For the last year, chip makers have painted a target on Nvidia’s back. In the next year, the company might face stiff competition to its graphics chips, which are the gold standard for running software that understands speech and learns people’s online preferences.
That has raised questions about how long Nvidia’s spellbinding growth can last. But there are few signs that the company feels strained. On Thursday, its quarterly earnings showed revenue and profit surging past Wall Street estimates.
Nvidia holds a formidable lead in chips that not only act like graphics engines in video game consoles but also like accelerators for machine learning code in data centers and autonomous cars. But no one knows if general-purpose chips or customized silicon will conquer machine learning. That gives Nvidia’s rivals – and major customers – some breathing room.
In May, Google revealed its new tensor processing unit – more commonly called the TPU – which has been upgraded to run algorithms through training exercises and make inferences when presented with new data. Previously, Google only had custom silicon for inferencing. For years, Nvidia has touted its ability to both train and apply models as a unique feature.
This year, Intel plans to release its Lake Crest chip using expertise it acquired from Nervana Systems. Advanced Micro Devices has new graphics chips waiting in the wings. Others support FPGAs that can be changed to reflect advances in neural networks, which enable deep learning. Still others have thrown weight behind digital signal processors.
The mounting pressure does not seem to bother Nvidia’s chief Jensen Huang, who describes his executive style as part cheerleader and part swashbuckler. Last year, he dismissed Google’s claims that its new accelerator – which falls into the category of ASICs – beat Nvidia’s chips at inferencing tasks.
Huang cast Google’s silicon as an imperfect clone in a Thursday call with financial analysts. “Nvidia’s GPU is basically a TPU that does a lot more,” he said, adding that versatility is vital for handling varying tasks in servers. “The cloud is not an appliance. It is not a toaster or a microphone.”
But investors fear that hardware customized for machine learning could start nibbling at Nvidia’s business in the long term. In addition to traditional rivals, many start-ups are trying to siege its stronghold from Graphcore – which is readying a unique graph processor for servers – to Groq – the secretive firm founded by former Google chip makers. Others target embedded intelligence, which could make Nvidia’s technology less essential.
These concerns hardly dampened Nvidia’s second quarter. The company reported profit of 92 cents per share versus estimates of 70 cents. Profits more than doubled to $583 million, up from $261 million in last year’s second quarter. Revenue rose to $2.23 billion from $1.43 billion a year ago.
The continued financial success comes as more businesses try tapping into vast storehouses of data to identify serious illnesses or organize vacation photos. Nvidia’s graphics chips runs such programs more efficiently by dividing software into smaller bites that can be processed simultaneously.
Nvidia predicts that its customers will need increasingly powerful silicon to handle neural network complexity that doubles every year. The company’s data center unit appears to be reacting as revenue grew to $416 million in the quarter, up from $151 million a year before, according to Nvidia’s accounting.
In the second quarter, Nvidia released new accelerators based on its Volta architecture, which contain specialized cores that run 120 trillion operations per second on deep learning workloads. The 21.1-billion-transistor chip started shipping last month under the Tesla V100 brand.
The complexion of Nvidia’s business has seen little change. Gaming sales account for around half its revenue, helped by strong demand for its chips in consoles like Nintendo’s Switch. The unit reaped revenue of $1.186 billion, up from $1.027 billion last quarter, and $781 million in last year’s second quarter.
Still, Nvidia’s executives see other opportunities to entice investors. In addition to supplying server chips, the company is also aiming to sell chips that control self-driving cars with minimal human training. In recent months, it signed deals with Baidu and Toyota as well as Volvo, which plans to sell fully autonomous cars by 2021.
Nvidia’s automotive sales totaled $142 million in the second quarter, up from $119 million in the same period a year ago. The company said that its Drive PX system is used by around 225 companies. But its automotive efforts helped drive operating expenses to $533 million, up 19% from the same span last year.
Nvidia’s graphics chips are also ideal for running algorithms that mine digital currencies like Ethereum. Over the last year, developers have emptied store shelves of products traditionally bought by gamers. Nvidia plans to release a new graphics card specifically for mining the blockchain.
“Our strategy is to stay alert to this fast-changing market, knowing that GPUs are highly efficient at running the algorithms used to mine cryptocurrencies,” said Nvidia’s chief financial officer Colette Kress. “This is a market that is not likely to go away soon,” added Huang.
The sum of these circumstances has more than tripled Nvidia’s share price in the last year. Nvidia’s potential winnings from machine learning have also excited investors like Softbank – the Japanese investment firm that also owns Arm – which this week disclosed that it bought 4.9% of Nvidia’s stock.
Despite the good vibes, Nvidia’s earnings elicited an incongruous response from investors. Nvidia’s stock price fell around 5% after it reported earnings on Thursday. The stock continued to fall from $157 after the market closed. The company predicted third quarter revenue of $2.35 billion.