JEDI put a decade of cloud risk into one award decision
The October 2019 JEDI award gave one company a cloud contract with a $10 billion ceiling over ten years, if all options were exercised. A single award concentrates schedule, capability and price risk in one decision. Its replacement spread them across a multiple-award vehicle.
Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.
Two-minute summaryThe dek and the first sentence of each section, quoted from the article
The October 2019 JEDI award gave one company a cloud contract with a $10 billion ceiling over ten years, if all options were exercised. A single award concentrates schedule, capability and price risk in one decision. Its replacement spread them across a multiple-award vehicle.
- The week On October 25, 2019, the Department of Defense’s daily contracts announcement led with a single entry from Washington Headquarters Services.
- How an IDIQ cloud contract allocates risk An indefinite-delivery, indefinite-quantity contract sets the terms for future purchases without fixing how much will be bought.
- Schedule risk: one protest can stop everything When one award covers an agency’s general-purpose cloud, every program that planned to use it waits on that award.
- Capability risk: one provider has to meet every need The 2018 strategy asked one General Purpose cloud to meet four tenets at once, including separate environments at every classification level and reach from central data centers to the tactical edge.
- Price risk: prices set by scenarios written years earlier A firm-fixed-price commercial cloud contract fixes unit prices or discounts, not total cost.
- Federal, state and local buyers: the same choice at different scales The department’s scale was unusual, but the choice it faced is common.
- The case for a single award The case for a single award is strong, and the strategy made much of it.
- Choosing where competition happens An agency planning a large cloud or platform procurement can make the competition choice explicitly rather than inherit it.
- With hindsight The award never delivered a general-purpose cloud.
The week
On October 25, 2019, the Department of Defense’s daily contracts announcement led with a single entry from Washington Headquarters Services. Microsoft Corporation of Redmond, Washington, had been awarded the Joint Enterprise Defense Infrastructure (JEDI) Cloud contract. It was a firm-fixed-price, indefinite-delivery/indefinite-quantity (IDIQ) contract with a ceiling value of $10 billion over ten years, if all options were exercised.
The announcement described the work in one sentence. JEDI would provide enterprise-level, commercial infrastructure as a service (IaaS) and platform as a service (PaaS) to support the department’s business and mission operations. The department obligated $1 million of fiscal 2020 operations and maintenance funds on a task order to cover the contract’s minimum guarantee. A second task order, for program management support to the Cloud Computing Program Office, was awarded at $0. The expected completion date, if all options were exercised, was October 24, 2029. The contracting activity was Washington Headquarters Services in Alexandria, Virginia, under contract number HQ0034-20-D-0001.
The award was the centerpiece of the DoD Cloud Strategy, a document dated December 2018. That strategy said the department would meet its mission needs through a multi-cloud, multi-vendor approach. It would combine one General Purpose cloud with Fit For Purpose clouds for needs the general one could not serve. JEDI was to be the General Purpose cloud, used for the majority of the department’s systems and applications.
The strategy explained the reasoning. A single enterprise-wide General Purpose cloud would let the department take advantage of economies of scale, provide common core services and aggregate data for analysis. It set four tenets for that cloud: offer IaaS and PaaS, offer separate environments at all classification levels, reach from centralized computing to the tactical edge, and enable emerging technologies such as artificial intelligence. It also said the department lacked large-scale, enterprise commercial cloud experience, which made an industry partner critical.
Fit For Purpose clouds were the exception. The strategy gave two examples: software-as-a-service applications such as email and collaboration, and the Defense Information Systems Agency’s milCloud 2.0 environment. Its primary bias was toward the General Purpose cloud. A mission owner who wanted a Fit For Purpose alternative would have to submit an exception brief to the Office of the DoD Chief Information Officer. The brief had to explain why the General Purpose cloud could not support the mission.
So the week’s announcement put a single vendor’s cloud at the center of a strategy that described itself as multi-cloud and multi-vendor. The ten-year ceiling set the scale. The $1 million minimum showed how little the department was committed to buy. Everything between the minimum and the ceiling would be bought through task orders placed with the one awardee.
The evaluation rested partly on price. The court record later showed that the solicitation, number HQ0034-18-R-0077, asked each offeror to price a set of hypothetical workloads, called price scenarios. An amendment told offerors to assume that data in the scenarios was highly accessible unless otherwise stated.
The award did not settle the matter. On November 22, 2019, Amazon Web Services filed a bid protest against it in the United States Court of Federal Claims. A court opinion issued in that case the following February recorded that the department had publicly announced the award to Microsoft on October 25.
JEDI is easy to remember as a fight between two technology companies. For anyone who buys cloud services for a government, the more useful way to read it is as a procurement design. The department decided to run its general-purpose cloud through one award, competed once, for up to ten years. That decision put three different risks into a single moment.
The first is schedule risk: when everything depends on one award, one protest can stop everything. The second is capability risk: one provider has to meet every need the contract covers, for the whole term. The third is price risk: the prices that decide the award are set by scenarios written before the award, and they bind work that will be defined later. A multiple-award design spreads each of these risks across many smaller decisions. It does not eliminate them. The choice an agency makes is where competition happens, and that choice decides where the three risks sit.
After a short account of how an IDIQ cloud contract allocates risk, the sections below take the three risks in turn. Three further sections follow: how the same choice looks to federal, state and local buyers; the strongest case for a single award; and how an agency can decide where its own competition should happen. Nothing here is legal advice, and procurement rules differ between federal agencies, states and localities. Contracting officers and counsel should be part of any of these decisions.
How an IDIQ cloud contract allocates risk
An indefinite-delivery, indefinite-quantity contract sets the terms for future purchases without fixing how much will be bought. It has a minimum the government must order and a ceiling it cannot exceed. Actual purchases happen through task orders placed during the contract’s life. JEDI’s minimum was $1 million and its ceiling $10 billion. The gap between them shows how much of the real buying was left to the future.
IDIQ contracts come in two shapes. A single-award IDIQ has one contractor, and task orders go to that contractor without further competition. A multiple-award IDIQ has several contractors, and each task order is competed among them. The Federal Acquisition Regulation tells contracting officers to give preference, to the maximum extent practicable, to multiple awards, and lists circumstances in which a single award must be made instead.
The two shapes put competition in different places. In a single award, competition happens once, at the start, and it is intense. Every vendor knows the winner will receive every order. In a multiple award, the initial competition decides who gets onto the vehicle. The real contest for each piece of work happens later, task order by task order, among vendors who already hold the contract.
What the minimum and ceiling mean for budgets
A contract ceiling is not a budget. JEDI’s $10 billion was the most the department could order, not money set aside. Funds are obligated task order by task order, from appropriations available at the time. The $1 million obligated at award came from fiscal 2020 operations and maintenance funds and covered only the minimum guarantee.
That structure matters for who controls cloud spending. Where the programs that move workloads to the cloud pay from their own budgets, a central contract sets the prices and terms, but the decisions that create cost are made by dozens of program offices. A single award gives the center control over the provider. It does not give the center control over consumption. That has to be built separately, through tagging, reporting and review of what each program orders.
The ten-year term was also not guaranteed. The announcement gave an expected completion date of October 24, 2029, if all options were exercised. Options are the government’s right, not its obligation. A long single award with options gives the buyer exit points, but each one is a decision the buyer has to be ready to make, with an alternative in hand.
Who carries each risk
The three risks land on different parts of an agency, which is one reason they are easy to underweight when the vehicle is designed.
- Schedule risk lands on program offices. They planned migrations and new systems around the vehicle, and they absorb the cost of extending old hosting while it is held up.
- Capability risk lands on mission owners. They live with whatever the provider can and cannot do, or spend time arguing for exceptions.
- Price risk lands on budget offices and, in the end, on appropriations. They see the total cost only as programs consume services over years.
The contracting office that designs the vehicle carries none of these directly. Bringing program, mission and budget owners into the vehicle decision, with their exposure stated, gives the design a fuller picture of what is at stake.
Firm-fixed-price, with prices set by scenario
JEDI was firm-fixed-price. In a commercial cloud contract priced this way, the government buys from catalogs of unit prices, such as the price of a unit of compute or storage. To compare offers at award, a solicitation describes hypothetical workloads and asks each bidder to price them. The court record in the JEDI protest shows the department used a set of price scenarios for this purpose. The injunction later turned on one of them.
Scenarios are a reasonable way to compare unit prices. They are also a forecast. They assume a pattern of use that may not match what the department actually builds over ten years. A single award fixes that forecast into the one decision that counts.
Where a protest can land
A disappointed bidder can protest an award to the Government Accountability Office or to the Court of Federal Claims. A protest of a single award challenges the whole contract. A protest of a task order under a multiple-award vehicle challenges one order. The design of the vehicle therefore decides how much work any one protest can hold up.
Schedule risk: one protest can stop everything
When one award covers an agency’s general-purpose cloud, every program that planned to use it waits on that award. If the award is protested and a court orders the work stopped, no program can start. The agency has bet its schedule on a single legal event it does not control.
This is not a hypothetical in JEDI’s case. The protest filed in November 2019 led, in February 2020, to an order barring the department from proceeding with contract activities while the case continued. Any program that had planned to run on the new general-purpose cloud had to wait or find another route.
Why the stakes make protests likelier
A single award raises the stakes of losing, and higher stakes make a protest more rational for the losing bidder. Ten years of exclusive access to an organization’s general-purpose workload is worth fighting for even if the odds are poor. In a multiple-award vehicle, a vendor that loses a task order can compete for the next one. Its incentive to litigate any single decision is much lower.
The incentive works on the agency too. Expecting a protest, an agency running a single award may write a longer, more defensive solicitation, run a longer evaluation and document every judgment more heavily. Each of those steps protects the award and delays it.
What a program office can do about it
A program that depends on a vehicle still being competed carries the vehicle’s schedule risk whether it wants to or not. The practical responses are to keep an interim path open and to avoid designing in features only one provider offers before the award is settled. An interim path might be an existing agency hosting environment, an existing government-wide vehicle, or a limited pilot under another contract.
Program offices should also record what the delay costs. Data center leases extended, hardware refresh deferred, staff held on legacy systems. That record is what the agency needs to weigh a single award against a multiple award the next time.
Planning protest time into the schedule
For a large award, a protest is a planning risk worth budgeting for. GAO’s bid protest rules say it shall decide a protest within 100 days after it is filed, and a protest there may require the agency to suspend performance. The Court of Federal Claims has no fixed deadline, and it can order a stop to work, as it did in JEDI. A program plan that assumes work begins the week after award has not counted the most likely delay.
The more useful planning step is to ask which program milestones depend on the award and which do not. Training, architecture work, security documentation and data clean-up may be able to proceed against any likely winner. Building features that rely on one provider’s specific services should wait until the award is secure. A schedule built this way loses less when the award is held up.
The cost of waiting should be budgeted, not hoped away. If a program depends on a new vehicle, its plan should include the cost of keeping existing hosting for a realistic protest period, and a decision date after which it proceeds on another vehicle. Without that date, programs can wait indefinitely for a contract that may never be usable.
Capability risk: one provider has to meet every need
The 2018 strategy asked one General Purpose cloud to meet four tenets at once, including separate environments at every classification level and reach from central data centers to the tactical edge. A single award means one company must deliver all of that, for the whole term, at the pace the department needs. A single award commits the buyer to the winner’s weaknesses as well as its strengths.
The strategy’s answer was the Fit For Purpose exception. A mission owner whose needs the general cloud could not meet could ask for an alternative, through an exception brief to the department’s CIO. That is a governance process, and it adds friction by design. The friction protects the economies of scale a single cloud offers. It also slows any program whose needs are better met elsewhere.
Exception requests are also information. Each one records a need the default platform could not meet. An agency that tracks them, by capability and by program, learns early where its single provider is falling short. A rising number of exceptions in one area is a sign that the vehicle design no longer fits the requirement, long before a formal review would say so. An agency that treats exceptions only as requests to approve or refuse discards that signal.
How capability risk grows over a long term
Cloud services change quickly. A ten-year term spans several generations of computing, storage, data and machine-learning services. A buyer who chooses a single provider at the start is betting on that provider’s roadmap as well as its current catalog. If a competitor releases a capability the mission needs, the buyer can wait, seek an exception, or go without.
The department’s own explanation when it later canceled JEDI points directly at this risk. It said its requirements had evolved, its own familiarity with cloud had grown, and industry had advanced. It pointed to new initiatives in joint all-domain command and control and in artificial intelligence and data acceleration, and to user requirements to use multiple cloud environments.
Concentration as a security and resilience question
Capability risk has a resilience side as well. A single general-purpose provider is a single dependency. An outage, a security incident or a failure of a shared identity or management service at that provider reaches every workload on it. The strategy recognized resilience as an objective and discussed using multiple regions and availability zones. Regions and zones protect against facility failures. They do not protect against a fault that affects the provider’s whole platform.
Multiple providers do not remove concentration automatically. A program may still run each workload on one provider. They do give the agency a place to move a workload if it must, and they keep the skills to do so alive inside the organization.
Classification levels as a capability test
The strategy’s second tenet, separate environments at all classification levels, is a demanding capability to buy. The strategy asked for a separate environment at each level, and each must be in place before programs at that level can use it. A single award asks the winner to deliver all of them on the department’s timetable. If one level lags, every program that needs it waits or seeks an exception.
Civilian agencies and states face a smaller version of the same test. Workloads with criminal justice information, federal tax information or health data come with their own security requirements. A provider that meets one set may not yet meet another in the region or service a program needs. Checking each regulated workload against each candidate provider’s current authorizations, before choosing a vehicle design, shows how much capability risk a single award would carry.
The tactical edge
The strategy also asked the General Purpose cloud to reach the tactical edge, meaning computing that runs close to forces in the field, often on limited or intermittent connections. Edge capability depends on hardware, software and the provider’s ability to support disconnected operation. Offerings in this area can change over a contract’s life. A ten-year single award fixes the department to one provider’s edge roadmap at the moment when that roadmap is least certain.
Price risk: prices set by scenarios written years earlier
A firm-fixed-price commercial cloud contract fixes unit prices or discounts, not total cost. Total cost depends on what the agency builds, how it architects it and how well it manages use. A single award competes unit prices once, against scenarios. After that, the agency has little pricing power for that work. It can still negotiate, but the vendor knows it holds every order.
The JEDI court record shows how much weight scenarios can carry. The preliminary injunction rested on how the department evaluated the winner’s pricing for one scenario. That scenario concerned a containerized data analysis framework and turned on what kind of storage the solicitation required. A ten-year award turned, at least for the purposes of the injunction, on how a hypothetical workload’s storage was priced.
Continuous competition as a pricing tool
A multiple-award vehicle keeps price competition alive. Each task order lets providers on the vehicle price the actual workload, not a scenario written years earlier. Cloud prices can change over a contract’s life, and a task-order competition can capture those changes. The cost is the effort of running more competitions and of managing more than one provider.
Buyers can also manage price risk inside any contract shape. Clear tagging of resources to programs, regular review of unused capacity, and commitments sized to measured use all reduce total cost regardless of who the provider is. Those practices can matter more than the vehicle. The vehicle decides whether price competition is available at all.
Exit costs are part of the price
The price that matters at the end of a contract is the cost of leaving. Moving large data sets out of a cloud, rebuilding services that depend on provider-specific features, and retraining staff all cost money and time. A single award with a long term can deepen dependence on one provider’s services, which raises exit cost. An agency should price exit as part of the award decision, not discover it at recompetition.
What scenarios leave out
Price scenarios compare the prices of the services they describe. Real cloud bills include much they do not. Data transfer out of the provider’s network, premium support tiers, managed services added after award, third-party software bought through the provider’s marketplace, and training all add cost. A scenario-based evaluation can pick the lowest bidder on the measured workload and still miss the provider whose total cost is lower once those items are counted.
The JEDI dispute showed a narrower version of the problem. The question of what kind of storage the solicitation required for a single scenario was enough, at the preliminary stage, to support an injunction. An agency writing scenarios should expect each assumption to be read closely by a disappointed bidder. Plain definitions, worked out with technical staff before release, cost far less than an amendment and a clarification after it.
Consumption drives the bill
Over the life of a cloud contract, how much an agency uses matters more than the unit prices it negotiated. Idle virtual machines, oversized databases, data kept in expensive storage tiers and test environments left running all cost money at any price. An agency with strong consumption management on a mediocre contract can spend less than one with weak management on an excellent contract.
Commitment discounts complicate the picture. Providers offer lower prices in exchange for committing to a level of use over one or more years. Those discounts are real savings, and they also deepen dependence on one provider. A program that commits heavily to one provider’s discount plan has given up some of the flexibility a multiple-award vehicle was meant to preserve. Commitments should be sized to measured, stable use, and reviewed against the cost of the flexibility they give up.
Federal, state and local buyers: the same choice at different scales
The department’s scale was unusual, but the choice it faced is common. Every public buyer of cloud services decides whether to compete once for a single provider or to qualify several and compete the work as it comes.
Three things change with the level of government. The first is scale, which decides whether a buyer can afford to manage several providers at all. The second is the protest regime, which differs between federal, state and local law and decides how exposed a single award is. The third is the set of vehicles already available, since a buyer that can order from an existing multiple-award contract may never face the single-award choice directly. The three subsections below take each level of government in turn, with those differences in mind. In each case the question is the same one the department faced: where competition should happen, and which of the three risks the buyer is prepared to hold.
Federal civilian agencies
Civilian agencies may have government-wide vehicles available, along with their own agency vehicles. Where they do, they can avoid the single-award question by ordering through an existing multiple-award contract. The risk for them is gradual fragmentation: many programs each choose a provider through a different route, and the agency ends up with several clouds and no common security or cost management. A vehicle design that allows competition still needs an enterprise layer of identity, logging, network and cost controls that works across providers.
A common security authorization baseline, where one applies, lowers the cost of using more than one provider, because each agency does not have to assess each provider from scratch. It does not settle capability questions. A provider can be authorized and still lack a service, a region or an integration a program needs.
States
A state may run enterprise technology through a central agency, or buy cloud through cooperative contracts or statewide master agreements with several resellers or providers. A state that consolidates hosting into one provider gains the same economies of scale the defense strategy sought, and takes on the same concentration. Its protest exposure differs, since state procurement law sets its own protest procedures, but the stakes logic is the same. A single statewide award draws harder challenges than a pool.
States that run shared services also face a billing question where the central technology agency recovers its costs by charging other agencies. If it buys from one provider, its rates reflect that provider’s prices. If it buys from several, it has to decide whether agencies choose a provider and pay its price or pay a blended rate. That decision shapes whether agencies have any reason to manage their consumption, and it deserves as much attention as the vehicle itself.
Cities and counties
At city and county scale, the vehicle design may matter less than it did for JEDI. The risk may lie instead in software-as-a-service applications, each bought separately, each with its own data and its own exit path. The lesson of JEDI for them is about term and exit. A long single-vendor commitment for a core system, such as public safety records or tax collection, deserves the same attention to exit cost and capability roadmap that the department’s cloud should have received.
Local buyers also have a resource the department did not: other governments. Cooperative purchasing agreements let a city or county order from contracts another public body competed. That spreads the cost of competition, but it imports someone else’s choice of where competition happens. A county that buys cloud services through a cooperative contract should check whether that contract allows it to compete orders among several providers or ties it to one.
The case for a single award
The case for a single award is strong, and the strategy made much of it. One provider means one security model, one set of tools, one identity integration and one team to train. It allows the economies of scale the strategy sought and makes data aggregation easier. The department said it lacked enterprise cloud experience. Learning one platform well is easier than learning several at once.
A single award also simplifies accreditation. Each environment that holds classified or sensitive data must be authorized, and authorizing one provider’s environments at each level is less work than authorizing several. Multi-cloud architectures carry real costs: duplicated skills, more complex networking, more places for configuration errors, and the temptation to build only to the lowest common denominator of services.
Finally, task-order competition is not free. Every competition takes staff time and can itself be protested. An agency running dozens of task orders a year needs contracting and technical capacity it may not have.
The workforce argument
The strongest practical case for one provider is people. Cloud engineering skills are tied to specific platforms. A team that knows one provider’s identity model, networking and security tools deeply is more effective than a team that knows three superficially. An agency that struggles to hire and keep cloud engineers has good reason to fear spreading them thin.
That argument supports a primary provider. It does not require a single award. An agency can concentrate most of its work, and most of its training, on one provider chosen through task-order competition, while keeping a small team able to use a second. That small team is the price of keeping options open, and it can be far smaller than the cost of an exit forced without one.
Why the argument still holds
These are real advantages, but they are advantages of standardization, and standardization can be achieved inside a multiple-award vehicle. An agency can choose a primary provider for most workloads through task-order competition and still keep the ability to compete later work. It can set one enterprise identity, logging and policy layer that every provider must integrate with. What it cannot do in a single award is change its mind without a new procurement.
There are cases where a single award is the better design. A narrow, stable requirement, such as one email and collaboration service for a whole organization, may gain little from repeated competition and lose a lot from fragmentation. A small agency without contracting staff to run task-order competitions may be better served by one well-negotiated contract with clear exit terms. The strategy itself treated software-as-a-service applications such as email as a separate category. The single-award case is weakest where JEDI applied it: a broad, general-purpose platform meant to serve requirements no one could yet describe.
The JEDI record suggests the department reached the same conclusion. Its replacement kept one contract vehicle but made it a multiple-award vehicle with several providers on it.
Choosing where competition happens
An agency planning a large cloud or platform procurement can make the competition choice explicitly rather than inherit it. Four questions help.
- How long will the requirement stay stable, and is the contract term longer than that?
- Can one provider meet every need on the roadmap, including classified or regulated environments, edge locations and emerging services?
- If the award is protested and stayed, what is the interim path, and what does each month of delay cost?
- What would it cost to move the largest workload to another provider in the final year of the term?
If the answers point to a stable requirement that one provider can meet, with a tolerable delay cost and a priced exit, a single award may be the right design. If not, competition belongs at the task order. In both cases, the enterprise layers of identity, logging, network and cost management should be designed so they do not depend on any one provider. Those layers decide whether a second provider can be used when it is needed.
The decision should be recorded with its reasons. A short memo stating where competition happens, why, and what risks remain gives auditors and inspectors general a clear basis for review. It also gives the next program office something to learn from.
Designing the enterprise layer
Whatever the vehicle, the parts of a cloud environment that should not depend on any provider are the ones that govern all of them. They include the agency’s identity system and how it federates into each cloud, a central log store that every environment feeds, and network connections to the agency’s own systems. They also include security configuration policies that can be applied to more than one platform, and a cost tagging scheme every program must follow.
These layers are where an agency keeps its independence. If identity, logging and policy are built only inside one provider’s native tools, a second provider is usable only after that work is repeated. If they are built to span providers from the start, adding a provider is a matter of integration rather than reconstruction. That design choice costs more at the beginning and less at every later decision point.
Governing where each workload goes
A multiple-award vehicle moves a decision from the contracting office to the program. Someone has to decide which provider each workload should use, and on what grounds. Without a rule, programs choose by habit or by which vendor reached them first, and the agency loses the benefits of competition anyway.
A placement process can be light. It needs criteria, such as required authorizations, data location, dependence on provider-specific services and expected cost, and it needs a person who decides. JEDI’s strategy used an exception brief to protect a single default. A multiple-award approach can use the same tool in reverse: a short written case whenever a program wants to depend on a feature only one provider offers. The aim is to make sure someone made that choice knowingly and wrote down the cost of reversing it.
Planning the end of the term at the start
Every cloud contract ends, and the end is when the vehicle design is tested hardest. Under a single award, recompetition means either keeping the incumbent or moving everything. Moving everything is costly enough that the incumbent starts with a large advantage, which weakens the competition the recompetition is meant to create. Under a multiple-award vehicle, workloads are already spread, and the next vehicle can be phased in order by order.
An agency can prepare for the end from the first year. It can keep current documentation of each workload’s dependencies, test data exports at scale at least once, and record which services would have to be rebuilt elsewhere. Those steps cost a little each year. They turn the end of the term from a cliff into a choice.
With hindsight
The award never delivered a general-purpose cloud. The bid protest filed on November 22, 2019, led the Court of Federal Claims to grant a preliminary injunction. The opinion was issued under seal on February 13, 2020, and released in redacted form on March 6. The court found that the protester was likely to show the department had erred in evaluating the winner’s pricing for Price Scenario 6. That scenario involved a containerized data analysis framework, and the dispute turned on what kind of storage the solicitation required. The court barred the department from proceeding with contract activities and required the protester to post $42 million in security.
On July 6, 2021, the department canceled the JEDI solicitation and began terminating the contract. It said JEDI no longer met its needs because of evolving requirements, increased familiarity with cloud and advances in industry. It said it still had unmet needs for enterprise-wide commercial cloud at all three classification levels, reaching the tactical edge at scale. It said those needs had only advanced with its joint all-domain command and control effort and its artificial intelligence and data acceleration initiative.
In the same announcement, the department described the replacement. The Joint Warfighter Cloud Capability (JWCC), as the release named it, would be a multi-cloud, multi-vendor IDIQ contract. The department said it intended to seek proposals from Microsoft and Amazon Web Services, which its market research suggested were the only providers able to meet its requirements. It said it would keep researching whether other U.S.-based hyperscale providers could.
On December 7, 2022, the department announced Joint Warfighting Cloud Capability awards to Amazon Web Services, Google Support Services, Microsoft and Oracle. It described JWCC as a multiple-award vehicle for buying commercial cloud directly from providers, at all classification levels, from headquarters to the tactical edge, with commercial pricing or better. The Defense Information Systems Agency’s Hosting and Compute Center was named as the place for customers to start. The announcement said the agency had built tools to make it easier for defense customers to purchase, provision and onboard. The announcement did not state a ceiling value. It listed the services the vehicle was meant to make available, among them elastic computing, storage and network infrastructure, advanced data analytics, and tactical edge devices.
Against the three risks, the record is clear. Schedule risk materialized first. The award was enjoined within months, and the department had no general-purpose cloud from JEDI at all. Capability risk was the reason the department gave for canceling. Price risk was the issue on which the injunction turned. In our reading, the replacement answered each one. Four providers sit on one multiple-award vehicle, so work can be competed order by order, and a protest of one order need not stop the others.
What did not change was the department’s goal. The 2018 strategy had already called for a multi-cloud, multi-vendor environment. JEDI’s single award was the part that conflicted with that description, and it was the part that was abandoned. The broader point for public buyers is that a cloud strategy’s architecture and its procurement design need to say the same thing. When they diverge, the contract, not the strategy, decides what gets built, and in JEDI’s case it decided that nothing would be built under that contract at all.
Spotted an error? Tell us and we will check it and correct the article where needed.