Runway is short. What do you cut from engineering first?

When runway is short, cut engineering costs in this order: idle cloud resources and unused software seats first, then new commitments and uncontrolled AI spending, followed by project scope. Cut people last, and make those decisions deliberately rather than through repeated rounds of layoffs. Protect the minimum maintenance work needed to keep production safe. Carta's data shows that engineering accounts for nearly half of payroll at companies valued between $1 million and $10 million. But cutting engineering too deeply can undermine what the company needs to ship before its next funding round.

Key takeaways

• Engineering is one of the largest and slowest payroll costs to reduce. Carta's data shows that it accounts for nearly half of payroll spending at companies valued between $1 million and $10 million. According to the Bureau of Labor Statistics, wages represent only 70.0% of what a US private employer pays per employee, so the full cost of employment goes beyond salary.
• Cloud spending is a practical place to start. Buyers estimate that 21% to 29% of their cloud spending is wasted, according to Harness (700 respondents) and Flexera (753 respondents). These figures are self-reported estimates, not audited measurements, but they point to potential savings that do not require cutting engineering capacity.
• Long-term cloud discounts can become a liability when cash is tight. AWS and Google Cloud offer commitment discounts of up to 72% and 70%, respectively, in exchange for commitments lasting one or three years. Google Cloud commitments cannot be cancelled, while an AWS Savings Plan can only be returned within seven days. If your runway is nine months, a commitment that lasts one or three years can outlive your available cash.
• The funding gap may be longer than expected. Carta reported a median of 616 days between seed and Series A in Q2 2025, while Crunchbase puts the interval at more than two years. Plan engineering spending around the time it may take to raise again, rather than assuming the next round will arrive quickly.
• Layoffs create costs beyond the immediate savings. Downsizing is associated with higher voluntary turnover among employees who remain. At a company with a median seed-stage team of four people, losing one person means losing 25% of the team. That can leave critical work uncovered and put delivery at risk.
• Do not cut below the minimum maintenance needed to keep the product secure and reliable. In audited commercial codebases, 92% contain open-source components that are four or more years out of date, while 80% of dependencies go a year without an upgrade. Public TLS certificate lifetimes are also set to fall to 47 days by March 2029. These figures highlight why maintenance cannot simply stop when budgets tighten.

In this article

• Why does the usual cost-cutting advice miss engineering?
• What does an engineering budget actually consist of?
• What should you cut first, and what last?
• How do you cut engineering costs in four weeks?
• When are cheaper engineering hours the right lever?
• Where RocketDevs fits
• Conclusion: cut waste before you cut engineering capacity
• Frequently asked questions

Why does the usual cost-cutting advice miss engineering?

Most cost-cutting advice treats engineering as a single expense: headcount. But at companies valued between $1 million and $10 million, engineering accounts for nearly half of payroll, according to Carta's 2022 compensation report. Engineering budgets also include several other costs, each of which can be reduced on a different timeline. General advice tells founders to cut tools and marketing, protect employees, and act quickly. It often fails to examine the engineering budget itself, where much of a software startup's spending sits.

Carta's 2022 report analysed more than 127,000 employee records and found that engineering was the largest job function at most startups. The trend continued into 2025: engineers accounted for approximately 29.7% of all new hires made by startups on Carta in the first half of that year, the largest share of any function, according to Carta's H1 2025 compensation report. For a seed-stage founder trying to reduce monthly spending, engineering is therefore one of the biggest areas to examine.

Financial pressure is already affecting many companies. In January 2026, VC-backed companies on Carta made 26,030 new hires, the lowest January total since 2018. During the same period, 12,161 employees left voluntarily and 8,217 were laid off, according to Carta's H2 2025 compensation report. Cost-cutting is no longer an unusual response to financial pressure. The challenge is making cuts that extend runway without damaging the business.

What the 2022 playbooks actually said

One of the best-known examples of startup cost-cutting advice is Sequoia's guidance from May 2022. Its Adapting to Endure presentation encouraged founders to review projects, research and development (R&D), marketing, and other expenses through a cost-cutting exercise. The goal was not necessarily to make every cut immediately, but to prepare to act within 30 days if needed.

Sequoia's companion presentation, Extending Your Runway, by Ravi Gupta and Sonya Huang, took a more detailed approach. It recommended separating operating expenses into compensation and non-compensation costs, including computer hardware, hosting, and subscriptions. It also made an important point: few cuts are both easy to implement and capable of producing significant savings.

Neither presentation explicitly says to make all cuts in one round. The idea of cutting once, rather than repeatedly, is a recommendation based on the potential consequences of successive layoffs, which the research discussed later in this article helps explain. It should not be presented as Sequoia's own advice.

The Extending Your Runway deck also sets out a useful fundraising target: start raising money 12 months before your cash runs out. The runway you need must therefore cover both the time required to reach your next milestone and an additional 12 months to secure funding.

Paul Graham's essay Default Alive or Default Dead? offers another way to assess a startup's financial position. It asks whether the business can reach profitability using its existing cash, given its current expenses and recent growth. On what kills funded startups, Graham is blunt: “Hiring too fast is by far the biggest killer of startups that raise money.”

Both approaches focus on the same underlying issue: a company's runway is the result of its spending and growth decisions. For a software startup, engineering is one of the most important budgets to examine. But it is not a single expense. It contains four different types of costs, each requiring a different approach to cutting.

What does an engineering budget actually consist of?

An engineering budget has four main parts: people, cloud infrastructure, tools and software licences, and AI usage. People are usually the largest expense and the slowest to reduce. Their cost also extends beyond salary. In Q2 2026, wages and salaries accounted for 70.0% of total compensation for US private industry workers, according to the Bureau of Labor Statistics (BLS). The remaining share reflects benefits and other employer compensation costs. The other three categories are generally faster to adjust and easier to reverse.

The best way to assess each category is by its flexibility: how much runway a saving buys, how quickly it takes effect, and whether you can reverse the decision if circumstances improve. A smaller saving that takes effect within a week and can be reversed may be more valuable than a larger saving that takes three months to deliver and locks the company into a decision.

People, cloud, tools, AI usage: four lines that cut differently

People. This is usually the largest engineering expense, and cutting it has consequences beyond the immediate saving. The BLS reported a US median annual wage of $135,980 for software developers in its 2025 Occupational Employment and Wage Statistics data, equivalent to $65.38 an hour. O*NET Online reports the same figures. Employers must budget for more than wages: BLS data puts total compensation for private industry workers at $46.89 an hour, compared with $32.82 in wages. That means total compensation is approximately 43% higher than wages alone, calculated from those BLS figures.

Reducing headcount is also slow and can be expensive. In the UK, statutory redundancy notice ranges from at least one week for employees with between one month and two years of service to one week per year of service, up to 12 weeks. Statutory redundancy pay generally applies to employees with at least two years of service. In the US, the Worker Adjustment and Retraining Notification (WARN) Act regulations require at least 60 days' notice for covered mass layoffs, but the rules generally apply to employers with 100 or more employees. Most seed-stage startups fall below that threshold. Employment contracts, local laws and severance arrangements vary, so check the applicable requirements before calculating the potential savings.

People also carry knowledge that may not be documented anywhere else. Carta's data puts the median seed-stage company at just four employees. In an automated analysis of 133 popular GitHub projects, Avelino and colleagues found that 65% had a truck factor of two or less. This means that losing just two developers could leave a project unable to continue effectively. A separate survey of 269 engineers found that knowledge is shared through code reviews and meetings as well as stored in code. Our article on what happens when one person knows everything explores this risk in more detail. For cost-cutting purposes, the key point is that removing an engineer can mean losing essential knowledge alongside the salary expense.

Cloud. Cloud spending is often easier to adjust quickly, making it a sensible early target. Flexera's 2026 State of the Cloud survey, which included 753 cloud decision-makers, found that respondents estimated waste at 29% of infrastructure-as-a-service (IaaS) and platform-as-a-service (PaaS) spending. This was the first increase after five years of decline. Harness, in a survey of 700 engineering leaders and developers, estimated cloud infrastructure waste at 21%. Both figures are self-reported estimates from companies that sell cost-management tools, so treat them as indications of where to investigate rather than as estimates of your own potential savings.

Cloud waste often comes from resources that remain running when nobody needs them. The FinOps Foundation identifies pre-production environments, such as development and testing systems, as the biggest targets for savings and recommends scheduling them to start and stop automatically. Amazon RDS databases, for example, can be stopped so that instance-hour charges cease, although storage and backup charges continue. For workloads that can tolerate interruption and be restarted, AWS Spot Instances offer discounts of up to 90% compared with On-Demand prices. If your cloud bill is rising for reasons beyond your team's control, our article on why cloud costs are rising explains the hardware pressures involved.

Cloud credits create another risk for startups. AWS Activate offers eligible pre-Series B companies up to $200,000 in credits. These credits can reduce the immediate bill, but they can also hide the actual cost of running the product. When the credits expire, the cash bill can rise sharply, potentially just as the company is running short of runway. Before making spending decisions, examine the full cloud bill before credits are applied.

At larger companies, cloud can become a major strategic expense. An Andreessen Horowitz analysis of public software companies in 2021 found that committed cloud spending averaged 50% of cost of revenue. A seed-stage company operates at a different scale, which is why removing idle resources and unnecessary usage is generally a better first step than redesigning the entire architecture.

Tools and seats. Software licences may be a relatively small expense, but unused subscriptions can quietly drain cash. Many developer tools charge for every assigned seat, even when the person rarely uses the product. GitHub Copilot Business costs $19 per granted seat per month, while Enterprise costs $39. Usage-based services create a different expense: GitHub Actions, for example, charges $0.006 per minute for its standard Linux runner.

Zylo, a software licence management company, reports that organisations leave an average of 36% of SaaS licences unused against recommended utilisation. This is a measure from Zylo's own customer base, not a figure specific to seed-stage startups. Nevertheless, the underlying problem is straightforward: companies continue paying for accounts nobody needs because no one is responsible for reviewing them. Audit every licence, identify its owner and usage, and remove seats that are genuinely unnecessary before reducing the tools your engineers rely on.

AI usage. AI spending is a newer budget category and can be difficult to track if developers use APIs without clear limits. The FinOps Foundation's 2026 State of FinOps survey, based on 1,192 respondents, found that 98% were managing AI spending, compared with 31% in 2024. For a startup, the immediate risk is an API key with no spending controls: usage can grow without a corresponding increase in the value delivered.

Both Anthropic and OpenAI provide ways to control API spending. Anthropic supports spending limits by usage tier, as well as custom spending and rate limits for workspaces. OpenAI supports project-level limits and hard spending limits designed to stop API traffic when the applicable limit is reached. Check how each provider applies its limits before relying on them as a guaranteed spending ceiling. Our comparison of hard caps on AI agent spending explains the differences between providers in more detail.

Why a three-year commitment is the wrong discount at nine months

Cloud commitment discounts reduce prices in exchange for agreeing to spend money over a set period. When a company has nine months of runway, flexibility may be more valuable than the discount. A lower unit price is not necessarily a saving if the company must keep paying after it runs out of cash.

AWS Savings Plans offer discounts of up to 72% in exchange for committing to a specified hourly amount of compute usage for one or three years. Google Cloud resource-based committed-use discounts offer savings of up to 70% for memory-optimised machines and 55% for other machine types. Google states that commitments cannot be cancelled after purchase and that customers continue to be billed monthly until the term ends.

The options for reversing these commitments are limited. An AWS Savings Plan can be returned only if its hourly commitment is $100 or less, it was purchased within the previous seven days, and the return occurs in the same calendar month. Azure reservations offer more flexibility through refunds, but refunds are subject to a USD 50,000 cancellation limit over a rolling 12-month period. Microsoft also reserves the right to impose a 12% early termination fee. From 1 February 2027, new reservations for services covered by savings plans will no longer be exchangeable.

The FinOps Foundation explains the trade-off: the more restrictive a discount arrangement is, the larger the potential saving. This can work well for a company with predictable usage and enough cash to support a long commitment. It is much less attractive for a startup that may need to shut down a service, change its architecture or reduce usage within nine months. A three-year commitment made with nine months of runway could leave the company paying for another 27 months after its current cash runs out.

Research suggests that even larger companies struggle to choose commitments accurately. In Harness's survey, 55% of developers said commitment purchases were ultimately based on guesswork. Flexera also found that fewer than half of organisations used any one commitment discount per provider. For a startup with nine months of runway, a commitment might still make sense if it is small, lasts one year, requires no upfront payment and covers a production database that must continue running regardless of roadmap changes. The expected savings must justify the remaining financial obligation.

Moving away from cloud infrastructure entirely is not usually a sensible first cut for a seed-stage company. 37signals reduced its annual cloud bill from approximately $3.2 million to $1.3 million after purchasing around $700,000 in Dell servers, according to The Register. That strategy required substantial upfront spending and an existing operations team. It is therefore very different from the immediate, reversible savings a startup with nine months of runway needs.

The maintenance floor

The maintenance floor is the minimum engineering work and expertise required to keep a product secure, available and recoverable. It includes someone who can respond when production breaks, regular dependency updates, certificate renewals and backups that have actually been tested through restoration. Cut below this level and the apparent saving can turn into an outage, security incident or failed recovery at exactly the wrong time.

Software becomes harder to maintain when nobody updates it. Black Duck's 2026 audit of 947 commercial codebases found that 92% contained components that were four or more years out of date. Sonatype reported that 80% of application dependencies remained un-upgraded for more than a year. It also found that when a vulnerable component was being used, a fixed version was already available 95% of the time. Both companies sell tools related to software security and maintenance, so their findings should be read in that context. However, the results from two separate datasets point to the same concern: neglecting maintenance can leave known problems unresolved.

Some maintenance tasks become more demanding because of changing technical requirements. The CA/Browser Forum's ballot SC-081v3 sets out a phased reduction in the maximum validity of public TLS certificates, from 398 days to 47 days by March 2029. Let's Encrypt has also announced plans to reduce its certificate lifetime to 45 days by 2028 and warns that a hardcoded 60-day renewal interval will no longer be sufficient. If the person responsible for certificate renewal leaves and nobody else understands the process, the company could discover the problem when its website or service stops working.

On-call coverage is another area where founders can underestimate the minimum staffing required. Google's Site Reliability Engineering (SRE) book recommends a minimum team of eight engineers for a single-site on-call rotation, with on-call duties capped at a quarter of each person's time. Most seed-stage startups cannot maintain a team of that size. Their maintenance floor must therefore be defined around the actual product and its risks.

At a minimum, two people should be able to deploy and roll back changes, patch the system and restore production. The team should also have written instructions for the three most common failures and alerts that reach a named person. If a planned headcount reduction leaves only one person capable of performing these tasks, the company has cut below its maintenance floor. That is not a sustainable saving; it is a new operational risk.

What should you cut first, and what last?

Cut idle and unowned spending first, and leave people until last. Buyers estimate that 21% to 29% of cloud spending is wasted, according to Harness and Flexera. Removing unnecessary cloud resources can reduce costs within days without affecting the product customers use. Layoffs take longer to implement and are associated with higher voluntary turnover among employees who remain, according to research by Trevor and Nyberg published in the Academy of Management Journal.

The ranking below considers three factors: cost, speed and reversibility. No dataset establishes exactly how many months of runway each type of cut buys a seed-stage startup. Treat this order as practical operating guidance based on documented costs, implementation timelines and the consequences of reversing a decision. Before acting, check your own spending data and calculate the actual savings.

The first three cuts have something important in common: they target spending that does not directly contribute to shipping the product. An idle staging environment, a Copilot seat assigned to someone who has left, or an API key still generating charges for an abandoned prototype can all drain cash without creating corresponding value. Removing these costs may require a little investigation, but it need not reduce engineering capacity.

Cut the spend that does not ship before cutting the people who make shipping possible.

The fourth step is different. You are not removing an existing expense; you are declining to create a new long-term obligation. When founders ask cloud providers how to reduce their bills, providers may recommend commitment discounts. Those discounts can be valuable when usage is predictable and the company has sufficient cash. With nine months of runway, however, reducing waste is usually a safer first move than accepting a commitment that may outlast the business's available funds.

Scope comes fifth because this is where cost-cutting begins to affect what the company delivers. Reduce the roadmap to the next milestone investors will use to judge the business. If the original plan assumed two years of runway and aimed for three milestones, a nine-month runway may require focusing on just one. Choosing the most important milestone deliberately is better than attempting all three and delivering none convincingly.

People come last because headcount reductions are difficult to reverse and can affect the entire company. Research by Trevor and Nyberg found that downsizing was associated with higher voluntary turnover among remaining employees, partly through reduced organisational commitment. The relationship was weaker when HR practices conveyed procedural fairness.

A meta-analysis by van Dierendonck and Jacobs, covering 37 samples and 11,256 people, also found that fairness was strongly associated with commitment after downsizing. For remaining employees, procedural justice, how fairly the process was run, mattered more than distributive justice, how fairly the outcomes were shared. Amabile and Conti's study of a large technology company found that creativity and many of the conditions supporting it declined significantly during downsizing, with only a modest recovery afterwards.

None of these studies specifically measures the effects of layoffs at seed-stage startups, so their effect sizes should not be assumed to apply directly. However, the findings support a cautious approach. If layoffs are unavoidable, make the decision carefully, size it to provide enough runway to reach the next milestone plus 12 months, explain the reasons clearly, and avoid repeated rounds of reductions whenever possible. A second round just two months later can signal to the people who remain that the first decision was not based on a sufficiently considered plan.

How do you cut engineering costs in four weeks?

Start by assigning an owner to every engineering expense in week one, then remove idle and unowned spending. In week two, reduce the cost of development environments and databases, and put limits on AI usage. In week three, narrow the roadmap to the next critical milestone. In week four, make any necessary staffing decisions once, while preserving the minimum engineering capacity needed to keep production safe.

Step 1: List every engineering expense and assign an owner (week 1). Start by collecting the full cloud bill before credits are applied, along with every software licence, API key, contractor agreement and annual contract connected to engineering. Assign a named person to each expense. The FinOps Foundation's guidance on cost allocation treats ownership as an essential part of managing spending. A seed-stage startup can begin with a simple spreadsheet containing each expense, its cost and the person responsible for it. Any expense nobody claims by the end of the week should be investigated for removal.

Step 2: Remove idle and unowned spending (week 1). Start with resources nobody can justify keeping. AWS EC2 Global View provides a consolidated view of instances across AWS Regions, helping teams identify forgotten resources. Remove software seats assigned to former employees and cancel duplicate subscriptions where one tool is being paid for more than once. Before deleting a resource, preserve any data or configuration that may be needed later.

In Harness's survey, enterprises took an average of 31 days to identify and remove idle cloud resources. A small startup may be able to complete the same exercise more quickly because its cloud bill and software inventory are easier to review line by line.

Step 3: Reduce the cost of development environments and databases (week 2). Schedule development and staging environments to stop outside working hours when they are not needed. AWS allows stopped RDS databases to avoid instance-hour charges, although storage and backup charges continue. A database stopped for seven consecutive days is automatically restarted by AWS, so build a regular stop schedule rather than relying on a single shutdown.

Review production database utilisation before changing its size. If the database consistently uses far less capacity than provisioned, consider moving to a smaller instance after checking performance and safety requirements. Do not buy a long-term cloud commitment as part of this exercise.

Step 4: Set spending limits for AI and cloud usage (week 2). Give each AI tool and environment its own API key so that usage can be tracked separately. Set spending limits for each workspace or project, and configure alerts to trigger before those limits are reached. This makes it easier to identify which tool or workload is driving costs and reduces the risk of uncontrolled usage.

Apply similar controls to cloud spending. AWS Budgets Actions can apply policies or target specific EC2 and RDS instances when spending thresholds are crossed. AWS also offers its first two action-enabled budgets free each month. Check the available controls and their conditions before relying on them to prevent unexpected charges.

Step 5: Reduce scope to the next milestone (week 3). Identify the single result the company must achieve before its next funding round, then list the engineering work needed to deliver it. Pause projects, features and refactoring that do not directly support that milestone, unless they are necessary to maintain security, reliability or production operations.

This is a deliberate change in priorities, not a decision to stop building the product. The objective is to concentrate the remaining engineering capacity on the work most likely to move the business forward.

Step 6: Make any staffing decision once (week 4). If the savings from cloud, tools, AI and reduced scope are insufficient, assess whether a headcount reduction is necessary. Calculate the staffing level required to reach the next milestone and retain enough cash for another 12 months. Before finalising the decision, check that the remaining team can still maintain and recover production safely.

If layoffs are unavoidable, explain the reasons clearly and handle the process with fairness and respect. Communicate with both departing employees and those who remain. As the research discussed earlier suggests, how a company handles downsizing can influence the commitment and retention of its remaining employees.

The same four-week process works for a team of two engineers or ten, although the savings available will differ. Smaller teams may have fewer subscriptions and cloud resources to eliminate, but they also have less spare capacity and greater exposure to losing critical knowledge when someone leaves.

For founders deciding whether to keep engineering in-house or use external developers at different stages of growth, our guide to in-house versus outsourced engineering by stage explores the options beyond an immediate runway crisis.

When are cheaper engineering hours the right lever?

Cheaper engineering hours make sense when the work is clearly defined and someone on your team can review the results. Labour costs vary significantly across the EU. Eurostat's 2025 data puts hourly labour costs at EUR 12 in Bulgaria and EUR 56.80 in Luxembourg, a difference of more than four times. That makes the cost of engineering hours a legitimate way to reduce spending, but lower rates do not automatically mean lower overall costs.

We provide engineering services ourselves, so this option deserves a balanced assessment. Paying less per hour can help when three conditions are met. First, the work is clearly specified, with defined tasks, agreed interfaces and tests that establish when each task is complete. Second, someone within the company can review the work and confirm that it meets the required standards. Code that nobody can properly review creates risk, regardless of who writes it. Third, the arrangement adds capacity or replaces a role that is already being phased out, rather than removing the only person who understands how a critical part of the system works.

If any of these conditions is missing, cheaper hours may not save money. Without someone capable of reviewing the work, you could end up with code that cannot be trusted or maintained. If the specifications exist only in a founder's head, a new engineer may spend their first few weeks trying to understand the requirements. That time still costs money, even if the hourly rate is lower.

The risk is greater when a cheaper replacement takes over from the only person who understands a critical system, such as the payment process. If that person leaves before transferring their knowledge, the company can end up with a truck factor of one: losing a single individual could seriously disrupt its ability to maintain the system.

The best time to consider lower-cost engineering capacity is after completing the first five cost-cutting steps. By then, you should have removed unnecessary spending, reviewed cloud and software costs, controlled AI usage and narrowed the roadmap to the next milestone. If the remaining work still costs more each month than the company's runway allows, reducing the cost per engineering hour becomes a more informed option.

At that point, you know what needs to be built, which skills are required and what work can be handed over safely. That makes it easier to compare the cost of different engineering arrangements without sacrificing the quality or continuity of the product. Our guide to how many engineers you need to raise a seed round explores the other side of this calculation.

Where RocketDevs fits

The cuts above remove spending that does not help ship the product. If the remaining problem is the cost of engineering hours, reducing the hourly rate may be the next lever to consider. That is where RocketDevs fits.

RocketDevs provides vetted remote developers at $9.99/hr for Associate roles, $21.99/hr for Mid-senior roles and $30.99/hr for Senior roles. Each developer completes 6-8 hours of vetting before placement, and only the top 2% are accepted. The 14-day risk-free trial is money-back and 100% honoured, so you can test the fit on real tasks before making further changes to your team. RocketDevs works best when the work is clearly defined and someone on your team can review the results.

If you have completed step five and engineering costs still exceed what your runway can support, visit RocketDevs to compare developer rates and find a vetted engineer for your next milestone. Start by identifying one well-defined task, then use the 14-day risk-free trial to assess the developer’s work before expanding the engagement.

"Elite Talent. Honest Price."

Conclusion: cut waste before you cut engineering capacity

When runway is short, the goal is not to spend as little as possible on engineering. It is to spend only on the work that gives the company the best chance of reaching its next milestone before the money runs out.

Start by removing unused software, unnecessary cloud spending and poorly controlled AI costs. Then narrow the roadmap to the work that must ship next. Protect the people, systems and security measures that keep the product reliable, and avoid long-term commitments that leave you paying for capacity you no longer need.

Only after those costs are under control should you consider reducing the cost of engineering hours. Lower rates can extend your runway, but only if the work is clearly defined, the developer has the right skills and someone on your team can review the results. Otherwise, the apparent saving can turn into rework, delays and a more expensive product.

Before making another cut, calculate how much the company can afford to spend each month, identify the specific milestone that spending must deliver, and work backwards to decide what engineering capacity you need to reach it. If the remaining work is clear but your current team costs too much, compare vetted developers at different rates and test the fit on a defined task before committing further.

The measure of a successful engineering cut is not how much you remove from the budget. It is whether you extend your runway without sacrificing the milestone your company needs to survive.

Frequently asked questions

What should a startup cut first when runway is short?

Start with spending that does not help the team ship the product. This includes unused cloud environments, oversized non-production databases, software seats assigned to former employees and API keys without spending limits. Harness and Flexera surveys estimate that 21% to 29% of cloud spending is wasted. Removing unnecessary costs can reduce your burn without slowing development. Start with reversible changes and take snapshots before removing anything that might be needed later.

Should you lay off engineers or cut salaries?

Neither option is cost-free. Research by Trevor and Nyberg, published in the Academy of Management Journal, found that downsizing can increase voluntary turnover among employees who remain. A study by Jerald Greenberg found that employee theft increased in workplaces affected by a temporary 15% pay cut, but decreased when management explained the reasons thoroughly and sensitively. If cuts are unavoidable, make the decision carefully, communicate the reasons honestly and avoid leaving employees uncertain about whether more cuts are coming.

How much runway can cutting cloud costs buy?

It depends on how much of your total spending goes towards cloud services and how much waste you can remove. If cloud accounts for 10% of your monthly burn and you eliminate 25% of that spending, your total burn falls by approximately 2.5%. That is worthwhile, but it is unlikely to transform your runway on its own. Use cloud savings alongside other cost reductions, and calculate the effect on your actual monthly spending rather than assuming the savings apply to your entire budget.

What is the minimum engineering team needed to keep a product running?

As a practical starting point, aim for at least two people who can each deploy changes, roll back releases, apply security patches and restore production systems. Relying on one person creates a truck factor of one: if that person becomes unavailable, critical work may stop. A study of popular GitHub projects found that 65% had a truck factor of two or less, highlighting how common this risk is.

Your team also needs clear operating instructions, alerts that reach a named person and dedicated time for maintenance, including dependency updates and certificate renewals. Google's site reliability engineering guidance recommends eight engineers for a full single-site on-call rotation. A smaller startup may not be able to afford that many, but it should recognise the compromises and plan coverage deliberately.

Do investors look at engineering cuts in the next funding round?

Investors want to understand what the company achieved with its money, how much runway remains and whether its spending supports a credible path to growth. Fundraising can take time: Carta reported a median of 616 days between seed and Series A in Q2 2025, while Crunchbase's analysis points to timelines exceeding two years. That makes extending runway important, but cutting costs alone is not enough. The company still needs to demonstrate progress.

One metric investors may consider is the burn multiple, calculated by dividing net burn by net new annual recurring revenue (ARR). It measures how much cash the company spends to generate additional recurring revenue. Sequoia's Extending Your Runway deck also notes that investors generally view a very short runway as a bad sign. If you are preparing for your next funding conversations, see our guide to top seed-stage VC firms to identify potential investors.

When should a startup consider hiring lower-cost remote developers?

Consider it after removing unnecessary spending, reviewing cloud and software costs, controlling AI usage and narrowing the roadmap to the next essential milestone. Lower hourly rates can help extend runway when the work is clearly defined and someone on your team can review the results. Without those conditions, savings can be lost to rework, delays or maintenance problems.

Compare developers based on the skills your milestone requires, not just their hourly rates. Start with a well-specified task, assess the quality of the work and expand the arrangement only if it meets your standards.

James Hitch, COO at RocketDevs.LinkedIn

Sources

• Carta, The state of startup compensation, H1 2022, Peter Walker, 27 June 2022
• Carta, State of Startup Compensation: H1 2025, Kevin Dowd and Peter Walker, 26 August 2025
• Carta, State of Startup Compensation H2 2025, Ashley Neville and Kevin Dowd, 4 May 2026
• Carta, The new state of Series A fundraising, Kevin Dowd, 19 September 2025
• Crunchbase News, In Charts: Seed Deals Keep Getting Bigger As Odds Of Reaching Series A Fall Dramatically, Gene Teare, 26 May 2026
• Sequoia Capital, Adapting to Endure, presentation of 16 May 2022, published 14 June 2022
• Sequoia Capital, Extending Your Runway, Ravi Gupta and Sonya Huang, May 2022
• Paul Graham, Default Alive or Default Dead?, October 2015
• David Sacks, The Burn Multiple, 23 April 2020
• US Bureau of Labor Statistics, Wages and salaries as a percent of total compensation, private industry, Employer Costs for Employee Compensation, Q2 2026
• US Bureau of Labor Statistics, OEWS 2025, Software Developers 15-1252, annual median, read 19 October 2026
• O*NET OnLine, Software Developers, 15-1252.00, BLS 2025 wage data
• Eurostat, Labour cost levels, lc_lci_lev, 2025
• Electronic Code of Federal Regulations, 20 CFR Part 639, Worker Adjustment and Retraining Notification
• GOV.UK, Redundancy: notice periods and Redundancy pay
• Trevor and Nyberg, Keeping Your Headcount When All About You Are Losing Theirs, Academy of Management Journal, 2008
• van Dierendonck and Jacobs, Survivors and Victims, a Meta-analytical Review of Fairness and Organizational Commitment after Downsizing, British Journal of Management
• Amabile and Conti, Changes in the Work Environment for Creativity During Downsizing, Academy of Management Journal, 1999
• Greenberg, Employee theft as a reaction to underpayment inequity: The hidden cost of pay cuts, Journal of Applied Psychology, 1990
• Avelino, Passos, Hora and Valente, A novel approach for estimating Truck Factors, IEEE ICPC 2016
• Jabrayilzade, Evtikhiev, Tuzun and Kovalenko, Bus Factor in Practice, ICSE-SEIP 2022
• Flexera, 2026 State of the Cloud Report, 18 March 2026
• Harness, FinOps in Focus 2025 press release, 26 February 2025
• FinOps Foundation, State of FinOps 2026
• FinOps Foundation, FinOps Framework: Rate Optimization, Workload Optimization and Allocation
• Amazon Web Services, What are Savings Plans? and Returning a purchased Savings Plan
• Google Cloud, Committed use discounts overview
• Microsoft Learn, Self-service exchanges and refunds for Azure Reservations, updated 29 July 2026
• Amazon Web Services, EC2 Spot Instances, Stopping an RDS DB instance, Stop and start EC2 instances, Configuring budget actions, AWS Budgets pricing and AWS Activate credits
• GitHub Docs, Plans for GitHub Copilot and Actions runner pricing
• Zylo, 2026 SaaS Management Index press release, 29 January 2026
• Anthropic, Claude API rate and spend limits
• OpenAI, API production best practices
• Google, Site Reliability Engineering, Being On-Call
• CA/Browser Forum, Ballot SC081v3, 11 April 2025
• Let's Encrypt, Decreasing Certificate Lifetimes to 45 Days, 2 December 2025
• Black Duck, 2026 OSSRA report, 25 February 2026
• Sonatype, 10th annual State of the Software Supply Chain, 10 October 2024
• Andreessen Horowitz, The Cost of Cloud, a Trillion-Dollar Paradox, Sarah Wang and Martin Casado, 27 May 2021
• The Register, Developer pockets $2M in savings from going cloud-free, Dan Robinson, 21 October 2024