Industry commentators warn that the AI infrastructure bubble narrative is reemerging as capex commitments accelerate and funding rounds consolidate at mega-scale.

The listing of SpaceX and forthcoming artificial intelligence floats bear similarities to the lead-up to the 2000 dot-com crash, which resulted in losses exceeding $5 trillion. Even survivors like Amazon, Microsoft, Cisco, Dell and eBay, which possessed sufficient capital to weather the turmoil, suffered massive share price declines that took years to recover.

Today, familiar mistakes around technology, investment approach, business models, valuation and oversight are being repeated. As the actress Tallulah Bankhead observed, investors appear to believe that if they had to live life again, they would make the same mistakes—only sooner.

The dot-com boom centered on the internet, its enabling infrastructure and retail commercialisation as applications developed. Investors with minimal technical knowledge piled in, seeking huge returns. Today's focus has shifted to space and artificial intelligence.

Consider SpaceX, an unwieldy conglomeration encompassing Starlink satellite operations, a space launch business, a controversial social media service, a struggling AI venture, and plans for orbital data centres, a moon base and interplanetary colonisation. While satellite broadband and the X platform rely on established technologies, the launch business's cost advantage depends on reusable rockets that remain a work in progress. Orbiting data centres and interplanetary colonies remain technically unproven. The SpaceX prospectus offered unhelpful techno-babble, speaking of extending "the light of consciousness to the stars" and harnessing the sun "to power a truth-seeking AI".

During booms, investors aggressively finance prospects with limited understanding and minimal due diligence, triggering herd-like tactics and poor business models that amplify risks and speculative excess. Though some lessons have been absorbed, others recur. SpaceX's launch revenues are underwritten by the US government. That business and Starlink will confront significant challenges from nationally sponsored and subsidised competitors, especially in Europe and Asia, reflecting increasing reluctance to outsource critical national infrastructure to US interests.

Other AI businesses are hitting the stock market in the months ahead. OpenAI's pivot from retail to enterprise focus, mirroring Anthropic's strategy, reflected lacklustre conversion of free users to subscriptions. Companies have resisted costs as suppliers shift from subscription models to token-based payments, with many users now capping usage.

Both OpenAI and Anthropic face headwinds from increasing competition by cheaper, open-sourced models—primarily from China—alongside threats of regulation and export bans due to potential AI security applications. Model development costs are rising because of skill scarcity and shortages of processors, electricity and water for cooling.

These businesses all possess uncertain paths to profitability. SpaceX, whose primary profitable business currently is Starlink, warned in its prospectus: "We have a history of net losses and may not achieve profitability in the future." Burning cash rapidly with large unfunded commitments, these enterprises depend on uncertain funding access.

Yet valuations have decoupled from reality. In October 1999, shortly before the dot-com crash, the market cap of 199 internet stocks tracked by Morgan Stanley was $450 billion, against annual sales of approximately $21 billion and collective losses of $6.2 billion. SpaceX, by contrast, raised around $75 billion in June based on a valuation of nearly $1.8 trillion—over 90 times current revenue and 220 times earnings. Morningstar's analysis argued that even with generous assumptions, SpaceX was worth less than half that amount. OpenAI and Anthropic are expected to be valued exceeding $1 trillion, figures higher than those implied by their latest funding rounds.

Current prices reflect not future free cash flows but tribal affiliation, identity and deep faith in technology. When SpaceX listed, an equity trader on the firm's trading floor proclaimed: "In 1969 we put a man on the moon… Now let's go to Mars!" For many, SpaceX shares were cheap on a new measure: a price-to-universe ratio.

With negative earnings and cash flow, values rest on unreliable indicators like 'eyeballs'—unique website visitors or page views. As in the late 1990s, loss-making companies trade at premiums to profitable firms, and promoters eschew profitability because it would lower valuations.

Corporate governance is conspicuously absent. Elon Musk, the world's first paper trillionaire, will control the world's first 'orbital infrastructure conglomerate' via a dual-class share structure that curtails shareholder oversight and ensures his removal remains impossible. His known disregard for governance and self-dealing strategy shifts were dismissed without consequence.

Banks, analysts and media, despite obvious conflicts of interest, amplify the 'new economy' narrative. Goldman Sachs, an underwriter, expects SpaceX's AI revenue to increase 100-fold by 2030. The SpaceX offering generated fees exceeding $500 million for involved banks. Exchanges desperate to boost listed tech stocks agreed to include SpaceX in indices under expedited entry rules, forcing investors—particularly passive investors—to sell existing holdings to accommodate SpaceX, Anthropic and OpenAI shares. Intermediaries benefit from substantial trading volumes.

The real purpose of these IPOs is to allow insiders to cash out, transferring risk to over-enthusiastic and unsuspecting investors. In 2000, once the 180-day lock-up period expired, allowing original funders and employees to sell restricted shares, widespread sell-offs flooded the market. Overvalued listings also furnish founders with currency for acquisitions. Musk may merge SpaceX with Tesla, consistent with his prior transactions involving SolarCity, Twitter and xAI. His unfettered SpaceX control would obviate expensive, heavily-leveraged negotiations to take Tesla private.

Like the dot-com episode, this cycle too is likely to end badly. To paraphrase historian Christian Wolmar writing of British railways, booms cannot sustain themselves on "little more than optimism feeding on itself".