These days, Meta may be the only trillion-dollar company that can swing 10% or 20% in a single day as if it were routine. It took just one quarter for Meta to fall from market darling to market castoff. After its third-quarter earnings, investors’ panic appeared only just to have been digested. More recently, as expectations for rate cuts returned, Meta rebounded 11% from its lows.

The narrative first cast Meta as an “AI darling”: the company had the broadest application scenarios, and AI investment would come with accelerating revenue. Then came a third-quarter guidance mix of heavier investment and slower revenue growth, which quickly recast it as a wasteful “AI spendthrift.”

But can Meta really be safely chased higher? “Notably larger than 2025” capex is not just a throwaway line. The company also reiterated in a smaller meeting that future depreciation expenses will rise sharply.

So in 2026, how should investors think about the opportunities and risks for Meta, a company that combines a powerful social moat with a seemingly manic willingness to spend? Dolphin starts with Meta’s two recent debt financings to lay out our latest view.

The detailed analysis follows:

Cash Flow Shock: Meaningful, but Not the Key Valuation Killer

After its Q3 earnings, Meta drew intense market attention with two successive financing moves: a direct $30 billion bond issuance and the much-discussed $27.3 billion “off-balance-sheet financing.” Clearly, Meta has a near-term cash flow problem.

Dolphin’s rough estimate is that if all of Meta’s more than $120 billion in 2026 capex were funded by 2026 operating cash flow, a company with a market value in the trillions would see its annual operating cash flow almost wiped out.

Of course, the giants will not sit around waiting for cash flow to go to zero. They have plenty of ways to address the problem. Aside from reducing shareholder returns, which is effectively a form of implicit equity financing, on- and off-balance-sheet debt financing has become standard practice.

Do Not Count on Shareholder Returns

In “Tencent: Starting With Stingy Buybacks, Is Super AI an Obvious Strategic Play?” we previously compared the differences between Meta’s and Tencent’s cash flow management. Meta follows a “cash neutral” approach: cash inflows come from profits generated by its core business, while cash outflows mainly go toward reinvestment in its own operations, including software, hardware and fixed assets, shareholder returns through buybacks and dividends, external investments, and other items such as tax withholding.

Since the second half of last year, however, rising capex outflows have rapidly depleted Meta’s cash on hand. By the end of Q3, its balance of cash plus short-term investments, mainly marketable securities, had fallen to just $44 billion, down nearly 40% from a year earlier.

Put another way, this year’s $70 billion in capex has consumed not only newly generated operating cash flow, but also more than $30 billion of existing cash. Including Q4, the total draw on existing cash will only be larger.

Looking directly at the cash flow breakdown below, net cash flow was negative in both the first and second quarters, reflecting the use of existing cash reserves. As “software and hardware investment,” which includes capex, increased quarter by quarter, the main areas squeezed were shareholder returns and external investments. By the third quarter, both had already fallen to very low levels, leaving little room for further compression.

So if next year’s capex rises by another $50 billion on top of this year’s $70 billion, how can Meta’s current cash balance of just over $40 billion be enough?

It Ultimately Relies on Meta-Guaranteed “Off-Balance-Sheet Financing”

That is why, immediately after reporting Q3 earnings, Meta issued $30 billion in corporate bonds in six tranches, with interest rates ranging from 4.2% to 5.8%. The size was three times the unsecured note issuance it completed in Q3 last year.

In addition to direct bond issuance, Meta completed a special financing before releasing its Q3 results. By setting up a separate project company through a joint venture with financial institution Blue Owl, it moved financing for data center infrastructure construction off its own balance sheet.

Under the joint venture structure, Meta holds only 20% of the equity, so it does not consolidate the joint venture’s financial statements. Instead, it records the equity stake as a long-term equity investment.

In substance, however, Meta is acting as the lessee and signing a lease with a “second landlord”: Laidley first leases the data center campus and then subleases it to Meta, as shown below.

Although the leased assets appear mainly to be cabinets, the subsequent chip deployment, campus property management, and long-term power operations are all Meta’s responsibility. Blue Owl is effectively serving as an intermediary that helps organize the financing.

The specific structure is shown below:

Meta’s Perspective: High Financing Cost = Off-Balance-Sheet Liabilities + Less Near-Term Cash Flow Pressure

On the surface, the data center property does not belong to Meta, and Meta is not the issuer of the $27.3 billion in notes.

But as the tenant, Meta provided a residual value guarantee in this transaction: during the first 16 years of the data center’s operating period, if Meta terminates the lease early, it must make up the difference between the remaining value of the data center and the agreed value, which was initially $28 billion.

So while the structure appears circuitous, it is effectively equivalent to the following:

During the first 16 years, Meta issues a $28 billion corporate bond to Blue Owl to fund data center construction. The interest paid each period is equivalent to annual rent. Under the lease terms for the first four years, for example, the annual rent is about $3 billion, implying a coupon rate of 3/28, or 10.7%.

That coupon rate looks very high, far above the interest rates on Meta’s corporate bonds issued half a month earlier. But by sacrificing future cash flow, Meta gets the following in return:

A mismatch between rights and obligations: Meta can immediately get funding and start construction, but the first $3.1 billion interest payment does not begin until 2029, four years later. Blue Owl bears the note interest during the first four years.

Lower pressure on current accounting profit: although Meta is the dominant operator of the data center in terms of actual responsibilities, the transaction is still recognized as an operating lease rather than a finance lease for accounting purposes, thanks to the deliberately uneven equity split and the absence of an arrangement for Meta to take back ownership.

A finance lease can also ease near-term cash flow pressure, but under U.S. GAAP, the current-period expense recognized for a finance lease, including depreciation and interest expense, is often significantly higher in the early years than the rent actually paid.

As we estimate in the chart below, assuming rent remains unchanged and the discount rate is 4.5%, using operating lease accounting from 2029 to 2036 means the expense recorded in opex will always be lower than under finance lease accounting, potentially reducing Meta’s near-term profit pressure.

Blue Owl and Creditors’ Perspective: A Nearly Risk-Free Deal

Meta is trying to make its statements look better, while Blue Owl, a pure financial institution, is getting paid with little effort. In this financing partnership, Blue Owl does not take on the dirty work of campus operations or management. Aside from fronting $1.8 billion in annual interest during the first four years, it faces almost no risk or additional cost.

After 2029, Blue Owl collects $3.1 billion in rent from Meta with one hand and pays $1.8 billion in interest with the other, pocketing a $1.3 billion net interest spread. If Meta stops midway, the $27.3 billion bond principal is also guaranteed by Meta.

Using a 4.5% risk-free rate as the discount rate, suppose that in 2044, the final year of Meta’s promised 16-year residual value guarantee, Meta chooses to exit. In reality, this would be loss-making for Meta and would not be rational, but the assumption is useful for calculating Blue Owl’s investment return.

For Blue Owl, Meta’s $28 billion principal guarantee would be used directly to redeem the $27.3 billion in corporate bonds. Over the 19-year period, the discounted total of the net interest spread would be $7.1 billion, implying a net investment return of 1.4% per year, equal to 7.1/27.3/19. In practice, however, this is achieved with almost no capital committed by Blue Owl.

For PIMCO and BlackRock, the deal is even sweeter: they lock in more than 20 years of 6.58% annual returns, 1 to 2 percentage points higher than buying Meta’s bonds directly. In a rate-cut cycle, that is a solid investment.

Declining Profit Efficiency: The Real Illness Suppressing Valuation

Beyond cash flow, the difference between operating leases and finance leases shows that Meta is also worried about the impact of AI investment on profit. Dolphin believes this is exactly the key factor that will suppress Meta’s valuation performance in 2026.

Depreciation Pressure Is Cumulative

By the third quarter, Meta’s capex had already reached 38% of current-period revenue, rising very quickly. Roughly speaking, if this investment ratio is maintained over the next year, and in reality it is likely to be higher, with the market expecting capex to reach 50% of revenue, and if the incremental spending is mainly on servers, then based on Meta’s current average server depreciation period of 5.5 years, this investment would pressure margins by as much as 7 percentage points per year, equal to 38% divided by 5.5, assuming revenue cannot effectively absorb the depreciation amount.

Even if some of this is replacement of old equipment, compared with the previous investment scale of less than 20% of revenue, it would still add at least 3 percentage points of margin pressure. In reality, however, the value of replaced old servers is far from enough to offset the steady stream of new investment. The book value of Meta’s “property and equipment” has been rising for four straight years.

Moreover, each year’s new investment brings cumulative depreciation cost pressure. Combining this with market expectations, Dolphin makes the following simple projection:

Assume 2026 revenue growth of 18%, and a 15% revenue CAGR from 2025 to 2030 under AI enablement.

Based on the current data center construction plan, Meta’s investment cost pressure will peak in 2026. It is expected to add 4 GW of computing power by 2028, half of which will come in 2026. Based on NVIDIA’s total deployment cost of about 50 billion yuan per GW, the company would need to invest an additional $100 billion in 2026 alone on the surface.

But given the current competitive environment in computing power, assume Meta uses TPUs in part, as rumors suggest: renting Google’s TPU cloud services first in 2026 and then directly buying TPU chips in 2027 to deploy in its own data centers. That would cause significant near-term fluctuations in capex.

Another possibility is that Meta negotiates an 80% discount with NVIDIA, reducing the cost to $160 billion. Given that there is less information about TPU cooperation and that technical adaptation is involved, Dolphin first calculates based on discounted NVIDIA GPU costs.

After subtracting the $29 billion in off-balance-sheet financing from Hyperion, meaning the portion provided by Blue Owl, the $160 billion total still leaves a huge $130 billion funding need, requiring another $50 billion to $60 billion in annual investment.

For other parts, including leased external computing power, land, and buildings, we assume a stable growth rate of 15% to 40%, depending on the pace of data center deployment, front-loaded and then slowing, with growth slightly above the long-term revenue CAGR of 15%. The expected capex changes are as follows. Note that the chart below does not include any sudden new data center deployment plans by Meta, though computing power shortages could still arise along the way.

Under Dolphin’s assumptions, no matter how later plans after 2028 change, AI investment will create very clear dual pressure on profit at least in 2026 and 2027. By our estimate, capex as a share of total revenue will rise to 50% in 2026 before gradually declining.

The impact of depreciation expenses on margin erosion will peak in 2027, dragging margins down by 3 percentage points in one year, before gradually easing after 2028. Upside surprises could come from a sharp drop in computing power costs, or from Meta making deeper concessions on operating expenses, such as workforce optimization. Margin erosion could bottom earlier than Dolphin expects, but the former would clearly be the more efficient lever.

Conversely, if Meta cannot cut costs, it can only hope to generate new revenue, such as a breakthrough from a general-purpose agent in Meta AI. But competition in general-purpose agents is also fierce. Although Mark Zuckerberg was repeatedly reported in the first half to be poaching talent at any cost, it has to be admitted that Meta’s Llama family of large models has lagged badly.

AI’s empowerment of advertising can indeed help improve ROI and encourage merchants to raise ad budgets, which in essence squeezes share from smaller platforms. But in terms of monetization efficiency, the “utilities” costs of the AI era cannot quickly be driven down to bargain levels.

At the same time, AI vastly expands the scope of user interaction, causing platform operating costs to rise exponentially. The internet-era effect of declining marginal costs is very small in the AI era. Relying on advertising alone, Meta will find it hard to restore its previous monetization efficiency.

At least in our model, Meta adds only 1 GW of new data center computing power in 2030, yet depreciation costs still grow 17%. If Meta wants to lift OPM back from 36% in 2027 to more than 40% in 2025 before then, 2030 revenue would need to reach $436 billion, more than double the sub-$200 billion level in 2025. In an interesting coincidence, the extra $200 billion-plus is also exactly OpenAI’s own revenue growth target for 2030.

Is Meta Being Wasteful? Not Fatal as Long as Competition Holds Up

After its near-term margin collapse, Meta quickly fell from the AI darling that everyone was bullish on at the start of the year, backed by the strongest social moat, to the bottom of the Magnificent Seven in a 2022 déjà vu. Google, which no one wanted at 19x P/E, has instead replaced the once-more-attractive Meta at $900 and NVIDIA, which rose to become the world’s most valuable company, as the “new AI king” with the cleanest logic right now.

In the AI era, the wheel turns quickly. The fact that narratives can swing from one extreme to the other shows precisely that more change is still ahead. Watch the tech giants take turns on stage, and do not rush to conclusions.

Meta now trades at 21x P/E, below its historical valuation midpoint. Its stock has recently found support and rebounded around 20x P/E as rate-cut expectations warmed up across the market, while cooperation with Google TPUs eased concerns that Meta was falling into a vicious AI capex bubble.

However, given the profit collapse discussed above, Dolphin believes that at least in 2026, this is not yet the time to be blindly bullish. Historically, Meta’s absolute bottom was 7x P/E in 2022. Although Dolphin’s operating profit expectation for 2026 has also fallen to single-digit growth, as we said in our Q3 earnings review, this is by no means a replay of the “2022 nightmare.” The key is that Meta’s core competitive logic has not suffered a fundamental flaw.

OpenAI may look aggressive, but Google Search has demonstrated the importance and underlying value of ecosystem control and entrance-point mindshare. To form a highly sticky consumer entry point, OpenAI still needs to lower barriers and keep cultivating its user ecosystem.

The market, however, has been sober for a while and has begun to lose faith in AI dreams that cannot be implemented. Even if capital is willing to dream, industrial heavyweights will not allow risks to be shifted onto them. In this equation, power supply and TSMC capacity are the choke points for AI.

Without the natural advantage of stable cash flow, how can a company win scheduling priority from industrial heavyweights? That makes OpenAI’s more important task right now finding a deep-pocketed backer to provide open-ended support for a trillion-dollar Stargate.

Meta therefore still has time, but it also faces trial-and-error risk. A detailed valuation analysis has been published in the article of the same title in the “Updates - Research” section of the Longbridge app.

Finally, Dolphin emphasizes again: Meta’s return will require more patience. In 2026, Meta’s opportunity lies in a recovery after falling below its floor value, not in an upside breakout during a burst of euphoria.