What SpaceX's AI Valuation Teaches About Ad Bidding Claims

The lesson starts with two pairs of numbers that should not be swallowed whole. SpaceX’s IPO valuation is reported at $1.77 trillion; Morningstar’s fair-value estimate is $780 billion. That is a $990 billion spread, and the useful question is not whether the optimistic side is impossible. It is what part of the number is independently observable, and what part asks the market to believe future AI economics before they show up in operating results [1].

The ad-platform version is familiar: Performance Max Target ROAS bidding is reported at 38% higher ROAS versus manual CPC in a Q1 2026 cross-industry average, and Meta Advantage+ Shopping is reported at 4.52x ROAS versus 3.70x for manual campaigns, a 22% improvement [2][3]. Those figures may describe real aggregate outcomes. They still do not tell a buyer whether the next dollar in a specific account should move from a controlled manual structure into a black-box campaign.

That is the structural parallel, not a claim that SpaceX and ad platforms run the same kind of business. Both cases put outsiders in front of an AI performance story produced or promoted by parties that benefit when the story is accepted. In finance, that story can support a trillion-dollar valuation. In paid media, it can redirect budget, rewrite targets, and leave the next person in the account explaining why platform-reported ROAS did not reconcile to booked revenue.

The SpaceX gap is a source problem before it is a valuation problem

The most important thing about the SpaceX valuation dispute is not that one analyst is more conservative than another. It is that the largest part of the upside depends on AI revenue that has not yet become ordinary, auditable operating performance. The reported total addressable market attached to SpaceX reaches $28.5 trillion, with 93% attributed to AI, implying roughly $26.5 trillion of potential tied to that future AI business [1].

Morningstar’s skepticism is not just a smaller spreadsheet. Forbes reported that Morningstar assigned a 43% probability to an “$81B+ capital destruction” scenario for the orbital AI data center plan [4]. That does not prove the plan will fail. It does mean the valuation cannot be treated as if the AI path is already de-risked.

Then there is the underwriter issue. Goldman Sachs projects $322 billion in SpaceX AI revenue by 2030, roughly 100x the current $3.2 billion revenue base cited in S-1-related materials, while also serving as an IPO underwriter [5][6]. The conflict is structural. A bank can produce a serious model and still have a financial role that makes the model’s assumptions worth inspecting harder.

The operating base makes the promise/fact split visible. The AI segment is reported at $3.2 billion in revenue, $6.4 billion in operating loss, and $12.7 billion in capex [6]. Losses that exceed revenue do not make a future AI business worthless. They do, however, force the denominator into the room: how much capital must be spent, over what period, to produce the revenue curve the valuation needs?

There is also a segmentation trap. The S-1-related data groups SpaceX’s AI segment and X’s advertising business in ways that should not be casually treated as one clean proof point for ads, Grok-related revenue, and orbital data centers all performing alike [6]. When a segment is broad, a headline number can be true and still be too blunt for the question being asked.

Why a platform lift is not an account forecast

Performance Max already manages more than 80% of enterprise ad spend, up from 55% in 2024, according to the cited 2026 market coverage [2]. That adoption matters because the question is no longer whether buyers will encounter AI bidding. Most already have. The question is whether the platform’s aggregate lift number survives contact with their own conversion data.

A 38% higher ROAS claim for PMax Target ROAS bidding is useful only after the buyer knows the baseline. Manual CPC against what structure? Same conversion action? Same product feed? Same geos? Same branded demand exposure? Same budget level? Same seasonality window? A manual campaign that carried prospecting, low-margin SKUs, or weaker landing pages is not a neutral control group for an automated campaign that can lean into easier demand.

That is where platform automation can be genuinely valuable and still overstated in the deck. Clean conversion data, enough budget, and a stable value signal can let bidding systems find pockets of efficient demand faster than a human operator can. The audit problem is not that AI bidding never works. It is that the reported lift has to be separated from spend reallocation, attribution expansion, and conversion-quality drift.

For a broader operating map of where automated bidding tends to help and where it breaks, the useful companion read is AI PPC automation. For the finance-side translation, the better standard is the one used in AI advertising ROI: prove the revenue, not the interface metric.

The Advantage+ number has the same denominator problem

Meta Advantage+ Shopping’s reported 4.52x ROAS versus 3.70x for manual campaigns is a cleaner-sounding number than most platform claims because the comparison is easy to repeat in a slide: 22% better [3]. But the same cited coverage also reports platform-wide Meta Ads ROAS across all industries at 1.86x in 2025 [2]. Those numbers can both be true. One is a specific automated shopping benchmark; the other is a broad platform average across industries and campaign types.

The mistake is importing the better number into an account where the category, offer, margin, audience saturation, and conversion window do not resemble the aggregate. A buyer selling replenishable beauty products to warm audiences should not inherit the same expectation as a buyer trying to generate qualified B2B demos with a six-month sales cycle. The lift claim is a starting hypothesis, not a budget justification.

What to check before accepting the lift

The practical audit starts before the test launches. If the manual baseline is weak, old, underfunded, or structurally different, the comparison will be easy for automation to win and hard for finance to trust. The buyer needs a record that survives after the test owner has moved on.

• Match the conversion action. Do not compare an AI campaign optimizing to purchases against a manual campaign optimized to add-to-cart, leads, or mixed events.
• Match the economic goal. ROAS should be checked against margin, cancellation rate, refunds, sales-qualified lead rate, or downstream revenue where those metrics matter.
• Match the demand pool. Separate brand, non-brand, retargeting, prospecting, existing customers, and new customers as far as the platform and backend data allow.
• Match the time window. Conversion lag can make an automated campaign look stronger or weaker depending on when the report is pulled.
• Match the product or offer mix. If the AI campaign shifts spend toward bestsellers, high-intent queries, or lower-discount products, that is a spend-allocation result, not automatically a bidding-efficiency result.

The uncomfortable part is that spend reallocation can be useful. If PMax finds profitable demand in a feed that a manual structure ignored, the buyer should not reject that because the platform’s slide was too confident. But the account should record what changed: which campaigns lost budget, which products gained impressions, whether branded or returning-customer demand increased, and whether backend revenue followed the same curve as platform-reported conversions.

This is also where metric gaming becomes less theoretical. If the system can improve the reported number by selecting easier conversions, the buyer needs a counterweight outside the platform. The same verification instinct behind Alex Karp’s tokenmaxxing warning applies here: an optimized metric is not the same as an optimized business outcome.

The AI ad boom depends on definitions too

Market-size numbers have the same denominator problem. One 2026 AI ad market figure is $57 billion, while a narrower definition puts it at $32 billion [3][2]. That gap is not a rounding issue. It reflects different decisions about what counts as “AI advertising”: AI-generated creative, automated bidding, ads adjacent to AI Overviews, chatbot inventory, or conventional paid search touched by machine learning.

That distinction matters because more than 80% of US AI ad spend is reported to run through traditional paid search listings alongside AI Overviews, not inside chatbots [3]. In other words, many search teams are already managing AI-labeled exposure without treating it as a separate exotic channel. The label changed faster than the measurement workflow.

The early ChatGPT ads data is a useful caution, provided it is dated properly. During the January–March 2026 pilot, ChatGPT ads generated a 0.91% CTR versus 6.4% on Google Search, and fewer than 20% of eligible users saw ads daily [7]. That does not tell buyers what chatbot ads will look like later. It does say that “AI ad inventory” should not be assumed to mean performance inventory.

For that measurement problem, the relevant operating issues are covered more directly in Why ChatGPT Ads Isn’t Working for Performance Advertisers and ChatGPT’s Data Privacy Design Blocks Ad Measurement. CTR is only one surface metric; if identity, attribution, and conversion feedback are limited, a buyer cannot repair the business case with a better headline rate.

Use the analyst questions inside the ad account

The SpaceX dispute gives media buyers a useful habit because the questions transfer cleanly: who benefits from belief, what is observable, what is projected, and what denominator controls the conclusion. Those questions are not anti-AI. They are the minimum standard for letting an opaque system influence capital allocation.

In campaign terms, that means the platform-reported number should sit next to an account-level reconciliation file. The file does not need to be elegant. It needs to show spend before and after the change, conversion definitions, excluded campaigns, baseline quality, attribution windows, backend revenue, customer quality, and where budget moved. If the lift survives that file, it is more useful than a generic benchmark. If it does not, the benchmark was never the account’s result.

That same audit reflex is why the broader verification cases matter. Microsoft’s Scout controversy is useful when a platform claim outruns inspectable evidence. Waymo’s transparency premium is useful when the question is how much more trust a system earns by exposing more of its operating record. Neither case gives a buyer a campaign answer by itself. Both reinforce the same standard: claims become more valuable when outsiders can inspect how they were produced.

The budget rule that follows is plain. Do not inherit a 38% or 22% AI bidding lift as an account expectation. Enter it as a claim to be reconciled against real spend, real customers, and real revenue. If analysts can challenge a trillion-dollar AI valuation gap by asking who benefits, what is observable, and what is projected, media buyers can ask the same questions before platform-reported lift steers the next budget move.

References

1. SpaceX Thinks Its AI Business Has $26.5 Trillion in Potential — Yahoo Finance
2. AI Ad Spending Will Reach $32 Billion In 2026 — Forbes
3. AI-powered ad spend will hit $57 billion in 2026 — eMarketer
4. SpaceX's AI Wing Casts Cloud Over IPO — Forbes
5. SpaceX IPO Faces a Credibility Test Over Goldman's AI Revenue Model — Investing.com
6. X Platform Statistics 2026: SpaceX S-1 Filing Data — Digital Applied
7. ChatGPT Ads Pilot had a Rocky Start — The Keyword