Sponsored by

OVHcloud

How To

May 29, 2024

How to grow as an AI startup without breaking the bank

The waters may be relatively calm at the moment, but how can startups make sure that they’re ready to ride out rougher waves when they arrive?

From mega rounds to high-profile commercial deals, the AI sector is enjoying huge growth. Just this month, French AI startup Holistic secured $35m in investment, and AI legaltech Leya closed a $10.5m seed round.

But, like with any sector that experiences the highest of highs, no period of growth can continue forever – and AI startups must strategise now to get ahead of a potential downturn.

“The biggest challenges to AI startups are regulation and ethics – but these also offer a big opportunity,” Filippo Sanesi, startup program manager at OVHcloud, tells Sifted. “If there's regulation, there are opportunities to serve customers because it’s there to protect customers and help companies do well. The challenge is to really understand what regulation is about and how it's going to affect the business.”

Advertisement

So when operating in a thriving but challenging sector like AI, what should startups work on to create a growth strategy? We quiz the experts about how to keep costs down. 

Tackling the cost of staying competitive 

Sri Ayangar is an investor at VC OpenOcean. He says that with the rise of transformer models and the commoditisation of machine learning (ML) models, the new generation of AI startups faces different challenges to those before them. 

Ayangar believes that data moats — what companies gain by collating unique data that competitors cannot easily replicate, like a protective castle wall around a company — are now “crucial for defensibility”.

In 2024, it's time for even early-stage businesses to adopt a multi-cloud strategy – not only for resilience in case something goes wrong, but to get the best pricing for a specific product in different cloud providers.

“Startups must find innovative ways to gain data moats through partnerships or synthetic data, enabling fine-tuning and personalisation,” Ayangar tells Sifted. 

However, Ayangar flags the challenge presented by high bills to achieve such efforts – specifically, the cost of hiring a ML team to build and maintain a product and the ML inference cost. He says that the cost of utilising these models poses challenges for CFOs and CEOs as they compete to hire top ML talent.

Sanesi agrees and adds that the high levels of expenditure have to be tackled in order to help AI startups thrive – and founders being mindful of cloud costs is crucial. He recommends that startups take a deep dive into their cloud bill to understand how much they are spending on storage, graphic processing units (GPUs) and data egress.

“Once you understand your pricing, you can optimise your costs,” says Sanesi. “Look for an alternative provider or other resources. In 2024, it's time for even early-stage businesses to adopt a multi-cloud strategy – not only for resilience in case something goes wrong, but to get the best pricing for a specific product in different cloud providers.”

Sanesi uses the analogy of having multiple bank accounts for different purposes – such as one for bills, another for savings and so on. He says an AI startup’s approach to their cloud expenditure should be similar.

“Prioritise spending money on storage for high value and crucial data, schedule your AI training and your GPU usage, and explore different pricing models,” he says.

Sanesi recommends using free credits from cloud providers, but choosing one that provides the best pricing once the credit is used up, too.

Advertisement

“If you show that you actually save costs on your cloud and reduce your CapEx, you can show investors that you're lean – and that could boost your valuation,” he adds.

The power of open source

Open source refers to source code that is made available to developers to use or modify – an example of this is Mozilla Firefox, a customisable internet browser.

Open source is community based and can provide more transparency on how things are done, building trust.

Lex Avstreikh, head of strategy at Hopsworks, says open source has the unique ability to help a technology reach a much wider audience — potentially faster — and has two inherent qualities that can help address visibility and validation.

“[Open source offers] visibility as it is simply more accessible; no one needs to contact the sales team or ask their supervisors to test something at home as there is no procurement process required to test an open source tool,” Avstreikh tells Sifted. “Validation by the market happens as open source directly collides with the needs of the market; if no one asks you about your open source technology or is trying to use it, do you really solve any problem worth solving?”

Sanesi believes AI startups should embrace open source because it can speed up development, reduce costs thanks to its interoperability and allow startups to own their data in a stack without being locked in with a vendor.

“Open source is community based and can provide more transparency on how things are done, building trust,” says Sanesi. “The open source community is massive and growing – why would you pay for something that's free?”

Embed sustainability ‘from day one’

Sanesi says there's often a misalignment with early-stage founders who are trying to focus on solving a problem, serving a customer and building a product – and don't often have sustainability in mind, seeing it as a luxury rather than a must-have.

Prioritise showing investors and customers that you care about the topic because the fact is, sustainability in general means using less. If you use less and if you optimise, you spend even less.

“It is necessary for startups in 2024 and beyond to embrace sustainability,” says Sanesi. “By this, we don’t just mean environmental sustainability, but rather it’s about making sure your organisation thinks long term. For example, having compliance in mind from day one doesn't mean that you have to become an expert, but trying to understand how you design your AI, algorithm and model so you can explain the output should be at the core.”

He says this approach of having a focus on product and customer while building with compliance in mind gives AI startups an edge to be sustainable in the long term.

“You have to show you have an optimised data strategy – focus on data quality rather than quantity and prioritise unbiased data,” adds Sanesi. “Prioritise showing investors and customers that you care about the topic because the fact is, sustainability, in general, means using less. If you use less and if you optimise, you spend even less.”

For example, OVHcloud has been developing, using and refining a water cooling system for its server since 2003 – in doing so, the cloud provider has less energy consumption and therefore lower energy bills. Sanesi says such efforts mean benefits can be passed down to customers.

“We have embedded sustainability since day one without compromising our business model and product,” says Sanesi. “As a result, we can now provide competitive pricing for our customers. It should be the same for all startups, and they need to get this right from the beginning.”

If you're building a successful, sustainable, innovative startup or scale-up, you can benefit from up to £100k in cloud credits with OVHcloud. Learn more about the program here