Feud between AI power startup Fermi and its fired CEO and top shareholder heats up over proposed sale

Fermi’s market cap has fallen to less than $3.2 billion from nearly $20 billion at its IPO last year, while its fired CEO is pushing for an immediate sale and the board is resisting. The company still lacks an anchor tenant for its Project Matador data center campus after a $150 million deal was canceled in December, underscoring execution risk. Management is searching for a new CEO as Fermi reviews strategic alternatives including investments, joint ventures, or a potential sale.

Analysis

The key takeaway is not just governance dysfunction; it is a financing and customer-confidence problem that can metastasize quickly in capital-intensive AI infrastructure. Once a project of this scale starts looking sponsor-fractured, prospective tenants and capital partners begin to treat every commercial term as revocable, which raises the cost of pre-lease commitments and lengthens the path to bankable project finance. That matters more than headline equity value because the equity was already a residual claim; the real risk is that the asset base becomes impaired by underutilization before it is ever fully built.

The board split also creates a timing asymmetry. A strategic sale is most plausible only if a buyer believes the land, power rights, and permitting path can be acquired before the tenant problem is solved; otherwise the asset is worth less as a “platform” than as a distressed option on future AI demand. If management can stabilize even one credible anchor relationship within the next 1-2 quarters, the liquidation narrative weakens materially; absent that, every month of delay compounds execution risk and increases the probability of covenant-like economic distress even without formal debt pressure.

The second-order winner is not necessarily a direct comparable, but incumbent power and data-center infrastructure suppliers with cleaner governance and faster delivery profiles. Any perception that this project is overhyped or politically noisy should redirect enterprise demand toward proven hyperscale capacity, benefiting names with existing campuses, utility-scale grid access, and easier procurement processes. The most exposed counterparties are niche nuclear/industrial partners whose reputations can be diluted if the project becomes the poster child for speculative AI infrastructure excess.

The market is likely underpricing how quickly this can move from a story-stock unraveling to an asset-sale process. The counterargument is that strategic value may still be real because land, power optionality, and public-sector support are scarce; that could attract a deep-pocketed infra buyer at a modest premium to the current equity if the company is forced to choose liquidity over control. But until tenant traction is proven, the stock’s upside is dominated by takeout optionality while the downside is tied to a slow burn of credibility loss.

Key Decisions for Investors

• Avoid initiating longs in NVDA-linked infrastructure proxies that depend on speculative campus buildouts until anchor-tenant confirmation improves; use a 1-2 quarter horizon and require evidence of signed capacity commitments before underwriting upside.
• Pair trade: long established data-center/infrastructure beneficiaries with visible execution and short names exposed to hype-driven AI campus buildouts; target a 3-6 month horizon where capital reallocates toward lower-governance-risk platforms.
• If you want optionality, express it via small-call-spread exposure on public AI infrastructure winners rather than bottom-fishing distressed project developers; risk/reward is better because downside is capped and the catalyst is broader AI capex strength, not one project’s execution.
• For event-driven accounts, monitor for a forced-sale announcement over the next 30-90 days; any bid below prior implied valuation should be treated as a negative signal for the broader private AI infrastructure complex.
• Use weakness in high-beta AI infrastructure sentiment to accumulate quality semicap and AI hardware names on a 6-12 month horizon, as capital should rotate away from governance-heavy “future capacity” stories toward picks-and-shovels with realizable orders.