EOCN

Sequoia Capital partner Shaun Maguire says Amazon.com Inc. (NASDAQ:AMZN), Alphabet Inc. (NASDAQ:GOOGL) and Microsoft Corp. (NASDAQ:MSFT) are “absolute jokes” at building physical infrastructure compared with Elon Musk’s SpaceX (NASDAQ:SPCX).

Maguire, whose firm holds a large SpaceX stake, predicts the company could generate $731 billion in annual revenue by 2028, largely by renting out computing power to AI companies.

“I’m forecasting that the hyperscalers and neoclouds have severe delays, and SpaceX is on time,” he wrote.

Why Maguire Thinks SpaceX Can Build Faster

Maguire argues that SpaceX can manufacture critical equipment and solve engineering problems internally rather than wait for outside suppliers.

SpaceX says it brought its Colossus II computing cluster online in 91 days from groundbreaking, compared with roughly two years to build a conventional 100-megawatt data center from scratch. Its original Colossus cluster took 122 days but used an existing factory.

Microsoft has already encountered some of the delays Maguire predicts. CFO Amy Hood said the company received some third-party computing capacity later than expected.

Google, one of Maguire’s three targets, is already a SpaceX customer. It agreed to pay $920 million a month at full capacity for access to approximately 110,000 Nvidia GPUs.

But rapid construction has caused problems at SpaceX itself. Some facilities reportedly lacked backup cooling and power systems, with outages interrupting AI training.

Musk Says the Forecast May Be Too Low

Maguire’s model projects SpaceX reaching 21 gigawatts of terrestrial computing capacity by the end of 2028, up from 1.7 GW in September.

His $731 billion revenue forecast is more than double Wall Street’s roughly $336 billion consensus estimate. That would mean growing roughly fourteenfold from the $51.5 billion he expects this year.

Musk responded to Maguire’s $731 billion forecast: “I think we can do better.”

“I think you can too!” Maguire replied. He described his model as a 25th-to-50th percentile estimate, suggesting he sees more upside than downside.

But his projections also depend on computing demand and rental prices remaining strong as capacity expands.

A drop in Nvidia H100 rental prices to $1 per hour for five consecutive days is one of six potential triggers in Polymarket‘s AI bubble market. Three must occur within 90 days to qualify.

Traders assign a 9% chance of a qualifying downturn by December 31, 2026, rising to 23% by June 30, 2027. The market has attracted $3.17 million in volume.

Maguire’s prediction rests on SpaceX continuing to build faster than its rivals, while demand for computing power grows faster than the industry can supply it.

Image: Shutterstock