Google is hoarding TPUs to develop Artificial General Intelligence

Alphabet said it is prioritizing AI accelerator allocation for AGI development and confirming TPU sales to some customers, while still boosting Cloud as AI demand stays strong. Google Cloud revenue rose 82% YoY to $24.75B with $8.8B profit (+214%), and quarterly revenue climbed 24% YoY to $119.8B (operating income +34% to $40.8B), but free cash flow turned negative at -$5.9B (first time since 2004). Management guided higher AI infrastructure spending to $195B–$205B for the year (vs. $180B–$190B) amid supply constraints, and shares fell ~4% after hours as investors focused on the cash burn.

Analysis

The market is likely reacting to the optics of negative free cash flow, but the deeper mechanism is a capital-allocation reset: Alphabet is choosing to defend frontier compute and enterprise retention over near-term balance-sheet cleanliness. That is usually bearish for the stock multiple in the next few weeks because it weakens the “cash machine” narrative and can slow repurchase support, but it is not automatically bearish for 6-18 month earnings power if the incremental compute translates into durable workload capture.

The immediate winners are the infrastructure owners that can monetize Alphabet’s supply gap without owning the customer relationship: DLR and EQIX are the cleanest public expressions, with second-order benefit to networking and rack density suppliers such as ANET and to AI capex beneficiaries like NVDA if hyperscaler spending stays elevated. The loser set is broader than just GOOGL: any cloud name trading on “efficient scale” vs “aggressive growth” gets a relative hit when one of the largest players signals it will keep spending into scarcity, because that can re-rate the whole sector toward lower current FCF multiples.

The key catalyst over the next 1-3 months is whether the company’s capex guide keeps drifting up while operating cash flow lags, or whether the market sees signs that AI-driven workload growth is absorbing the spend. The thesis breaks if Search/Cloud monetization softens, if buyback pace visibly slows, or if management implies the current capex cadence is structural rather than bridging. The contrarian read is that investors may be over-penalizing the quarter’s cash flow optics and underestimating that capacity scarcity itself is a moat: forcing customers into longer commitments can improve lifetime value even when near-term margins look messy.

Key Decisions for Investors

• Long DLR or EQIX on weakness for a 1-3 month trade; use Alphabet’s disclosed compute shortages as a proxy for lease-up demand. Risk/reward improves if Q3 commentary confirms continued third-party capacity use; invalidate if hyperscaler leasing demand does not show up in backlog conversion.
• Relative value: long MSFT / short GOOGL for the next earnings cycle if the market keeps rewarding FCF discipline over AI reinvestment. This is a valuation/multiple trade, not a business-quality call; cover if GOOGL shows FCF recovery or if MSFT capex inflects similarly upward.
• Stay long NVDA or ANET on pullbacks as a basketed AI-infrastructure hedge; Alphabet’s spend trajectory argues that compute demand is still outpacing supply. Cut if hyperscaler capex guides roll over broadly or if Alphabet signals TPU/internal silicon is displacing external hardware faster than expected.
• Alert, not recommendation: if GOOGL trades another 5%-7% below the post-earnings level without any downgrade to Search/Cloud demand, consider buying the dip for a 6-12 month horizon. The thesis is invalidated if next-quarter FCF remains negative and buybacks are curtailed more than expected.