On June 24, in front of a white goose and a field of golden eggs, Masayoshi Son wrote the AI trade as an equation. Market capitalization equals egg value plus goose value. ¥74 trillion of NAV, plus α. The vision is a quadrillion. The bridge is a letter.

Here is the problem with that plus sign. A few pages deeper, the same deck puts SoftBank’s market cap at ¥37 trillion. That is half of NAV. So at the exact moment the headline formula asks you to add a premium for the goose, the market is taking roughly fifty percent off the eggs. Son gave you the image. The market gave you the math.

SoftBank asks investors to add α. The market is still subtracting the discount.

This is not a scorecard. It is a reflection on how the story is told, because Masayoshi Son is the best corporate storyteller alive and the SoftBank AGM is the closest thing finance has to a keynote. The interesting question is not whether the deck passes a checklist. It is how a narrative this good can leave the one number that matters sitting just outside the frame. So watch the machine work.

How the story is told

It opens, as it always does, with a photograph. A young Son in front of a circuit board, pasted into a book, the founder reading his own origin back to himself. The device tells you, before a single figure, that this is not a results presentation. It is a chapter in a life’s argument, and you are being invited to believe in a person, not audit a balance sheet.

Then the thesis, built to be repeated. Artificial Super Intelligence, which Son sizes at ten thousand times human intelligence, on two legs. A brain computer, the model layer, anchored by OpenAI. And Physical ASI, intelligence with a body, robots for the factory, the disaster zone, the labor shortage, anchored by Arm silicon and the ABB Robotics acquisition. Underneath, the architecture: model plus chip plus infrastructure plus robot, SoftBank in the center as the self-described No.1 ASI Platform Provider. It is clean, it is memorable, and it hands every asset a seat in one sentence. As storytelling structure, it beats most IPO roadshows that had a year to prepare.

Then the escalation, which is the part everyone missed while screenshotting the bird. Son does not stop at “we own valuable eggs.” One slide opens the goose up to show clockwork inside and labels it the Golden Egg factory. Yes, that is a cutaway diagram of a goose in an investor deck. It is exactly as strange as it sounds, and it is also the most important rhetorical move in the room. The claim has quietly changed. It is no longer that SoftBank holds good assets. It is that SoftBank is the machine that manufactures good assets, on a loop, and a machine that keeps laying is worth more than the eggs on the floor. That gap, between the eggs and the factory, is the entire α. It is a genuinely good argument. It is also the one the story declines to price.

And when the bubble question came, Son did not reach for a discount rate. He called the doubt blasphemy against AI. That is the register the deck plays in from the first slide to the last. Not finance. Faith.

What the story does brilliantly, and what it leaves as a feeling

Give the narrative its due, because it earns it. It names its engine without flinching: Arm is the chip layer under everything, the toll road the whole AI build has to cross. It draws its flywheel on one page: lay an egg, sell an egg, fund the next clutch, feed the machine. Naming the engine and showing the loop are the two moves most investor stories fumble, and Son lands both in pictures you still remember on the train home.

And he does not only assert the goose, he dates it. The slides walk sixteen years: a NAV of three trillion yen back then, three golden eggs, and seventy four trillion now. The craft is in turning his own record into the argument. For the length of that slide, the premium is not a feeling. It is a history.

What the story will not do is carry that history forward into a figure. The record tells you the goose is real. It does not tell you what the next clutch is worth, and that is exactly where α sits. The vision-to-value line, the one sentence that turns ASI into cash per share, never quite arrives. The hierarchy a CFO wants, which business pays the bills today and which is a beautiful maybe, dissolves into one shade of ASI gold: SoftBank Corp and Arm royalties, the parts that actually make cash, are never set apart from OpenAI and the robots, the eggs that have not hatched. And the return on the next dollar, against OpenAI at its current mark, is left to faith. None of it is hidden. It is told as a feeling where a number was owed.

Where the story stops tying out

A number means what it is standing next to. On the headline slide, ¥74 trillion stands next to a plus sign and an α, and you are invited to add a premium for the bird. A few pages later, the same ¥74 trillion stands next to ¥37 trillion of market cap. SoftBank’s NAV is roughly double what the market will pay for it, and that gap is the audience’s reply to the story.

Run the sixteen years forward to the share price, and the argument cuts both ways. Back then, the market paid egg for egg: three trillion of NAV, three trillion of market cap. Today it pays half an egg. By the deck’s own arithmetic, the goose may have laid the eggs, but the market is not paying extra for the laying machine.

The market has reasons. The same factory that hatched Alibaba also hatched WeWork, and the NAV moves with every quarter’s marks. A low goose value is not only doubt. It is memory.

To be fair to the accounting, the minus is not missing from the books. SoftBank defines NAV correctly in the appendix, holdings less net debt, and it shows the debt. The minus is missing from the one place it travels: the emotional headline equation. “Market cap equals NAV plus goose” reads as a sum waiting to grow. It quietly drops the only operation the market is actually performing, which is subtraction.

The substance is in the appendix. The meme is in the meeting. Markets price the gap between the two.

How a CFO would tell the same story

The fix is not less story. It is the same story with the number inside it.

1. Put a figure on the goose. If α is real, frame it. The premium is the value of recycling eggs into new eggs, so show the realized return on recycled capital across the cycles. A goose with a track record can be priced. A letter cannot.

2. Keep the minus in the headline. NAV ¥74 trillion, less a fifty percent discount, equals ¥37 trillion, and here is what closes it. Say it on the slide everyone screenshots, not only in the appendix.

3. Separate the laying hens from the unhatched eggs. SoftBank Corp and Arm royalties make cash today. OpenAI and the robots are optionality. Rank them in one breath. Audiences discount what they cannot tell apart.

4. Underwrite the next dollar. At the current OpenAI mark, what is the next ¥6 trillion buying? A hurdle rate is a sentence. Conviction is not a number.

The Tie-Out

The goose is the best investor metaphor of the year, and it is also the tell. When the honest number is the problem, and here the honest number is the discount, you do not reach for a better number. You reach for a better image. Son reached for a bird.

Seventy four trillion of NAV. Thirty seven of market cap. The goose is the claimed premium above the eggs. The market has already told you what it thinks that premium is worth: nothing. The story is magnificent. It just stops at the edge of the one figure it exists to defend. The eggs tie out. The goose is still on credit.

The story serves the truth, or it does not tie out.

Sources. SoftBank Group, 46th Annual General Meeting business-strategy presentation, 66 pages, June 24 2026 · SoftBank Group Investor Relations. NAV of ¥74 trillion, market cap of ¥37 trillion, and the ¥1 quadrillion aspiration are SoftBank’s own figures from the deck. Primary sources, linked and not reproduced.

Watch the goose for yourself. SoftBank Group, 46th AGM presentation, all 66 pages.

Researched with AI, judged by a CFO.

The stock fell about 10 percent in extended trading.

A beat, and the stock falls. The explanation is not in the quarter. It is in the bridge from the quarter to the year. Cerebras disclosed the numbers. It did not disclose the story in a form an investor could use.

We read Cerebras at the IPO and found a profit that came from a one-time gain, growth that came from a single related party, and a backlog that came from one counterparty. The lesson then was that a number means what it’s standing next to. This quarter, the number that mattered was the guide, and the company left its explanation off the page.

The margin cliff

Start where the market started. Core gross margin this quarter was 47 percent. Cerebras guides next quarter to 36 to 38 percent, and the full year to 38 to 41 percent. Core operating margin was minus 2 percent this quarter, near breakeven, guided to minus 30 to minus 32 percent next quarter.

The best margin quarter of the year is the one already reported. The beat was backward-looking. The guide was not. A company can beat the quarter and still tell investors the quarter was the high point, and that is what happened here.

The explanation that wasn’t written down

The release gives no reason for the drop. The reason exists, but you had to be on the call to hear it.

Management explained that Cerebras is renting back data center capacity from an existing customer, and that the full rental cost is carried in the guide, alongside the cost of new capacity coming online. That is the driver of a ten-point gross-margin fall, and it appears in none of the written materials.

On the same call the CFO said margins will land higher than the company expected at the start of the year, citing pricing power and a rising price floor. Read carefully, that is measured against an earlier internal plan, not a promise that margin climbs from here. But notice the shape of it. The discouraging half of the margin story is written down and precise. The reassuring half is spoken, unquantified, and points the other way. A reader of the guide sees only the fall. A listener hears the hope. The two halves never meet on a page.

This is the disclosure failure, and it is the spine of the quarter. Not a hidden number. A missing narrative. The cause of the single most important negative in the print lived only in the audio.

The profit that wasn’t, again

Now the vocabulary, because it runs through everything else. Almost every comfortable figure in the release is a metric Cerebras calls “Core.” And the profit it led with is one more.

Cerebras reported Adjusted EBITDA of positive $12.7 million. Set it next to the line it is built from. GAAP loss from operations was $15.0 million. The positive figure appears only after adding back $18.2 million of depreciation and amortization and $9.6 million of stock-based compensation. This is not fake; the company reconciles every step. But it is not operating profit. Strip the addbacks and the business ran a loss.

This is the same shape we flagged at the IPO, with a new mechanism. Then the reported profit came from a one-time, non-cash gain. Now it comes from an addback stack. In both cases the headline is positive and the operating line is not.

A scorecard that drifts in the company’s favor

Look at how Core revenue is built. Start with GAAP revenue of $193.4 million. Subtract $4.1 million of pass-through. Add back $2.1 million of warrant amortization. Core revenue, $191.3 million.

That add-back is the tell. Cerebras grants customers warrants, and as they vest the value is booked as a discount against revenue, called contra-revenue. The charge was small this quarter, $2.1 million. But the warrant asset on the balance sheet jumped to $516 million, from $152 million at year end. That $516 million is queued to amortize against revenue for years. The charge grows from here.

So Core revenue is the line that adds back a cost set to get larger. As the warrants bite, Core revenue will print further and further above GAAP revenue, and the company will guide on the figure that flatters as the gap widens. (It is the early version of a move we have watched mature elsewhere: change the scorecard before the honest line turns down.) That is the metric drift the rubric exists to catch.

Cash is king, and this cash is borrowed

Operating cash flow was positive, $12.3 million. Set it next to capital spending of $132.0 million. Free cash flow was about negative $120 million. The $3.3 billion cash pile is real, and it did not come from operations. It came from $2.0 billion of financing in the quarter, a $1.0 billion preferred raise and a $1.0 billion working-capital loan from a customer.

That customer is OpenAI. The loan sits on the balance sheet at $983 million. OpenAI is the largest customer, the lender behind the cash, and a warrant holder, all at once. The $20 billion-plus, 750-megawatt agreement the company placed at the top of the release is the bull case and the concentration risk in a single sentence. At the IPO the dependence was on two UAE entities. The names have shifted toward OpenAI. The shape has not.

The cash is not a war chest at rest. Management is stepping up the pace of fundraising after the IPO and has named data center deployment as the priority use. Capacity is the stated limit on growth, so the negative free cash flow is the running cost of a buildout the company must keep financing to grow at all.

The deck that wasn’t, one quarter on

At the IPO we marked Cerebras down for communication. Everything was in the S-1, and almost nothing was presented. The first public quarter was the test of whether that changed. It did not.

The earnings slides are a title card, a legal disclaimer, a series of executive portraits, a single slide boasting the chip’s transistor count against an Nvidia package, three data tables that restate the press release line for line, and a Q&A slide. The things that actually move the thesis, why gross margin drops, where the second-half growth comes from, how fast the company can add capacity, the speed advantage over a GPU, were delivered out loud or in a live demo, and committed to no document an investor can later navigate.

The demo, by accounts, was genuinely striking, a real speed gap against a GPU. That is the point, not the counterpoint. The value is real and the company is impressive. It is the disclosure that is thin. When the substance of a public quarter lives only in ninety minutes of audio, the honest instruction to any investor is the one the company should have made unnecessary: listen to the webcast, because the written record will not tie the quarter out for you.

The CFO Rewrite

A cleaner version of this quarter was available, and it would not have cost a point of revenue. Lead with GAAP and let Core support it, not replace it. Put the margin guide-down on the page with its reason, the rented-back capacity, rather than leaving the cause to the audio. Name the warrant amortization path and what it does to revenue over the coming year. Give a forward GAAP frame, or explain the gap rather than declining to reconcile it. Answer the capacity and concentration questions you were always going to be asked. The numbers were strong enough to carry a cleaner explanation.

The Boardroom Test

Five questions a director should have asked before the release went out.

• Is the headline profit from operations? No. Adjusted EBITDA turns positive only after adding back depreciation and stock comp to a GAAP operating loss. Fail.

• Is the cash earned? No. Free cash flow was about negative $120 million. The cash balance grew on $2 billion of financing and a customer loan. Fail.

• Has the customer base broadened? Not yet. The risk factors name the same handful, and OpenAI is now customer, lender, and warrant holder at once. Fail.

• Was the historical quarter reconciled? Yes. The GAAP figures sit ahead of Core in the revenue line, and every non-GAAP measure is reconciled in the filing. Pass.

• Was the forward guide explainable from the written materials? No. The margin cliff is in the guide as a number, but the cause, rented-back capacity, came only on the call, and the concentration and capacity specifics were asked and declined. Fail.

One of five. The quarter it reconciled, it passes. The year it guided to, it cannot explain on paper. That is the same split we found at the IPO, one step further down the page.

The Tie-Out

Cerebras did not have a weak quarter. It had a strong quarter with a weak explanation.

Revenue nearly doubled. The product story is real. The speed advantage matters. The OpenAI agreement is large enough to define the company. But the public quarter asked investors to accept a new Core scorecard, a falling Core margin guide, a growing warrant adjustment, and a cash balance built with financing, while leaving the operational reason for the margin fall outside the written record.

That is why the stock fell after a beat. The market did not reject the quarter. It rejected the bridge from the quarter to the year. The picture below is six weeks of that arithmetic: the pop to $386, the slide, the recovery into the print, and the drop when the guide landed.

Beat the quarter, guide down the year. The story serves the truth, or it does not tie out. This quarter, the truth was in the guide, and half of it was never written down.

Sources. Cerebras Form 8-K and press release dated June 23, 2026, and the condensed consolidated financial statements and non-GAAP reconciliations therein. Q1 2026 earnings call, June 23, 2026. Consensus and extended-trading move per company filing and market data. Primary sources, linked and not reproduced.

Researched with AI, judged by a CFO.

That same week, a sophisticated seller agreed to sell at €300, and looked glad to do it.

One of those numbers was a market price. The other existed because almost nobody could trade the stock.

This is a story about which prices are real. The company is OHB, Germany’s national space champion, a listed satellite and defence group. The seller is KKR, which bought into OHB at €44 a share in 2023 and is now placing part of that stake at €300. The headlines call €300 a discount, because it sits 26% below the screen. The more useful question is why a seller sitting on close to seven times its money is happy to take it.

The screen price was never real

Start with what almost no one could do with OHB stock: trade it.

The Fuchs family controls about two thirds of the company. KKR held most of the rest. That left a free float of 5.7%. Roughly one share in eighteen was actually available to the market.

On that sliver, the shares ran up 234% this year, touched €688 in May, then gave back about 45% before the deal was priced. None of the move tied to cash generation. It tied to scarcity. When only five percent of a company trades, the last price is not a valuation. It is inventory.

A thin float behaves like a thin order book in any market. A handful of buyers meeting almost no sellers will print a high number, and that number looks like a price. It is not. It is what the marginal share cost when supply was artificially absent.

The placement did not create a discount. It exposed one.

Here is the part worth keeping.

Before the placement, OHB’s price was set by a market that barely existed. A few trades, a constrained supply, a number on a screen. A quote, with no market underneath it.

The placement changes the machine that makes the price. Instead of a few shares crossing on the open market, a syndicate runs a bookbuild. It calls institutions, collects real orders at real sizes, and finds the price that clears actual demand against a much larger supply. That is what price discovery is supposed to look like. It produced €300.

So the two numbers are not “price” and “discounted price.” They are “fiction” and “first real clearing price.” The placement did not knock 26% off OHB’s value. It revealed that the screen price had never been a market price to begin with.

This is not unique to OHB. Any time a stock carries a tiny float, a dominant shareholder, or a large secondary on the way, the same question applies. Are you looking at a market price, or just the last share that happened to trade? The distinction reaches far beyond German space stocks.

One caveat, because the story serves the truth. €300 is a large block clearing into a deadline, so it prints at the low end of what a market might bear. It is not proof of fair value. But it is built from real orders, and a price built from real orders beats a quote built from no supply, every time.

The seller is on a clock

Why the hurry, and why the willingness to take €300?

Because the shareholder agreement says it has to. KKR’s deal terms require it to cut its OHB stake by June 30. The block was first set to launch on June 12, then pulled when SpaceX’s own listing absorbed the institutional cash it needed. The demand disappeared. The contractual deadline did not. So the paper clears now, at the price a real book will bear.

And KKR does not feel the discount the way you would. It sells up to 1.23 million shares for up to €368 million and walks away with close to a 6.8x on a €44 entry.

A discount only stings if your cost basis is the screen. KKR’s cost basis is €44. Yours, if you bought the rally, is €405.

The irony is that none of this says much about OHB itself

Step away from the share register and the operating company looks strong.

A record order backlog of €3.35 billion, up 45%. Adjusted operating profit up 63%. A new €10 billion satellite communications joint venture with Rheinmetall. Output guided above €2 billion by 2028.

That is a real company with a real book. So hold two things at once. The business is doing well. The screen price was still fiction. Both are true, because a share price and a company are not the same object, and a thin float is exactly what pries them apart.

Separate the operating company from the share mechanics, and the placement stops reading as a warning about OHB. It reads as a correction to how OHB was priced.

What ties out

The enthusiasm narrative does not tie to the clearing price. “Space is hot, SpaceX proved it, pay up for exposure” is the story being sold into. The evidence is a 5.7% float, a 26% block discount, and a seller booking close to seven times on the way out. That is not a stock the smart money is fighting to own at €405. It is one the smart money is content to exit at €300.

The backlog does tie to something. If you like European space and defence at €300 and a genuine float, that is a thesis you can underwrite. If you are paying €405 because the screen said so and SpaceX is in the headlines, you are buying the quote, not the company.

Five moves to steal

1. Trust the market that clears over the market that prints. A bookbuild is real demand meeting real supply. The last tick on a 5% float is neither.

2. Check the float before you trust the multiple. Scarcity inflates a price the way it inflates any collectible. Cash flow does not care how many shares trade.

3. A forced seller’s discount is not your bargain. Ask the seller’s cost basis. KKR at €44 does not feel a 26% cut the way a buyer at €405 does.

4. Separate the operating company from the share mechanics. A discounted placement can sit on top of a record order book. Do not let the tape rewrite the business.

5. Read who sells and who holds. The outside investor takes the deadline. The family keeps control. Neither is selling you the enthusiasm.

The verdict

OHB is a great company, sold at an honest price, by an investor who had to sell.

Markets do not always lie. Sometimes they just have too little information.

A 5.7% float gave investors a quote. A 19% float finally gave them a market.

That is what tied out.

Sources. OHB SE investor relations (offering announcement, Q1 2026 figures) · KKR 2023 takeover documents (€44.00 offer price) · Reuters. Primary sources, linked and not reproduced.

Researched with AI, judged by a CFO.

The market just said something worse.

AI is real. It is not yet big enough to offset the slowdown in the core enterprise machine.

Stop at the income statement and the quarter looks fine. Revenue was $18.7 billion, up 6 percent. Operating margin was 17 percent, up 20 basis points. EPS was $3.80, up 9 percent. Free cash flow was $3.6 billion.

Then look at the order book.

For ten straight quarters, Accenture told investors exactly how much AI it was selling. It introduced an “advanced AI” bookings line in 2023 to size the prize and prove its lead. By the first quarter of this fiscal year the figure had reached $2.2 billion in a single quarter, about $1.1 billion of it revenue.

Then Accenture said that would be the last time it broke the number out.

The reason is fair on its face. AI is now embedded across nearly everything the firm does, so a standalone line tells less of the story. But watch the timing. The proof metric went dark just before the total book rolled over.

Because the book did roll over. New bookings came in at $19.3 billion, down 2 percent on the year and down 13 percent from the record $22.1 billion quarter right before it. A quarter ago the book was filling 1.2 times faster than it drained. This quarter, bookings barely cleared revenue at all.

And the part that softened is the part investors most want to keep. Consulting bookings actually beat. Managed services, the run-it-for-you book, is where it slipped, with management pointing to large deals pushed into fiscal 2027. The part investors most want to treat as durable suddenly looks less automatic. It is the read some analysts drew: clients may buy the AI strategy, then do more of the build in house.

So when Accenture says AI demand is strong, investors now have to work harder to prove it. The one line that would have shown whether AI was offsetting the core fade is the line the company just retired.

The income statement is the rear-view mirror.

Bookings are the windshield.

One says fine. The other says slower.

Tie-Out keeps running one move: read the number standing next to the headline number.

Cerebras was the clearest case, and the Tie-Out most of you read. The company reported a net profit of about $87 million. Almost all of it came from non-cash and non-operating items. Strip those out and the operating business had lost $146 million. The headline number was real. The number beside it was the truth.

Accenture is the same test on a much healthier company. The income statement is honest. The forward book sitting beside it is the warning. With Cerebras the catch hid inside the profit. With Accenture it sits next to a clean quarter.

Same discipline either way. Read the number standing next to the headline number.

The price said it faster than the release could.

Accenture fell almost 18 percent on the day, reported as its worst single session on record. It now sits near its 52-week low, down 56 percent in a year, from about $291 to $128.

That is not an earnings wobble.

That is a reset.

And the reset matters for AI, because Accenture sits as close to enterprise AI demand as any public company. It sees the budgets, the pilots, the data problems, the cloud migrations, the security reviews, the procurement cycles and the boardroom pressure.

So when Accenture says AI demand is strong while bookings weaken and guidance narrows, that is the tell.

The IPO story wants a clean chain.

• Better models.

• More demand.

• More revenue.

• Operating leverage.

• Public-market trust.

Accenture is showing the break.

Enterprise AI does not move from demo to value in a straight line. It has to pass through legacy systems, data readiness, governance, cybersecurity, compliance, operating-model redesign and people who still have to change how they work.

That is not a model problem.

It is an enterprise problem.

It is exactly where the AI story gets slow and expensive.

And here is the contradiction. AI matters more to this business every quarter. Accenture named the ecosystem it is building on: OpenAI, Anthropic, Nvidia, Palantir, Databricks and Snowflake. It logged 104 client bookings of $100 million or more this year, up 13 percent. It is spending $4.18 billion on cybersecurity acquisitions to stay in front of demand.

The strategy has never read cleaner.

But the proof has become harder to isolate, and the book is no longer accelerating.

For a private AI company, that is the warning label.

• A great model is not enterprise adoption.

• Adoption is not revenue.

• Revenue is not margin.

• Margin is not public-market trust.

Cerebras will have to prove that AI infrastructure becomes durable economics.

OpenAI and Anthropic will have to prove that frontier demand survives the cost structure.

Accenture just showed the customer side of the same problem. Even when demand is real, the enterprise still has to absorb it.

And absorption is slow.

That is the bottleneck.

• Not the chip.

• Not the model.

• Not the demo.

The enterprise.

Sources. Accenture Q3 FY2026 results and Form 8-K, quarter ended May 31, 2026 · Accenture Q1 FY2026 conference call transcript for the advanced-AI reporting change · closing price 18 June 2026. Primary sources, linked, not reproduced.

Researched with AI, judged by a CFO.

That number looks absurd. It also may not be the number that matters.

The Dutch FD headline gets the market reaction right. Experts are not shocked by the loss itself. They are asking whether OpenAI can finance the losses long enough to turn scale into economics.

The first tie out is accounting, and it cuts both ways.

OpenAI generated about $13.1 billion of revenue in 2025 against roughly $34 billion of cost, per audited figures reported by the Financial Times.

The $38.5 billion headline has a middle bar. A $41.55 billion non cash charge from the conversion toward a for profit structure. Real accounting. Not operating burn.

Strip it out and the bears lose their headline.

But the figure some circulate in the other direction, a loss closer to $8 billion, strips out too much. That is the bulls’ scoreboard, and it flatters.

The number that actually ties to operations sits between them. The operating loss was $20.9 billion, per the same reported figures.

That is the honest scoreboard.

Not the bears’ $38.5 billion.

Not the bulls’ $8 billion.

Nearly $21 billion of operations spending more than they earned.

Caption to paste under the image: OpenAI’s 2025 loss, three ways. The headline carries a $41.55B non-cash conversion charge; the cited $8B strips too much out. Operations lost $20.9B. Source: figures reported by the FT and The Information, not yet filed.

That does not make the economics good.

It just changes the question.

The real IPO test is not the headline loss. It is whether investors can isolate the operating burn, the compute cost curve, and the path to gross margin.

Because the IPO question was never whether OpenAI lost money.

Of course it did.

The question is whether public investors can underwrite the bridge from usage to pricing power, and from compute spending to durable returns.

OpenAI is already operating at extraordinary scale. Reuters, citing The Information, reported that OpenAI burned $3.7 billion in Q1 2026 while generating $5.7 billion of revenue. The burn profile is still heavy. But revenue is no longer theoretical, and that is why the story is investable at all.

The second tie out is capital structure.

OpenAI closed a $122 billion funding round at an $852 billion post money valuation, according to its own announcement in March.

That is not a normal venture round.

That is pre public infrastructure financing.

The third tie out is dependence.

Microsoft has invested more than $13 billion in OpenAI, and Reuters reported that the two companies capped OpenAI’s revenue sharing payments to Microsoft at $38 billion.

The cap matters because IPO investors will want to know who really captures the economics.

• OpenAI.

• Microsoft.

• The cloud providers.

• The chip suppliers.

Or the capital stack underneath all of it.

One caveat on all of this: these figures come from leaked audited documents and shareholder reporting, not from a public S 1. OpenAI’s IPO filing was confidential. Treat the numbers as reported, not yet disclosed.

So the loss number is not the verdict.

It is the entrance exam.

For an AI IPO, the market will need five disclosures before it can price the story honestly:

• Revenue quality.

• Gross margin after compute.

• Customer concentration.

• Capital commitments.

The path from usage growth to free cash flow.

Until then, OpenAI is not a normal software IPO.

It is a financing vehicle for the AI infrastructure race.

That may still be worth a trillion dollars.

But it has to tie out.

SpaceX shares jumped today after the company disclosed it is officially acquiring Anysphere, the artificial intelligence startup behind Cursor, for $60B in stock. While the headline number is significant, the transaction itself was not entirely new news. SpaceX’s S1 had already highlighted an April arrangement giving the company the option to acquire Cursor for $60B or pay $10B if it did not proceed.

That makes today’s move the third IPO catalyst in quick succession: first the IPO itself, then Elon Musk’s Sunday tweets around SpaceX’s AI ambitions, and now the formal Cursor acquisition announcement. The stock reaction suggests investors are still repricing the AI story, even though the Cursor transaction had already been signposted in the IPO materials. By midday Tuesday the stock was up about 10 percent and roughly 55 percent above the $135 IPO price in three sessions, briefly making SpaceX the world’s fourth most valuable company, ahead of Microsoft.

Cursor has emerged as one of the leading AI coding agent platforms, with more than 1M developers using its products. Its tools are used by major technology and enterprise customers, including Nvidia, Adobe, Figma, and OpenAI. More than half of Cursor’s revenue is believed to come from corporate customers, driven by both new enterprise adoption and expansion among existing accounts.

The strategic rationale is straightforward: Cursor gives SpaceX an enterprise AI distribution layer and a strong developer product, while SpaceX brings large scale compute infrastructure, including its Colossus training supercomputer. Together, the combination strengthens SpaceX’s position in AI coding and knowledge work, where Cursor competes with OpenAI, Anthropic, Google, and others.

The deal also comes amid rapid growth at Cursor. In November 2025, the company raised about $2.3B at a $29.3B post money valuation in a Series D round backed by existing investors including Accel, Thrive, and Andreessen Horowitz, alongside new investors such as Coatue, Nvidia, and Google. By March, Cursor had reportedly surpassed $2B in annual revenue run rate, reflecting strong corporate demand.

The transaction significantly expands Elon Musk’s technology platform into enterprise AI and could support SpaceX’s longer term AI revenue ambitions. Goldman Sachs recently forecast that SpaceX’s AI related revenue could increase 100 times by 2030, with Cursor expected to contribute meaningfully to that growth.

What there is, is a 300-page S-1. The key disclosures are in it. But the filing leaves the reader to assemble the story themselves.

That isn’t dishonesty. It’s a different kind of investor-communication problem.

When SpaceX went public, there was a story to inspect: the S-1, a fact sheet, the roadshow deck and a CFO-narrated video, all in one controlled place (spacexipo.com). You could disagree with the framing, but you could find it.

With Cerebras, the filing is the story. And when the filing leads with the flattering numbers, those numbers become the narrative by default. It’s why, asked to summarise Cerebras last week, an AI told me the company “turned profitable on revenue up 76%,” fluent, sourced, and wrong in the two places that decide whether either number is real. (The profit that wasn’t.)

Here’s the rule underneath all of it: a number means what it’s standing next to. A profit means one thing beside an operating profit and another beside an operating loss. Growth means one thing beside a broad customer base and another beside related-party concentration. A backlog means one thing from the open market and another from a customer who also bankrolls it. Cerebras disclosed every one of those qualifiers. It just didn’t put them in the frame. Three numbers show what that costs.

First, what’s on the table

Cerebras priced its IPO on May 13 at $185 a share, above its marketed range, and raised about $5.55 billion — the largest US tech listing in years. It opened the next morning at $350, touched $386 intraday, and closed its first day up 68%.

Then the numbers the listing was built on. FY2025 revenue was $510 million, up 76% (from $290.3 million); the bottom line was a $237.8 million GAAP profit; and the figure now doing the heavy lifting is a $24.6 billion backlog, roughly 48 times that revenue. A profit, a growth rate, a backlog. Take each one and look at what it was presented beside.

The profit, without the operating line

A $237.8 million GAAP profit reads as a company comfortably in the black — a swing from a $481.6 million net loss the year before. The operating business tells the opposite story: a $145.9 million operating loss, wider than 2024’s $101.4 million. The entire distance between the two is non-operating: chiefly a one-time, non-cash gain of about $363 million from extinguishing a forward-contract liability tied to G42’s original investment. It’s an accounting event, not a year of selling chips. Strip it out, as the company does in its own non-GAAP measure, and 2025 was a $75.7 million net loss.

None of that is concealed; it’s all on the income statement. But the presentation lets “net income, positive” stand on its own and a positive bottom line is what gets repeated, by an AI summariser, a headline, a reader in a hurry. What would tie it out is one line, in frame: operating result → the non-cash gain → net income. Shown together, fewer people mistake a liability revaluation for a business turning profitable. Shown apart, almost everyone does.

The growth, without the customer

“Revenue up 76%” reads like product-market fit. The qualifier sits in the notes: around 86% of 2025 sales came from two related UAE entities — MBZUAI at 62% and G42 at 24% — and one of them alone accounted for 77.9% of receivables. US-billed revenue actually shrank year-over-year. The prior year’s healthy-looking operating cash was largely a prepayment from that same group — the customer funding its own orders.

Growth and the source of growth belong in the same frame. “Grew 76%” and “86% from two related parties” aren’t two facts; they’re one fact told in halves. Put the growth rate beside revenue-by-relationship and you see what the number is: not a market forming, but a relationship deepening.

The backlog, without the counterparty

This is the one that matters most at today’s price, because it’s the $24.6 billion backlog, not the $510 million of revenue, that the valuation is leaning on. Framed simply as “backlog,” it reads as demand.

Read it next to its counterparty and it becomes a different asset. The majority of that backlog is tied to one customer: OpenAI, under a multi-year agreement for 750 megawatts of compute through 2028. And OpenAI is not an ordinary arm’s-length buyer: it is also advancing Cerebras roughly $1 billion and receiving warrants. That makes it a customer, a financier and a prospective owner.

A backlog from the open market and a backlog from a customer also bankrolling the supplier are not the same quality of asset. Only one of them tells you demand is truly independent. The disclosure says which one this is. The framing, “$24.6 billion backlog”, does not.

To be fair to the forward story: AWS and other channels may broaden the customer base. That’s exactly the development that would change this picture, if it shows up in the revenue rather than the narrative.

The CFO Rewrite

If I had twenty-four hours before this listing, this is the slide I’d insist on.

One slide, three panels, every qualifier in frame: the operating-to-net-income bridge; revenue by customer relationship; backlog by counterparty. Not a single new disclosure: everything on it is already in the S-1. Just the same numbers, set beside the things that decide what they mean. It costs the story its flattering reading, and that’s the point: a number presented with its qualifier in view is a number an investor can trust.

The Boardroom Test

Take an independent director’s seat and ask four questions:

• Is the profit operating profit? No, a one-time non-cash gain sitting on a $146 million operating loss. Fail.

• Is the growth broad-based? No, 86% from two related parties. Fail.

• Is the backlog independent demand? No, mostly one customer, who is also lender and prospective owner. Fail.

• Is everything an investor needs disclosed? Yes, it’s all in the S-1. Pass.

One of four. And even the pass carries an asterisk: it’s all disclosed, but only inside a 300-page filing. No deck, no fact sheet, no investor-grade communication around it. Disclosure passes; communication doesn’t.

The price is doing the framing now

Here’s where the three threads tie together and they don’t need me to tie them. Since the first trade, the market has done the reframing the presentation didn’t: it put the operating line, the customer and the counterparty back beside the headline, and repriced. The stock walked from a $386 first-day high back toward the $185 it was sold at. As of early June it was trading in the low-to-mid $200s, a band of roughly $205–$240 since analyst coverage began. At around $220 it’s still up on the offer price and well down from where it closed day one and which of those numbers you quote depends, once again, entirely on what you stand it next to.

That isn’t the market calling Cerebras a bad company. It’s the market doing a CFO’s job in public: pricing in the qualifiers the framing left out.

What the first earnings call will settle

Cerebras hasn’t reported as a public company yet and that call, not the S-1, is where management first frames the story in its own words. Three things to listen for: whether they bridge net income through the operating line themselves; whether they give revenue by customer relationship unprompted; and whether the backlog is presented beside its counterparty. If those show up, the framing was launch-week roughness. If they don’t, it was the strategy. Let’s wait for June 23 when Cerebras ties out their Q1 results.

The Tie-Out

The tie-out is simple. The profit only ties out beside the operating line. The growth only ties out beside the customer base. The backlog only ties out beside the counterparty. Cerebras disclosed those qualifiers; it didn’t frame them. The gap, between a complete filing and clear investor communication, is where the market has stepped in.

The lesson goes beyond Cerebras. Read every headline number beside the thing that gives it meaning: profit beside operations, growth beside concentration, backlog beside counterparty quality. Put the qualifier back in frame, and the story either survives the move or it doesn’t.

With Cerebras, for now, the tape is doing the work.

What’s Next at Tie-Out

AI IPO Tie-Out

✓ SpaceX
✓ Cerebras IPO Review
→ Cerebras Q1 Results: Did the story improve?
→ OpenAI IPO Filing
→ Anthropic IPO Readiness
→ Perplexity’s path to 2028

The CFO Deck Lab

→ Investor deck best practices
→ Award-winning annual reports
→ CFO storytelling frameworks

In Development

→ Annual Report Benchmark Project (US vs Europe)
→ Investor Relations Scorecards

Tie-Out reads below the line. Researched with AI, judged by a CFO.

Most investor presentations start with the business and end with the vision. SpaceX does the opposite. And that, it turns out, is exactly why it works — and exactly where it leaves a CFO wanting more.

So this isn’t a review of the deck; plenty of those are coming. It’s something you can reuse: a five-part test for whether any investor story ties out to actual value — run, live, on the most anticipated IPO of the decade. SpaceX is the case study. The framework is the point. Next time it’s an earnings deck, an activist presentation, or the next AI listing, the five tests don’t change.

The Tie-Out Framework

A great investor story makes you believe. A great investor case shows you what to do with that belief. Narrative gets investors into the room; capital allocation gets them to sign. Five things tell you whether a story closes that gap — whether it ties out:

1. The vision-to-value bridge — does the deck explain, in one line, how the mission becomes cash flow?

2. Capital allocation — if they raise another dollar, do you know where it goes and what it’s expected to earn?

3. A clear business hierarchy — which business makes the cash today, which buys optionality, and in what order?

4. A named economic engine — is there one business, clearly identified, that powers the rest?

5. The flywheel — is the value-creation loop drawn on a single page?

Here is how SpaceX’s story scores as it heads to market:

The Tie-Out Framework, scored on SpaceX — 1 of 5.

I’ve rarely seen a wider gap between the quality of a narrative and the quality of the value-creation bridge beneath it. On the five things that tie a story to value, almost everything is still “partial.” That may sound contradictory. It isn’t: great stories create belief, and great investor communication explains how belief becomes cash. The A+ for storytelling and the B- for investor communication are those two jobs, scored apart.

First, what’s on the table

SpaceX is attempting the largest IPO in history — raising about $75 billion at a valuation near $1.75 trillion, with demand reportedly running about twice the shares on offer. It prices June 11 and trades June 12 on Nasdaq as SPCX.

What’s unusual is how they’re communicating. Rather than bolt an IR tab onto spacex.com, SpaceX built a separate, controlled destination — spacexipo.com — with everything in one place: the S-1 filing, a plain-English fact sheet / investor Q&A, the full roadshow deck, and a seventeen-minute video narrated by CFO Bret Johnsen. A company that builds a bespoke narrative destination instead of a compliance page is telling you, before you read a number, that it intends to control the story. So I read all of it the way a CFO reads a board pack.

Why the deck earns the A+

Two ways to sequence a deck. SpaceX puts the mission first and the financials last.

A normal deck marches through market, product, customers, financials, outlook. SpaceX runs mission → problem → technology → economic engine → future civilisation, and saves the financials for the appendix. That isn’t a presentation structure. It’s a movie script — challenge, obstacles, solution, destination.

Put plainly: most companies ask investors to understand the business before they’ll believe the mission. SpaceX asks investors to believe the mission before they understand the business. That single inversion is the whole deck, and most investor-relations teams never dare to try it.

It works because every business reinforces one line — everything lowers the cost of becoming a spacefaring civilisation — so a sprawling conglomerate reads as a single idea. It sells inevitability the way Amazon did with the cloud and Nvidia did with AI, widening the frame until you’re thinking in decades, not quarters. And it leans on pictures over spreadsheets, because a reusable booster landing builds more conviction than another EBITDA bridge. On storytelling, it is genuinely world-class.

Where the story doesn’t tie out yet

The grade drops to a B- on the five tests above — not because anything is hidden, but because the deck assumes investors will connect technology to cash flow on their own. Public markets aren’t that generous.

The bridge (test 1) is missing. The deck explains the mission, the technology, and the opportunity. It never quite explains the economic architecture that turns ambition into shareholder return. There’s no one-line answer to how does this become cash flow? So here is the slide I’d hand them — the mission on one side, compounding cash on the other, and the four mechanisms that carry one across to the other:

The vision-to-value bridge — the four mechanisms that carry the mission to cash.

Capital allocation (test 2) is thin — for a $75 billion raise. The more ambitious a company, the more its capital-allocation story matters, not less, because that’s where storytelling meets accountability. SpaceX asks investors to back many futures at once — launch, internet, AI, the Moon, Mars. The deck shows segment revenue, profitability and capex, but not the framework that says this is how we choose between them, and this is what we expect each to earn.

Hierarchy and the engine (tests 3 and 4). Here’s the one the deck gets half-right. The financials make clear that connectivity — Starlink — is the profit centre that powers the rest; on test 4, the engine is visible in the numbers. But the narrative treats Starlink as one pillar of three rather than the load-bearing one. Every great corporate story needs a named engine: Amazon had AWS, Apple the iPhone, Microsoft enterprise software. Starlink is the business that turns a visionary engineering company into something investors can actually model — and it deserves to be the hero, not a peer.

The CFO Rewrite

If I had twenty-four hours before the roadshow, this is the one slide I would insist on adding.

If the bridge above is the map, this is its engine. The flywheel is the loop that makes the whole thing turn — and if I could add only one slide, it’s this one.

Suggested caption: The CFO Rewrite — the one slide I’d add.

No addressable-market chart. One picture. Starlink, the mainspring, generates the cash. The cash funds Starship and AI. Starship lowers the cost of launch. Lower cost expands the opportunity. The opportunity lifts demand. Demand grows Starlink — and the loop closes.

The deck already gestures at this with a “repeatable business model” list. But a list reads as strategy; a loop reads as a machine. Investors love flywheels because they make complicated businesses look simple — and the best companies run complicated businesses. This one slide is the bridge between the A+ and the B-: it answers test 1 (the value bridge) and test 5 (the loop) in a single frame, and it costs the narrative nothing.

The Boardroom Test

Strip away the theatre and put yourself in an independent director’s seat. Management asks you to approve another $10 billion of investment. Can you answer four questions?

• Which business makes the cash today? Yes — Starlink, clearly. Pass.

• Where does the next $10 billion go? Not from this deck. Fail.

• What return is it expected to earn? No hurdle, no segment ROIC. Fail.

• Which dollar is cash generation, and which is optionality? Discernible — connectivity earns, AI and Mars are bets. Pass.

Two of four. That isn’t a verdict on SpaceX; it’s the test you can run on any company that asks you to fund a future — Nvidia, Tesla, the next AI listing, an activist’s slide deck. Same four questions, every time.

The Tie-Out

The lesson was never about rockets. It’s that vision attracts capital, and capital allocation keeps it — and the best investor presentations do both. SpaceX has mastered the first half as well as any company in a generation. The second half is the next evolution of its investor communication, and the five tests above are where it starts.

Every great investor deck makes a promise. The best ones explain how that promise becomes cash. SpaceX has written one of the most compelling promises in corporate history; the next decade will decide whether the economics are as extraordinary as the narrative.

That’s the tie-out. Because in the end investors don’t fund spreadsheets — they fund futures, and the spreadsheet only decides whether the future is believable. When SpaceX prices next week, we run the same five tests against the number, and we find out.

Tie-Out reads below the line. Researched with AI, judged by a CFO. Next — “SpaceX priced. Did the story tie out to a number?” — the same five tests, re-scored once it lists.