Google is paying SpaceX nearly $1 billion a month for AI — buy the data center suppliers they rely on

Verdict: The AI capex story is real, but insiders and a weak Vistra say wait for the triggers

The strongest argument for this idea is the demand evidence: per CNBC's June 5, 2026 report, Google is paying SpaceX roughly $920 million a month for compute capacity, and the two supply names backing the thesis are performing — Arista's latest quarter showed revenue up 12.1% sequentially to $3.0B with operating margin around 45.3%, while Vertiv grew revenue 23.6% to $3.3B and lifted free cash flow 41.6%. The strongest argument against: in the ownership filings covering the period ended June 30, 2026, insiders at all three names were net open-market sellers — about $865M at ANET, $9.3M at VST, and $3.9M at VRT — and Vistra's latest quarter showed revenue down 12% and net income down 70.4%, the weakest print of the trio. The strategy evidence is thin but real: the 60-month ANET backtest returned 48.1% across just 5 trades with an 80% win rate and a 17.9% maximum drawdown, while the 12- and 24-month windows failed to reproduce, and no robust parameter setup was established because the sensitivity evaluation exceeded its time budget. None of the entry rules has fired yet — VST sits just 0.71 points above its 50-day average of $149.28, VRT is $13.61 below its $282.66 line, and ANET would need a pullback toward $183.57 followed by a fresh upward cross. That combination argues for waiting: the thesis is coherent and ANET's and VRT's fundamentals are strong, but insider selling, Vistra's lumpiness, and unconfirmed triggers mean patience costs nothing. What flips the verdict is confirmation — a completed MACD bullish cross above a 50-day moving-average cross on at least the two supply names, paired with a Vistra earnings print that stabilizes net income.

Trade now: three setups, none triggered yet

None of the three entry rule sets has fired yet, so the honest instruction today is wait — not because the rules lack evidence, but because price has not reached the trigger lines. ANET closed at $195.79, sitting 12.22 points above its 50-day moving average of $183.57, so the price-above-average condition is already met but a fresh upward cross would require a pullback first. VRT closed at $269.05, still $13.61 below its 50-day average of $282.66 — the farthest of the three from its line despite the source note. VST closed at $149.99, just 0.71 above its 50-day average of $149.28, so the cross could happen on a single strong close. Each entry also requires the MACD line to cross above its signal line, plus a defensive confirmation: the day's low must tag the nearest support level while the close holds above it. That first support sits at $193.41 for ANET, around $272.93 for VRT, and $149.19 for VST. Once a position opens, the exits are mechanical: a take-profit signal at 14%, a hard stop at roughly 2.3% below entry, and a protective stop if price closes below the second-ranked support level. That asymmetry — up to 14% target against a 2.3% stop — is what shapes the reward-to-risk profile; the position cap is 25% per name with roughly 2.3% of equity risked per trade. The evidence comes from the completed backtest on the 60-month window: a 48.1% return across 5 trades with an 80% win rate and a 17.9% maximum drawdown on ANET. That is a sparse but coherent record — 5 trades in 5 years means this is a patient setup, with long stretches of no signal punctuated by short holding periods capped at 14 days. Note that no robust parameter setup was established: the sensitivity evaluation exceeded its time budget, so the published rules run exactly as written, with no nearby variant substituted. So what does wait mean concretely? Set alerts at ANET $183.57, VRT $282.66, and VST $149.28 for the moving-average crosses, and watch the MACD-versus-signal relationship on each daily bar. Do not chase entries above the trigger — the 2.3% stop assumes entry near the cross, and buying extended prices turns a controlled setup into an uncontrolled one.

Hyperscaler capex is flowing into the exact names this thesis targets

The thesis rests on a real, citable capital-flow event: per CNBC's June 5, 2026 report, Google committed to pay SpaceX roughly $920 million a month for compute capacity at xAI data centers — over $11 billion a year of hyperscaler spending chasing physical infrastructure. The Bloomberg piece from the same day frames AI-driven capex as the key to sustained gains. This is exactly the demand signal that flows down to the companies supplying the data center build-out: Arista for networking, Vertiv for power and cooling, and Vistra for electricity. The thesis direction (long these suppliers) is coherent with the news catalyst. The fundamentals of the two supply-chain names are genuinely strong right now. Arista closed FY2025 with $9.0 billion of revenue up 28.6% year over year, a 42.8% operating margin, a 64.1% gross margin, and $4.3 billion of free cash flow — putting it in the 98th percentile of its sector on both margin and free cash flow. Its Q2 2026 quarter (period ended June 30, 2026) showed revenue accelerating again to $3.0 billion, up 12.1% sequentially, with net income up 18.6% and operating margin expanding to 45.4%. Vertiv, the direct beneficiary of cooling and power build-out, grew FY2025 revenue 27.7% to $10.2 billion with free cash flow of $1.9 billion (98th percentile in its sector) and a 33.8% return on equity; its Q2 2026 quarter showed revenue of $3.3 billion, up 23.6% sequentially, with operating margin improving to 19.5%. These are not fading suppliers; the numbers show accelerating orders and margins, not decelerating ones. The backtest evidence supports the momentum framework rather than just the narrative. On the daily timeframe over a 60-month window, the strategy on Arista produced a 48.1% return across 5 completed trades with an 80% win rate and a maximum drawdown of 17.9%. That is a limited sample, and the caveat matters: exits were filled on daily bars rather than intrabar data, so reported win rate and drawdown are coarse. But the trades that did complete align with the thesis window — the equity curve went essentially flat through 2021-2023 and then compounded sharply from mid-2024 onward, exactly when hyperscaler capex announcements began hitting the tape. Note also that Vistra and the shorter backtest windows did not produce completed results (an engine error on the MACD indicator for Arista), so the quantified edge is specific to Arista over…

Scores

• Conviction score breakdown: 51
• Thesis support: 72
• Trade readiness: 35
• Risk quality: 48
• Backtest evidence: 40
• Fundamentals trend: 62

Watch items

• VST — Close vs 50-day SMA
• VST — MACD line vs signal line
• VRT — Close vs 50-day SMA
• ANET — Close vs 50-day SMA
• ANET — Low vs nearest support
• ANET — Insider open-market activity
• VST — Next quarterly report