AKAM: The AI Contract Is Huge. So Is the Cash Bill.

The 60-second read

Five things to know

1. The cash bill arrived. Q3 operating cash of $433.0M was overtaken by $605.2M of cash capex, producing negative $172.1M FCF.
2. The contract predates this print. The $11.6B Anthropic agreement was disclosed in September; Q3 adds explicit 2027 CIS and 2028 company revenue targets.
3. Cash and accrual capex switched places. Cash spending exceeded accrual additions by $141.0M this quarter. Neither is a substitute for a complete investment schedule.
4. The warrant has two economic effects. Potential exercise can dilute shares, while its grant-date fair value is recognized as a future reduction of customer revenue.
5. The price and the call changed. The prior Q2 SELL used $81 value at $110.54. The current HOLD uses $75 at $83.09, with a wide $30–$120 scenario range.

Three scenarios, one probability-weighted number

Author assumptions, not issuer forecasts; values rounded to usable references.

Signposts: what would change our mind

These are analyst monitoring thresholds, not company promises or confirmed earnings dates. Passing one threshold does not automatically change the rating.

The price reset

Akamai closed November 6 at $83.09, up 6.74% from the preceding $77.84 close. The market rewarded the updated growth outlook, but the stock remains approximately 24.8% below the $110.54 reference in our Q2 episode. Those are different comparisons. The first describes one session around the earnings release; the second describes the change since our prior research. Neither return establishes what the business is worth.

The Q2 call was SELL with conviction four out of five and an $81 base value. This quarter moves to HOLD, conviction three, with a lower $75 base. There is no contradiction in becoming less negative on a stock while reducing its estimated value: the share price fell faster than the modeled value. The change also recognizes a much larger cloud opportunity, but only after reserving for the substantial cash investment required to deliver it.

A former high is not a valuation anchor. The saved quote shows a 52-week range of $77.37 to $165.45 and a market capitalization of $11.941B. We use the current quote date consistently and do not characterize the last trade as a live executable price. The historical reaction chart uses a fixed next-session-close versus release-date-close convention. It describes price behavior, not a causal event study controlling for market and sector moves.

Source: Saved FMP quote and price history. Supporting statements and calculations are retained in the episode dataset.

The current print and the outlook

Revenue was $1,121.961M, up 6.4% from $1,054.630M a year earlier. The company rounds reported growth to 6% and constant-currency growth to 7%. GAAP diluted EPS fell to $0.77 from $0.97, while adjusted EPS fell to $1.77 from $1.86. Both earnings measures declined despite higher revenue. The adjusted figure is useful for comparisons, but it is not interchangeable with GAAP profit or cash available to owners.

The saved FMP earnings row contains a $1,117.943M revenue estimate and $1.67 adjusted EPS estimate. Actuals exceed those figures by a computed 0.36% and 5.99%. These are provider estimates retrieved after the event, not an independently frozen pre-announcement analyst survey. A small revenue beat and an EPS beat can coexist with year-over-year profit pressure and a lower annual earnings guide.

Q4 revenue guidance is $1.145B to $1.170B, adjusted EPS $1.50 to $1.70 and adjusted operating margin 24%. Full-year revenue guidance is $4.440B to $4.465B and adjusted EPS $6.47 to $6.68. The annual EPS midpoint of $6.575 is 2.2% below the preceding $6.725 midpoint and 7.7% below 2025 adjusted EPS of $7.12. This is a new reduction in the near-term earnings path, not evidence that the announced cloud contract has already repaired profitability.

Management now projects 2027 CIS revenue of $1.1B to $1.3B, total company constant-currency growth of 17% to 22%, and 2028 revenue above $7B. These are forward-looking issuer targets. We do not apply the constant-currency range directly as a guaranteed reported-dollar growth rate. Capacity delivery, customer demand, exchange rates and contract accounting all influence the reported result.

Source: Q3 release, financial statements and outlook. Supporting statements and calculations are retained in the episode dataset.

The franchise funding the expansion

Security remains the largest business: $621.892M of Q3 revenue, 55.4% of company sales, up 9.4% year over year. Delivery and other cloud applications contributed $391.266M, down 3.4%, while CIS contributed $108.803M, up 34.0%. CIS is strategically important, but it is still only 9.7% of current quarterly revenue. A large future contract should not erase the need to monitor the security business that supports today’s economics.

Security added $53.455M of revenue versus the prior-year quarter, CIS added $27.624M and delivery plus other applications lost $13.748M. The growth engines more than offset the shrinking category in revenue dollars. That arithmetic says nothing by itself about incremental profit. Company gross profit actually declined by $7.827M, because the cost of revenue rose faster than sales.

The reporting categories changed in 2026. Earlier releases show security, delivery and cloud computing; current releases separate CIS while including other cloud applications with delivery. The dataset retains each historical original table and its definition. The nine-quarter total revenue series remains comparable, but we do not splice the old cloud-computing numbers into a continuous CIS growth chart. Current comparative figures are taken from the current release.

Akamai reports one operating segment and does not disclose a standalone GAAP operating-profit series for each solution category. We therefore show solution revenue and consolidated margins, rather than inventing category profit. Our sum-of-parts multiples are explicitly assumptions about economic differences, not evidence that the company disclosed a cloud or security segment margin. That distinction matters when infrastructure utilization and shared costs are changing rapidly.

Source: Solution revenue and comparative definitions. Supporting statements and calculations are retained in the episode dataset.

What reaches shareholders

GAAP operating income was $108.721M, a computed 9.7% margin, versus $166.024M and 15.7% a year earlier. Adjusted operating income was $285.654M, a 25.5% margin, down from $321.985M and 30.5%. The rounded release margins are 10% and 25% for the current quarter. Precision should explain the calculation, not imply that our numbers are more authoritative than the filed statements.

The $176.933M operating adjustment includes $131.917M of stock compensation, $26.331M of acquired-intangible amortization, $16.028M of amortized capitalized stock compensation and interest, $2.166M restructuring and $0.491M acquisition costs. Stock compensation is a real economic cost even when it does not consume current operating cash. We keep it in the economic valuation rather than treating all adjusted profit as freely distributable.

Below the operating line, $41.719M of interest and marketable-securities income outweighed $10.630M interest expense and $1.376M other expense. Pretax income was $138.434M; the $21.330M tax provision left $117.104M net income. Investment income is meaningful because the debt proceeds created a large pool of liquid assets. It should not be confused with operating profitability from cloud capacity that has not yet been built.

The adjusted EPS denominator is also different. GAAP diluted shares were 151.166M, while adjusted diluted shares were 147.717M after recognizing 3.449M shares of note-hedge benefit. The company explains why the hedges economically offset convertible dilution even though GAAP recognition differs. We show both denominators and use a separate 152M model assumption. Calling every difference between GAAP and adjusted EPS a profit adjustment would miss this share-count effect.

Source: GAAP and non-GAAP reconciliations. Supporting statements and calculations are retained in the episode dataset.

Nine checks on earnings quality

Operating cash flow of $433.037M was strong in absolute terms and only modestly below $441.832M a year earlier. But cash capex rose to $605.155M from $195.016M. Our simple free-cash-flow definition, operating cash less property, equipment and capitalized software cash payments, therefore moves from positive $246.816M to negative $172.118M. The current cash problem is reinvestment intensity, not an absence of operating cash generation.

The cash-versus-accrual distinction reverses the direction emphasized in our Q2 research. This quarter cash capex exceeds accrual capex of $464.146M by $141.009M. In Q2 cash capex was $225.753M while accrual additions were $346.536M. Payment timing moves costs between periods; it does not eliminate them. Nine-month cash capex of $1.022755B is now close to accrual additions of $1.017019B, leaving only $49.056M of cash FCF for the nine months.

Accounts receivable reached $958.463M. The ending-balance receivable-days proxy is 76.9 days, calculated as receivables divided by quarterly revenue times ninety. It is not an average-balance collection measure. Inventory days are not meaningful for this presentation because the consolidated balance sheet does not separately identify an ordinary inventory line; prepaid components and future capacity commitments require reading the notes instead.

Current disclosure controls were effective at September 30. Historical material weaknesses were remediated and are retained as history, not treated as a permanent prohibition on coverage. The earnings-quality conclusion rests on the current statements: recurring compensation adjustments, greater investment intensity, denominator differences and a lower near-term guide. None alone proves misstatement or misconduct. Together they explain why a low adjusted P/E is an incomplete description of shareholder economics.

Source: Q3 cash flows, accounting policies and controls. Supporting statements and calculations are retained in the episode dataset.

The contract and its funding burden

The Anthropic agreement was signed in September and disclosed in the September 24 8-K. It is not a newly signed November earnings surprise. Two project plans represent approximately $11.6B of aggregate payments over initial seven-year terms beginning on their respective service start dates. Delivery and availability conditions apply, as do specified termination rights. Dividing the headline by seven is only an average, not the contractual annual revenue schedule.

The accompanying issuer announcement estimates approximately $5.5B of associated capital expenditures, including approximately $1.7B of additional 2026 spending to secure components such as memory. The 8-K describes a Jabil request authorizing approximately $1.7B of memory purchases, with payment on supplier receipt and consignment arrangements before use. This is a tangible investment obligation supporting future service capacity. It is not an operating expense that can simply be subtracted from the contract headline to calculate profit.

At September 30, cash and current plus noncurrent marketable securities totaled $4.238579B. Convertible-note carrying value totaled $7.567687B, leaving $3.329108B of net debt on our stated convention. Carrying value includes accounting deductions and differs from principal repayment requirements. The model uses the carrying value consistently; refinancing and maturity analysis must also consider contractual principal and the note hedges. Operating leases remain inside cash-flow assumptions rather than being deducted a second time from enterprise value.

This quarter illustrates why backlog, liquidity and profit must stay separate. The commitment can improve confidence in demand while increasing the amount of cash at risk before revenue arrives. Existing liquid assets provide funding flexibility, but much of that liquidity came from financing. A failure to deliver capacity, a customer payment dispute, hardware cost inflation or a material outage could delay revenues after capital has already been committed. The filing explicitly identifies increasing Anthropic concentration as a future risk.

Source: September 24 agreement filing. Supporting statements and calculations are retained in the episode dataset.

Source: September 24 issuer capex announcement. Supporting statements and calculations are retained in the episode dataset.

Three ways to value the transition

Our base fair value is $75, with very high uncertainty and a $30 to $120 scenario range. Three independently structured routes cluster around $72 to $81. They are cross-checks sharing assumptions, not statistically independent observations. The conclusion is therefore less certain than the closeness of the three outputs might suggest. We require a chosen twenty-five percent discount to the base, or $56.25, before considering a new purchase while the operating case remains intact.

The enterprise DCF uses assumed revenues of $5.3B, $7.0B, $7.6B, $8.2B and $8.8B for 2027 through 2031. The first cash bucket combines remaining Q4 2026 and FY2027 at negative $1.2B; subsequent annual economic cash flows are negative $0.3B, then positive $0.7B, $1.15B and $1.5B. The combined opening bucket assumes negative $1.3B in remaining 2026 and positive $0.1B in 2027, explicitly retaining the late-2026 component funding burden. It is discounted 1.25 years from the September balance-sheet date; later buckets follow annually. Revenue figures remain calendar-year assumptions, so the opening cash bucket is not a comparable annual cash margin. These cash flows include recurring compensation as an economic cost, capital spending, working capital and operating occupancy costs. They are analyst assumptions, not management cash guidance. The early cash deficits reserve for the build; we do not subtract the entire $5.5B commitment again after discounting those flows.

At a ten percent discount rate and three percent perpetual growth, enterprise value is $14.265B. Subtracting $3.329B of model net debt and dividing by 152M shares gives $71.94. Roughly 93.8% of enterprise value comes from the terminal value. This is a highly duration-sensitive result. Future maintenance spending, replacement cycles and mature utilization matter far more than a few cents of this quarter’s adjusted EPS.

The sum-of-parts cross-check assumes 2027 security revenue of $2.75B at five times sales, delivery and other applications of $1.45B at one times, and CIS of $1.10B at four times. That produces $19.6B before a separate $4.0B future-build reserve. After the same net debt and shares, value is $80.73. The reserve is a rough capital adjustment in this route only, not a second deduction from DCF. These sales multiples are analyst choices, not observed transaction prices or disclosed solution profitability.

The owner-earnings cross-check assumes normalized 2028 EPS of $6 after recurring compensation, applies fifteen times earnings and discounts two years at ten percent, giving $74.38. It is not fifteen times the company’s adjusted EPS guidance. The market’s approximately 12.6 times current adjusted guide looks inexpensive, but that denominator excludes compensation and does not capture the near-term cash build. Our base needs material improvement in mature economics even though the current adjusted multiple is low.

The price-implied calculation fixes the early DCF cash flows and solves for the final-year and terminal cash flow together. At $83.09, the required 2031 economic cash flow is about $1.68B, above our $1.50B base. This is an inference from our chosen model, not proof that all investors hold the same forecast. The sensitivity grid shows how modest discount-rate and growth changes can overwhelm near-term earnings differences.

Source: Balance-sheet inputs; forecast assumptions computed in dataset. Supporting statements and calculations are retained in the episode dataset.

Our call versus Wall Street

The provider aggregate rating is Hold, comprising two strong buys, twenty-four buys, twenty-five holds and one sell. Its aggregate target is $144, with a low of $85 and a high of $225. This is a collection of analyst marks from different dates, not a fresh unanimous valuation of the latest print. The dispersion and timestamp mix are more informative than treating the average as an objective destination.

Post-release November 6 target entries include FBN at $120, KeyBanc at $175, RBC at $85, UBS at $90 and Evercore at $150. The first four entries are described as target reductions in the saved provider headlines; Evercore reiterated its rating. We report the numerical targets and dates as provider evidence, rather than claiming access to the firms’ complete proprietary models. Their differing assumptions about capital intensity, cloud margins and valuation horizon can explain wide disagreement.

Our HOLD label matches the aggregate rating but our $75 base is materially lower than the aggregate target. Rating alignment does not mean price-target alignment. The model explicitly carries early negative economic cash flow and a meaningful capital reserve. The optimistic counterargument is that these reserves are excessive, the security franchise deserves a higher multiple and contracted cloud demand makes future cash flows more durable than our discount rate implies.

The saved 2028 consensus revenue estimate is approximately $6.467B, below management’s projection above $7B. That mismatch may reflect different forecast dates or skepticism about execution. We do not quietly replace the provider series with management’s target. The dataset retains annual revenue, EBIT, net income, EPS and analyst counts so a later update can identify whether the gap closes through estimate revisions or through changed company expectations.

Source: FMP ratings, target news and annual estimate snapshots. Supporting statements and calculations are retained in the episode dataset.

Management and the customer warrant

The September warrant gives Anthropic the right to purchase up to 387,051 nonvoting Series B preferred shares, each initially convertible into twenty common shares under the specified transfer conditions. The full common-equivalent amount is 7.741M shares. The exercise price of $2,226.60 per preferred share equals $111.33 per common-equivalent share. These are not free shares handed over at the current $83.09 price, and no warrant shares had vested or been exercised at September 30.

Forty percent of the warrant, equivalent to 3.096M common shares, is tied to the first specified customer payment. The remaining tranches require additional contractual commitments in $3B increments, totaling another $9B. We do not count that optional expansion as signed base-case revenue. The company recorded $149.1M of grant-date fair value for the first tranche it expected to vest, with corresponding prepaid and other assets.

The accounting effect is unusually important: as related services are delivered, the asset is amortized as a reduction of revenue. Potential share dilution and reduced reported revenue are separate effects. The model reserves shares and includes contract economics in its cash assumptions; it does not subtract the entire warrant fair value as a new cash payment. At share prices above the exercise price, a full dilution illustration must also recognize cash exercise proceeds. Counting all shares without proceeds would overstate the simple dilution effect.

Capital allocation also deserves scrutiny. The company spent $615.744M on buybacks in the first nine months, all before Q3, while expanding debt-funded liquidity and infrastructure commitments. Buybacks can be sensible below intrinsic value, but they compete with funding flexibility. The current quarter had no common repurchases. Management should now be judged on delivery milestones, cash collection, capital discipline and security retention, not only the size of the headline contract.

Our previous Q2 concerns remain relevant but require updating. Security growth stayed near its prior pace, adjusted margin declined further, cash spending caught up with accrued investment, and the annual earnings midpoint fell again. The prior under-$75 reconsideration level was a conditional research reference, not an automatic buy order. The enlarged investment program changes the evidence required before acting on a lower price.

Source: Equity, compensation, warrant and controls notes. Supporting statements and calculations are retained in the episode dataset.

The strongest bull and bear cases

The strongest bull case is that Akamai already has the operational infrastructure and customer credibility needed to turn contracted AI demand into a larger recurring cash stream. Security provides an established base, the delivery decline is moderating, and the enlarged cloud relationship may improve utilization and purchasing scale. If mature cash economics arrive faster, the market could reasonably assign a higher value than our conservative base.

The strongest bear case is not that the contract is imaginary. It is that a real contract can still earn an inadequate return after hardware spending, power, maintenance, customer incentives, financing and execution risk. Revenue concentrated in one expanding customer is less diversified than the same revenue spread across many security clients. The contract’s termination and service conditions make delivery quality central to cash realization.

Our judgment sits between those cases. We credit substantial future growth and eventual cash recovery, but decline to capitalize the gross commitment as if it were guaranteed profit. The low adjusted earnings multiple helps explain why a blanket bearish call is less compelling at this price. The wide distribution of possible cash outcomes explains why the same price does not yet offer our preferred margin of safety.

There are additional accounting and competitive risks. Capitalized software and infrastructure useful lives affect the timing of expense recognition. A solution mix shift can make old margins poor guides to new capacity returns. Competitors may reduce prices, customers can optimize workloads, and energy or memory costs can rise before contracts are repriced. Our forecast is deliberately falsifiable through cash spending, service starts, reported revenue and the customer-warrant reconciliation.

Source: Risk factors and concentration disclosure. Supporting statements and calculations are retained in the episode dataset.

What to check next

The next useful catalyst is a reconciled investment and revenue schedule, not another repetition of the total contract value. Investors need to know when capacity becomes service-ready, when cash payments occur, how much spending is already included in reported capex and how the remaining program is funded. Management’s 2027 CIS target supplies a revenue checkpoint, but does not by itself answer those cash questions.

We will review Q4 results by February 28, 2027 and first-quarter evidence by May 15. These are research deadlines rather than confirmed announcement dates. The provider’s calendar estimate is not used as a company promise. Any official scheduling announcement should replace the provisional calendar only after verification.

A positive reassessment would combine security growth near current levels, delivery of at least the Q4 revenue floor, an adjusted margin consistent with the guide and a credible capital bridge. A price decline without those operating conditions would not automatically make the stock attractive. Conversely, a higher price can be justified if verified cash economics improve enough to raise intrinsic value.

The adverse checkpoints are equally explicit: missed capacity milestones, unexplained spending materially above the plan, annual CIS expectations slipping below the current range, or a debt increase without corresponding contracted capacity evidence. We would revise the model at the point the evidence changes, rather than defend the old $75 estimate. A scenario table is useful only when it leads to identifiable decisions and later accountability.

Data, definitions and source trail

The nine-quarter dataset retains original issuer statement rows, full non-GAAP reconciliation rows, cash-flow rows, balance sheets and solution definitions. Amounts in those raw source tables are in thousands; the analytical tables here use millions unless labeled per share. Annual cash capex includes both property and equipment and capitalized internal-use software. Free cash flow in this packet excludes acquisitions, buybacks, dividends and financing transactions; it is a diagnostic, not a claim that all remaining cash can be distributed.

Akamai is a Nasdaq-listed U.S. common-share issuer, not an ADS or a foreign VIE wrapper. PricewaterhouseCoopers is identified as the annual-report auditor. The newly designated customer preferred stock is nonvoting except where law requires otherwise and was not outstanding at September 30. The company does not currently pay dividends. Current stock-compensation and note-hedge effects are discussed above rather than hidden inside a single headline share count.

The financial model is unlevered at the enterprise level before its net-debt deduction. Lease occupancy costs remain in forecast cash economics. The earnings route is an equity valuation and therefore does not subtract debt a second time. The sum-of-parts route uses its own explicit future-build reserve. These conventions prevent the same capital requirement from being counted twice. They do not eliminate uncertainty about the eventual expenditure schedule.

All forward margins, discount rates, probabilities, valuation multiples and monitoring thresholds are author assumptions. The source release provides the company’s actual reported figures and separately labeled outlook. FMP supplies market data and third-party consensus snapshots. Sources were retrieved for this research on November 8, 2026. Numerical ratios and valuation arithmetic are computed in the retained calculation script, and the prior published episode remains unchanged.