "Mini Hard" data center building about two or three times a month in 2027 to hit the 20-gigawatt target, assuming each building holds about 220,000 chips. Herbert pushes back that the panel should already be seeing more of these buildings under construction.
• The panel's best guess. Asked for a forecast, Brian guesses SpaceX lands around 18 gigawatts, about 2 gigawatts short on its own buildings plus several gigawatts filled with other companies' chips, and puts the resulting yearly revenue run rate in the hundreds of billions of dollars. Cern says he would agree with those numbers if the gigawatt target is met. These are personal estimates, not company guidance.

Newsworthy insights

These are claims, predictions and opinions as the video presents them. GNOUP has not verified them, and other analysts may disagree.

• Big tech says compute is tight. In Herbert's intro, he says Amazon has racks of chips waiting in warehouses, Microsoft's CEO has complained about chips he cannot plug in, and Google said on its earnings call that it is renting outside computing capacity because supply is constrained. Brian says that rented capacity comes from SpaceX, and that Google and Anthropic have both said they are leasing more from it. He cites a filing from Anthropic that he says shows its monthly spending on leased capacity rising sharply. GNOUP has not checked these statements against the original calls or filings.
• Why everyone might benefit. The guests argue that idle chips hurt everyone: the companies that paid for them, customers waiting on backlogged orders, and NVIDIA, which they say increasingly shares in revenue once chips are switched on. Brian compares the risk to the "dark fiber" glut of the early 2000s dot-com crash, when too much unused capacity sat idle.
• A tour of the data center sites. Cern walks through aerial and satellite images of SpaceX's Colossus 2 site in the Memphis area, pointing out a newer, much smaller building that he says holds about as many chips as an older one more than twice its size, plus cooling equipment, Megapacks and a power plant under construction. He then shows two buildings going up near Starbase, Texas, which he estimates may be about 60% larger than the Memphis design, and says the 444-acre site might hold six to eight such buildings.
• Other possible locations. The panel mentions a July Reuters report that SpaceX was looking at the Austin and Bastrop areas, and Brian speculates that a future spaceport in Louisiana would need so much power to make liquid oxygen that data centers could be built there too. Both are speculation.
• Small "pods" instead of big buildings. Herbert raises an analyst note from Pierre Ferragu of New Street Research suggesting thousands of small computing sites at the edge of the power grid. Brian says such modular pods could speed up construction but will take time to scale, and Cern questions how cooling and noise would work if they sit in parking lots near chargers.
• Delays are not cancellations. Brian says that if SpaceX misses its 2027 target, he expects it would simply be a quarter or two late rather than not happen at all.

Watch the video

EXCLUSIVE: SpaceX's Second $500B (10 Gigawatts) Explained on YouTube (official channel "Brighter with Herbert," @BrighterwithHerbert, posted October 31, 2026, about 57 minutes)

Episode guide

Chapter times come from the video's own chapter list. Each time opens the video at that point on YouTube.

• 00:00 How SpaceX gets to 20 gigawatts
• 01:04 SpaceX builds its own power
• 03:02 Up to an additional 20 gigawatts in 2027
• 04:35 SpaceX labor costs kept relatively flat
• 05:35 Stranded compute in partially complete data centers
• 07:47 Leasing a billion dollars a month from SpaceX
• 08:52 SpaceX's multiple sources of energy
• 10:27 Tesla relationships and virtual power plants for grid power
• 14:01 SpaceX making $30 billion per gigawatt per year
• 17:06 Hyperscalers and Anthropic as end customers
• 17:54 Where the GPUs and memory come from
• 21:50 Dot-com crash and dark fiber comparison
• 24:56 Elon: no longer a zero percent probability of 20 gigawatts
• 25:19 What a Mini Hard data center looks like
• 29:09 Mini Hard across State Line Road in Memphis
• 32:16 New data center location in Texas at Starbase
• 44:03 Spaceport in Louisiana needs huge amounts of power
• 45:00 Pierre Ferragu's theory: Megapods and NVIDIA DSX pods

About this summary

This recap was prepared from the video's title, its description and chapter list on YouTube, and its automatic transcript. Automatic transcripts can garble names and numbers (this one spells the data center names several different ways), so we kept to points we could follow clearly, used the spellings from the video's chapter list, and used guest names as the description lists them. Figures, filings, earnings-call remarks and images discussed on the show are described as the video presents them; we have not checked them against the original sources. We also left out the channel's website, membership, merchandise, referral, sponsor, product and course links. GNOUP has no connection to any product, offer or company mentioned and does not promote them. For the exact words and context, watch the video. This article is general information only. It is not investment, financial, legal or tax advice, it does not recommend buying or selling any security, and it is not a substitute for advice from a licensed financial professional.