Matt Levine: prediction markets are mostly sports betting, SpaceX IPO could suck oxygen from AI fundraising, private credit in AI isn't the 2008 scenario

Summary

Matt Levine (Bloomberg Opinion) offers a bracingly unsentimental read on prediction markets, the SpaceX IPO pipeline, private credit in AI, and the broader question of whether the current tech cycle has a 2008-style systemic risk lurking inside it.

Prediction Markets Are Mostly a Sports Betting Play

The oracle-of-the-future framing around prediction markets is largely marketing. By Levine's own rough estimate, sports gambling accounts for roughly 80% of volume on platforms like Polymarket and Kalshi, and the companies themselves see sports betting as the core customer acquisition engine. The sophisticated-information-market narrative is real but secondary.

The exchange structure is the key financial nuance. Prediction market platforms do not trade against customers the way traditional sportsbooks do, which means the actual money-making layer may ultimately sit with market makers, essentially hedge funds or high-frequency trading firms, rather than the platforms themselves. Susquehanna is already operating a sports desk; other quantitative trading firms are moving in the same direction. The regulatory interface, commodity exchange rules rather than securities law, makes it easier for these firms to participate without the ban risk they face on traditional books.

On insider trading, Levine says founders are philosophically in favor of informed markets but publicly committed to prohibition because CFTC approval requires it. Commodity market insider trading rules are historically more permissive than securities rules, and that philosophy may bleed in over time. With fraud enforcement deprioritized under the current administration, some actors will get away with violations in the near term. The more headline-driven enforcement will stay in sports, where the DOJ has already pursued baseball players and leagues have institutional incentives to police the integrity of their product.

A structural tax quirk may be quietly routing volume toward prediction markets. The current tax bill caps deductions for gambling losses on sportsbooks at 90%, while prediction market losses may be fully deductible as commodity trading losses. For professional bettors running near break-even, that difference can flip profitability entirely.

SpaceX IPO Could Crowd Out AI Fundraising

A reported $30 billion SpaceX IPO at a $1.5 trillion valuation, premised on building data centers in orbit, enters a capital market already straining under $10 trillion in projected AI CapEx demand. Levine takes seriously the thesis that a listing at that scale could compress the available oxygen for competing AI raises from OpenAI, Anthropic, and others, not by making capital unavailable in absolute terms, but by absorbing attention and institutional allocation at a moment when everyone is chasing the same narrative.

The xAI roll-up scenario, folding xAI into SpaceX and taking it public under a space-data-center story, is plausible on paper. A private SpaceX acquiring xAI would avoid the shareholder litigation exposure that a Tesla acquisition would face, particularly given Tesla's existing AI team and the Solar City precedent. SpaceX's current revenue is roughly $22 billion, making a $1.5 trillion valuation a deep multiple that requires a compelling decade-long story, which the data center and xAI narrative could provide.

Private Credit in AI Is Not the 2008 Setup

The 2008 analogy fails on structure. The capital underwriting AI infrastructure buildout is predominantly insurance companies and long-duration institutional investors in investment-grade paper, not leveraged short-term instruments. More importantly, a significant portion of the debt is ultimately backstopped by Meta, Microsoft, Google, and other hyperscalers with strong cash flows and the operational flexibility to scale CapEx down if needed. Meta may have destroyed equity value in its metaverse pivot, but it paid its vendor bills throughout.

The deeper risk question is whether the debt load is commensurate with value actually being created. Levine's answer is cautiously affirmative relative to crypto. AI has tangible consumer products, real enterprise use cases, Disney licensing its characters to OpenAI, developer tools like Cursor with measurable productivity effects. The circularity concern, foundation model companies funding each other's GPU spend, is real but not structurally equivalent to the token-on-token circular economy that characterized FTX-era crypto. Sam Bankman-Fried never believed in crypto as a product; the AI ecosystem has genuine true believers and genuine products. That is not a guarantee of commensurate returns on $10 trillion of CapEx, but it removes the purely illusory quality that characterized the prior cycle.

The dot-com parallel Levine finds more apt. Even in a significant shakeout, there will be an Amazon-scale survivor. That outcome was not structurally guaranteed in crypto. It seems close to inevitable in AI.