MAG 7 valuation report

Microsoft's 27% stake in OpenAI Group PBC was officially disclosed at $135 billion when OpenAI restructured in October 2025 at a $500B valuation. Then in March-April 2026 OpenAI raised $122B at a $852B post-money valuation — which mathematically marks Microsoft's stake at roughly $230B (a 17.6x paper return on $13B invested, $11.6B of which is actually deployed). Microsoft carries this at cost. If OpenAI IPOs in late 2026 or 2027 — which Reuters and CFO Sarah Friar have signaled is plausible — that hidden value would move onto the balance sheet at market. For context: $230B is roughly 7% of Microsoft's current $3.14T market cap, sitting off-book today.

In the October 2025 restructuring, Microsoft gave up its cloud right-of-first-refusal — the bear case headline. What actually happened: OpenAI signed a binding $250B incremental Azure contract, Microsoft's IP rights extended through 2032 (from 2030) and now cover post-AGI models, and Microsoft won the right to pursue AGI independently. Pre-deal, Microsoft needed OpenAI to succeed. Post-deal, Microsoft gets paid whether OpenAI succeeds, stumbles, or gets disrupted — because OpenAI is now a locked-in cash customer, a 27% equity position, and a licensed IP source Microsoft can build on alone.

Reality Labs has lost $83.6 billion cumulatively since 2020, including $19.19B in 2025 alone (Q4 2025: $6.02B loss on just $955M of revenue). The market treats all of Meta as one business and multiples it accordingly. Here's the asymmetry: Meta's 2025 total operating income was $83.3B — meaning the Family of Apps (core advertising business) actually generated roughly $102.5B of operating income standalone before Reality Labs wiped $19B off the top. At the current $1.74T market cap, the core business is trading at approximately 21x its standalone earnings — reasonable for a business growing revenue at 22%. Zuckerberg told analysts 2026 Reality Labs losses will be "similar to 2025 levels" and "likely be the peak" before gradually declining. Every dollar Reality Labs stops bleeding flows straight to consolidated earnings.

Quest VR is a rounding error. Ray-Ban Meta smart glasses — developed with EssilorLuxottica — have sold 2+ million units, tripled YoY, and the $799 Display model sold out store inventory in 48 hours at launch. EssilorLuxottica is expanding capacity to 10M units per year by end of 2026. Meta cut 1,500 Reality Labs staff in January and slashed the VR budget 30% to fund the glasses business. If glasses become the next smartphone-scale platform — Zuckerberg's explicit bet — Meta owns the operating system and the silicon. The bear case assumes Reality Labs is a permanent money pit. The pivot already happened; the market just hasn't noticed the re-allocation inside the line item.

At GTC 2026, Jensen disclosed that the top 5 hyperscalers represent 60% of Nvidia's revenue — and the remaining 40% spans sovereign clouds, enterprise, industrial, robotics, and edge. The bear case assumes hyperscaler concentration is 80%+ and that their in-house silicon efforts (Trainium, TPU, Maia, MTIA) will eat Nvidia's lunch. The disclosed 40% non-hyperscaler revenue is the fastest-growing segment and is structurally unaddressable by custom silicon: sovereign AI buyers like Saudi Arabia's Level42 ($4.8B) don't design chips, they buy national capacity. Huang also said on Dwarkesh's podcast that even inside the hyperscaler 60%, most of that compute serves external customers — AWS renting to enterprises — not the cloud's internal workloads. Meaning the chunk that's actually at risk of in-house replacement is much smaller than the concentration headline suggests.

Nvidia has committed up to $100B to OpenAI, owns roughly 9–13% of CoreWeave after doubling down with another $2B in January 2026 (post-deal stake value ~$4.7B), backstops $6.3B of CoreWeave's excess capacity through 2032, committed $2B to xAI, and invested $2B each into Lumentum, Coherent, and Marvell. Critics call this "circular financing" and draw dot-com parallels. The less obvious read: Nvidia's equity stakes give portfolio companies access to Nvidia's AA- credit rating, letting OpenAI and CoreWeave borrow at 6–9% instead of the 15% they'd pay standalone. Each dollar Nvidia invests in a portfolio company creates multiples of forward chip demand backed by cheaper capital. This is vendor financing at industrial scale — the same playbook GE used to dominate jet engines for 50 years.

The forward PS model in the valuation table yields $182 — below the current $202 price. That's the DCF's quiet warning: at $216B TTM revenue, sustaining even 10x price-to-sales requires FY27 to hit $367B (70% growth on a $5T company). Consensus says yes, Morningstar's $240 FVE says yes, but the margin for error narrows with every doubling. The bull case works. The bear case — where Rubin ramp slows or hyperscaler silicon captures even 15% share — is also mathematically intact. This is why the blended FV is only $222 vs. Morningstar's $240: the PS model is dragging the average down, and that tension is honest math, not a bug.