Deeptech startups face a singular administration problem: how do you construct a enterprise round a know-how that may take years to develop, when market tendencies and buyer wants might shift each few months?
Massive firms like Nvidia, Google, or Meta can tackle these sorts of issues with massive, extremely specialised engineering groups and deep reserves of capital. Even then, success isn’t assured. Startups don’t have that luxurious. They function with restricted sources, and any sudden change out there can render years of painstaking work irrelevant.
Typical startup knowledge suggests analysing the market, figuring out a distinct segment, after which constructing a product to suit it. However in deeptech, this mannequin usually breaks down. By the point the know-how is prepared, the “good area of interest” might have disappeared or developed into one thing fully totally different.
Why put money into an engineering problem over a market area of interest
When GPU Audio launched in 2017, the group made a deliberate selection: slightly than chasing an present area of interest, they determined to resolve a elementary engineering downside that would survive a number of shifts out there. Their focus was on adapting graphics processing items (GPUs), which have been initially designed for parallel rendering of graphics, to course of audio knowledge in actual time.
It wasn’t a fast win. Between 2012 and 2015, the group constructed a number of prototypes and failed extra usually than they succeeded. A turning level got here in 2017, when a number one researcher — a former advisor to Qualcomm’s founders and a recognised authority in ray-tracing engine design — joined the mission.
Even then, it took years earlier than the primary working demonstrations appeared in 2020 and 2021, practically a decade after early experiments started. Throughout that point, GPU architectures modified, AI went mainstream, and shopper tendencies shifted dramatically. However the group stayed targeted on their core technical aim, resisting the temptation to pivot towards short-lived alternatives.
On this case, prioritising long-term technological resilience helped the group navigate a number of market cycles and uncover alternatives that weren’t on the radar at the beginning. For groups in comparable conditions, it reveals that this path — whereas dangerous and demanding an extended planning horizon — can typically open sudden doorways.
The depth and complexity of deeptech engineering challenges
The obstacles GPU Audio confronted illustrate why deeptech startups should usually assume in a different way. Two issues specifically stood out.
First, GPUs have been by no means designed for audio. For graphics, small delays are acceptable — a dropped body might go unnoticed. However audio requires near-instant precision. Even a tiny delay of only a few milliseconds could be audible, disrupting the expertise fully.
Second, the best way sound is processed is essentially totally different from graphics. GPUs excel at dealing with thousands and thousands of similar, impartial duties in parallel — the computational equal of a manufacturing unit filled with employees all performing the identical motion concurrently. Audio, alternatively, is a sequence of small, interdependent steps. Every calculation will depend on the results of the one earlier than it. Getting GPUs to deal with this sort of workload required greater than optimisation — it demanded a reinvention of how audio processing itself may very well be structured.
Additionally Learn: From pilot to scale: Why conventional VC metrics don’t work for local weather deeptech
Challenges of this scale have an effect on extra than simply the engineering roadmap — they form how a group operates. Lengthy growth cycles name for clear, ongoing communication with each traders and inside groups. In our case, we made some extent of sharing intermediate milestones to maintain everybody aligned, even when the street to a completed product stretched over years.
Creativity in engineering: Borrowing concepts from adjoining fields
With out the deep pockets of a tech big, GPU Audio wanted greater than persistence; they wanted creativity. The group needed to rethink audio algorithms from scratch, trying to find methods to interrupt down sequential duties into parallelisable ones.
The breakthrough got here by borrowing concepts from one other area: ray tracing in 3D graphics. The corporate’s chief scientist had in depth expertise on this discipline, having constructed one of many quickest ray-tracing engines on the planet. Ray tracing calculates reflections, shadows, and interactions throughout numerous objects directly — issues not not like the a whole lot of processes required in real-time audio.
Making use of these ideas, the group constructed a brand new sort of audio course of supervisor — a scheduling system that would mixture audio streams, distribute workloads effectively, and preserve the responsiveness required for real-time sound. What the business had lengthy dismissed as technically unimaginable out of the blue turned possible.
Shifting from shopper merchandise to developer SDKs
Fixing the core technical downside was solely half the battle. Subsequent got here the query of product-market match. Breakthrough engineering doesn’t mechanically translate into buyer demand — particularly in markets the place preferences shift shortly.
Initially, GPU Audio launched shopper software program for musicians and sound engineers, instruments that ran on GPUs slightly than CPUs. Whereas helpful, this method wasn’t scalable. The group realised that as an alternative of constructing end-user merchandise themselves, they may multiply their attain by providing a software program growth package (SDK) to different utility builders.
Additionally Learn: Funding deeptech: Balancing potential and complexity within the seek for capital
This shift made the know-how extra versatile, much less tied to short-term shopper tendencies, and much more engaging to potential companions. It additionally created a pathway into industries that the founders hadn’t initially focused. For some deeptech startups, shifting from direct-to-consumer merchandise to an SDK mannequin can present extra flexibility and make it simpler to maintain tempo with altering business wants — that was the case right here.
Discovering unobvious alternatives
Shifting to an SDK unlocked a stunning new vertical: the automotive business.
Automobile audio techniques have been lagging behind broader automotive innovation. Whilst electrical autos, superior infotainment techniques, and autonomous driving turned extra subtle, most in-car audio processing nonetheless relied on outdated DSP chips. GPU Audio noticed a chance to modernise this layer.
The corporate developed zoned audio know-how — permitting totally different passengers to listen to totally different content material concurrently with out interference. A driver might take a name over the entrance audio system whereas kids within the again seat loved a film, all with out overlapping sound.
This innovation didn’t simply enhance in-car leisure; it opened the door to thoroughly new use instances, from personalised multimedia techniques to interactive voice-based providers. It additionally confirmed how a deeptech startup might scale by partnering with established industries, repurposing its core know-how to fulfill wants far past the unique imaginative and prescient.
The takeaway right here is to remain open to markets past the unique goal section. Usually, significant scale comes from industries that haven’t but gone by way of a full wave of innovation.
—
Editor’s be aware: e27 goals to foster thought management by publishing views from the group. Share your opinion by submitting an article, video, podcast, or infographic.
Loved this learn? Don’t miss out on the following perception. Be part of our WhatsApp channel for real-time drops.
Picture courtesy: Canva Professional
The put up Deeptech’s secret: Ignore the market, grasp the engineering, and let alternative discover you appeared first on e27.
