The world’s largest tech firms show no signs of easing up on artificial intelligence spending, a record wave that is propelling hardware providers like Samsung and SK Hynix.
It comes despite persisting doubt about the staying power of AI demand to justify all that capital.
Meta Platforms alone revealed ambitions to spend as much as $135 billion this year – one of the biggest planned outlays of the business sphere. Its suppliers have responded in kind: on Thursday, SK Hynix said it plans a “considerable increase” in capex, and Samsung said it’s ratcheting up spending on its memory production capacity.
Investors continue to reward ambition – so long as companies can show growth. Microsoft shares slid 6.1 per cent after revealing a slowdown in cloud services growth, while Meta climbed 6.6 per cent.
A procession of industry linchpins reported results this week that underscored how voracious the appetite for AI hardware has grown – and how that is likely to extend well into 2026.
Meta, Microsoft and fellow hyperscalers such as Amazon and Alphabet are driving a wave of global spending on chips, servers and computers that is firing up hardware suppliers around the world, particularly in Asia.
Samsung and SK Hynix – two makers of the memory chips that go into Nvidia’s essential AI accelerators as well as data centre servers – reported profits that grew several-fold. ASML Holding NV, the only provider of the cutting-edge lithography machines needed for advanced semiconductors – also smashed estimates.
The “real headline is the relentless ramp” in capital expenditures, said Matt Britzman, an analyst at Hargreaves Lansdown. “It wouldn’t be surprising to see the share reaction cool as investors absorb those aggressive investment plans,” he said, referring to Meta.
At the same time, that enormous demand is worsening a global chip demand-supply imbalance that threatens to disrupt industries from smartphones and electronics to cars.
While the appetite for Nvidia and Advanced Micro Devices accelerators needed to develop and operate AI has long outstripped supply, investors are growing increasingly concerned about a similar deficit in more basic memory.
The availability of semiconductors will be a big bottleneck to growth for companies including Tesla, and may necessitate building a Tesla TeraFab – a factory that can make logic and memory chips and provide packaging, Elon Musk said in a recent podcast with Peter Diamandis, founder of the X Prize Foundation.
“We’re going to hit a chip wall if we don’t do the fab,” Mr Musk said. “We’ve got two choices: hit the chip wall or make a fab.”
Memory manufacturers are reallocating production lines towards lucrative HBM to satisfy the needs of AI data centres. Because HBM – high-bandwidth memory – requires about three times the wafer capacity of standard Dram – dynamic random-access memory – for the same amount of memory, this shift has reduced supply for the consumer electronics industry. That’s threatening double-digit price increases for PC makers and smaller electronics companies.
Concerns about end-demand for AI persist, however. Microsoft’s capex grew a larger-than-anticipated 66 per cent in the quarter, but its Azure cloud-computing unit posted a 38 per cent revenue gain – one percentage point slower than the prior three months.
Meta chief Mark Zuckerberg talked about “a major AI acceleration” that has been brewing within the tech industry for over a year. “I expect our first models will be good, but more importantly, we’ll show the rapid trajectory that we’re on,” he said.
Samsung’s shares were down about 0.4 per cent during trading in Seoul on Thursday, even though it said it plans to start shipments of its next-generation high-bandwidth memory, HBM4, in February – a key step in its bid to catch up with SK Hynix in the lucrative segment. SK Hynix’s stock rose about 2 per cent.
Samsung’s chip unit reported a five-fold gain in profit, well above the average of analyst estimates. It also said it would buy back 3.57 trillion won ($2.5 billion) of its shares, and announced a special dividend payout that raises its fourth-quarter payout to 3.75 trillion won.
In Asia, attention is on the race for leadership in next-generation HBM4, which is set to be integrated with Nvidia’s upcoming flagship Rubin processors. Samsung is close to obtaining certification from Nvidia for the latest version of its AI memory chip.
The AI building rush is also helping to bolster Samsung’s foundry business, which competes with industry leader Taiwan Semiconductor Manufacturing. Samsung’s contract chipmaking arm expects two-nanometer orders to rise about 130 per cent in 2026, with the company in active discussions with customers in the US and China, executives said.
