Griffin on Tech: Lessons from Nvidia’s record run as Intel languishes
Chip maker Nvidia on Wednesday became only the third company in history to reach a US$3 trillion valuation.
Not bad for a company that for most of its 30-year history was a respected but niche player, supplying (GPUs) Graphics Processing Units primarily for video games. In late 2019 Nvidia’s market cap on the Nasdaq was US$100 billion.
Then came the generative AI boom which presented a compelling use case for Nvidia’s GPUs and sparked a flurry of orders for chips that continues to grow. Nvidia supplies over 70% of chips for AI applications, the bulk of those orders from big data centre operators like AWS, Microsoft, and Google.
Competitors, including those big public cloud players themselves, are now hot on Nvidia’s heels developing their own chips to avoid paying high prices for Nvidia’s. But with a few years’ head start, Nvidia has scope to reap healthy profits and maintain its market value - as long as belief in the promise of AI holds up.
Rise of the GPGPU
The big turning point for Nvidia, from owning a lucrative niche, the GPU market, to becoming the go-to chip maker for all things AI, was a decision it made in 2006 to release CUDA, a programming language that allowed its graphics cards parallel processing capabilities to be used for more general purpose computing processes. It spawned the rise of the General Purpose Graphics Processing Unit (GPGPU)
As Vox reported, Nvidia’s chips “could now do a lot of heavy lifting for tasks unrelated to pumping out pretty game graphics, and it turned out that graphics cards could multitask even better than the CPU (central processing unit), what’s often called the central ‘brain’ of a computer”.
While scientists clocked the potential of this early for maths-heavy calculations, the chip orders really started to stack up for non-gaming and visualisation applications when Bitcoin miners realised Nvidia’s chips could help them complete calculations faster and add blocks to the Bitcoin blockchain, gaining bitcoin in the process. Before China clamped down on Bitcoin mining operations, charter jets full of Nvidia GPUs were making their way to China on a weekly basis.
What happened to Intel?
It was a logical move to apply the GPUs to training the large language models that started to emerged around 2020 - 2021. Now Nvidia’s chips power ChatGPT, the AI functions of Tesla electric vehicles, and many of the AI functions of X and Meta.
Around 20 years ago, during a break in a visit to Intel’s fabrication plant in San Jose, I experienced my first earthquake, a feeling that would become all too familiar following my move to Wellington in 2005. Back then Intel was still the reigning chip maker for computers and servers, but was about to see a seismic shift of a different kind in the market it dominated.
The era of the smartphone had arrived but Intel saw it as a flash in the pan and didn’t devote enough resources to developing chips customised for the power and processing needs of mobile devices. Rival ARM filled that need and by the time Intel had chips in the market, it was too late. Apple went its own way and developed its own chips.
A strength becomes a liability
A series of management missteps and technical deadends followed for Intel, which also saw its biggest strength become its weakness. As both a designer and manufacturer of chips, Intel could capture more value and tightly integrate its divisions for faster innovation and efficiency. But that structure has now become a liability as Intel faces competitors specialising in either design or manufacturing, and out-innovating Intel on both fronts.
Nvidia doesn’t manufacture its high-end chips for AI, but outsources the job to Taiwan’s TSMC, the world’s biggest chip manufacturer. TSMC has picked up market share from Intel.
As Reuters’ Robbery Cyran pointed out in September: “With demand falling, Intel’s factories are running below full capacity. Its operating margin is projected to be only 5% this year, according to LSEG data, down from over 30% in the company’s heyday. TSMC’s margin is expected to be over 40%”.
The only way now for Intel to survive may be to split the business in two and aggressively court chip designers who want an alternative to Nvidia and are looking for a manufacturing partner.
Avoiding the Kodak moment
Intel needs some massive technical wins to become a force in the AI market too. Its market cap is currently languishing at US$131 billion. I don’t think Intel will become another Kodak, it still has a healthy PC and server chip business. The US wants to revive manufacturing and is tipping billions into the semiconductor industry. But Intel has a hugely capital-intensive operation and a lot of competition, so the next few years are pivotal to its future.
It helps that Nvidia has a visionary leader in Jensen Huang who has been there from the start over 30 years ago. He saw the potential to apply Nvidia’s chip technology to AI and deep learning very early. Intel was too preoccupied with internal problems and beholden to its outdated business model.
I wouldn’t write off Intel, but Nvidia has huge momentum, maybe even too much to avoid a fall if the US economy falters and AI uptake turns out to be lower than forecast.
