META Reports Tonight: The $145 Billion Question
The Bargo 2.5x tokenomics model says capex must rise. The model says the spending is clearing. Here's what the numbers say before the print.
META reports Q2 2026 earnings tonight after the close. The Street expects EPS of $7.13 on roughly $56.3B in revenue. But nobody cares about the quarter. The entire market is watching one number: capex.
META already raised its 2026 capex guidance once this year, from $115–135B to $125–145B. The question tonight is whether Mark Zuckerberg raises it again. The Bargo tokenomics model says the answer is probably yes, and here's why.
META has been sliding since mid-July, down from roughly $640 to $593. The 50-day SMA is rolling over. The stock enters earnings below its 50-day, RSI at 47, neutral, not washed out. The options market is long-gamma, meaning dealers pin the stock and dampen volatility. But the straddle is pricing in a big move.
The Bargo 2.5x Prediction Model: Explained
Before we get to the capex call, you need to understand the engine behind it. The Bargo tokenomics model is a deterministic structural model of the AI economy, published daily. It doesn't predict stock prices. It models the physical flows: how many tokens the world consumes, how much compute that requires, and whether the hyperscalers are building enough capacity to meet demand.
The model's name comes from its core finding: observed token demand is growing at 2.5x per year. That number is independently verified from company disclosures: Google's token volume grew 18.4x, OpenAI's API volume grew 5.1x, Gemini's monthly active users grew 2.1x. The model's capability-derived cross-check (using AI benchmark data completely independent of any revenue series) says the growth could be even faster at 3.2x. Either way, demand is doubling-plus every year.
The model has three layers that feed into a capex prediction:
Layer 1: Kappa (token efficiency). Kappa measures how much of the installed GPU fleet's theoretical capacity is actually being used for tokens. It started at 0.5% in early 2025 and has risen to 2.33% today. That gap, from 0.5% to 2.33%, is the demand story. As AI models get better and more people use them, more of the fleet lights up. At 2.33%, there is still enormous headroom before the fleet is saturated.
Layer 2: Supply-cover. The model compares projected token demand against projected GPU supply. Supply-cover is currently 1.17x, meaning demand is pressing right up against capacity. When supply-cover is tight, hyperscalers have to build more. When it loosens, they can slow down. Through 2027, the model projects supply-cover staying between 1.08x and 1.40x. That is tight enough to force continued spending.
Layer 3: The NVDA lead-lag. This is the model's most powerful predictive relationship. When you shift combined hyperscaler capex (Alphabet, Microsoft, Amazon, Meta) forward by three quarters and line it up against NVDA's quarterly revenue, the correlation is 0.99 across eight consecutive data points from mid-2023 through today. Every NVDA print telegraphed the hyperscaler capex print nine months later. NVDA's latest quarter came in at $81.6 billion. If the 0.99 relationship holds, that maps to roughly $210 billion in combined hyperscaler capex by Q4 2026. GOOGL already confirmed the direction last week with a raise to $195–205B.
Together, these three layers form the 2.5x model's capex prediction: if demand is growing at 2.5x, supply-cover is tight, and NVDA's 3-quarter lead with a 0.99 correlation says capex accelerates, then the hyperscalers must raise. They don't have a choice.
The CoreWeave Tell
In April, META signed a $21B deal with CoreWeave for external GPU capacity. If you're spending $21B on rented GPUs, your internal buildout is not keeping up with demand. That deal alone suggests the current $135B midpoint is too low.
The ROI Math
The B200 cloud TCO model calculates a 35.7% project IRR and 77.5% equity IRR at current rental rates. At $6.90/hr on-demand and $0.046 per million tokens inference cost, every dollar of capex deployed into AI infrastructure earns returns that would make any CFO approve more spending. The constraint is not willingness to spend. It's how fast you can physically build.
What the Model Expects
| Signal | Reading |
|---|---|
| Token demand growth | 2.5x/year |
| Supply-cover (Q4 2026) | 1.17x (tight) |
| NVDA 3Q lead-lag | 0.99 correlation, says capex accelerates now |
| GOOGL precedent | Already raised to $195–205B |
| B200 equity IRR | 77.5% |
| META current capex guide | $125–145B (midpoint $135B) |
| Model capex prediction | $140–155B midpoint |
The model assigns a 70–75% probability that META raises the 2026 capex midpoint from $135B to $140–155B tonight. The range would likely expand from $125–145B to $135–160B. If Zuck doesn't raise tonight, he will almost certainly signal an acceleration for 2027.
The Street
57 analysts cover META with a consensus Strong Buy rating and a mean price target of $825. That's 39% upside from $593. The low target is $664 (still +12% above current). The high is $1,015. No analyst has a Sell rating. The Street is positioned for this to work.
The Risk
The market is not uniformly bullish. Credit default swaps on Big Tech debt hit record highs. Hyperscaler capex now eats 98% of operating cash flow. META's free cash flow was $13.2B last quarter against $19B in capex. If META raises capex without delivering strong ad revenue growth, the stock will sell off, and the options market is pricing a big move.
But the 2.5x model says the demand is real. Supply-cover is tight. The inference revenue pool is scaling from $183B to $384B by year-end. The GPUs are earning their keep. If the model is right, a capex raise tonight is not a reason to sell. It's confirmation that the AI buildout is still in its early innings.
More research at bargo.ai/research.