Richard Waters

Can today’s tech giants adapt fast enough to the age of AI?

Like all disruptive technologies, artificial intelligence has the potential to upset the processes, products and business models on which today’s most successful companies are founded. Computing platform shifts such as this usually leave at least some industry leaders out in the cold: today’s incumbents are all racing to make sure that doesn’t include them.

Two events this week show the different ways in which AI is coming to dominate Big Tech’s agenda.

First, Meta announced a second round of job cuts, jettisoning 10,000 workers on top of the 11,000 it said it would let go late last year. Along with scrapping 5,000 more open positions, this has gone a long way to assuaging Wall Street’s anger over the company’s earlier refusal to dial back its heavy spending in the face of an advertising slowdown.

Chief executive Mark Zuckerberg made clear, though, that there was a lot more to this latest overhaul than simply slashing costs. In his view, Meta has to slim down, cut layers of management and move much faster to capitalise on big changes sweeping through the tech industry — with AI the most significant.

It is less than 18 months since Zuckerberg renamed his company to reflect the central role virtual worlds played in his vision of the future, but priorities have clearly shifted since then.

AI already represented Meta’s biggest single tech investment, Zuckerberg said this week. He also pointedly singled it out ahead of the metaverse when discussing the technologies with the most importance to Meta’s business.

AI had a key role to play in the company’s core services by improving the way people “creatively expressed” themselves and found content, he said, as well as boosting the productivity and speed of the company’s engineers. As the Financial Times reported late last month, AI is also now an important engine of Meta’s advertising, helping to overcome some of the damage to its ad-targeting capabilities caused by Apple’s privacy changes.

Of course, AI is an important technology behind the metaverse as well, and Zuckerberg is hardly backing down from his long-range VR vision. But in terms of the issues that have the most immediate impact on his company’s competitiveness — countering the rise of TikTok, boosting engagement, lifting revenues, honing the effectiveness of Meta’s engineering teams — AI is now central.

Microsoft, meanwhile, faces a very different AI challenge. While Zuckerberg is trying to re-equip his company while also managing a significant retrenchment, Microsoft has a head start. The close ties it forged with OpenAI three years ago have put it in prime position roll out generative AI to the masses. The question now is to change the way hundreds of millions of people use its software.

This week, Microsoft outlined how it plans to push new AI features into widely used software applications such as Word and Excel. This means putting generative AI tools into the hands of workers to make them more creative; allowing people to control their applications using natural language commands; and introducing software “agents” that can dig through a user’s work to find and highlight relevant material.

This sounds all well and good in theory, but are the world’s “information workers” ready for the big changes in work practices this technology will bring? And will it really make them more productive?

Generative AI has quickly become something no tech company wants to be left without. There is a race on to implant it into widely used software and internet services. It will take trial and error to find out which of promises that have been made for the technology pan out, and which are empty hype.

The optimists claim that AI lends itself better than other technologies to mass adoption. The viral success of ChatGPT shows the appeal of using natural language to control computers. This could have special relevance for Microsoft’s customers: as the company has added more features to its applications, making them more complicated, workers have struggled to keep up. Life might be much easier when all software responds to simple natural language commands.

But Microsoft has bet on supposedly transformative new computing interfaces before and come up wanting, from redesigning Windows to be touch-first (PC users hated it) to promising that previous voice assistants such as Cortana would completely change computing. Will this time be different? As Microsoft and the rest of Big Tech rush to show that AI can make their services newly relevant, we are about to find out.

richard.waters@ft.com

The flood of cash into the tech start-up scene in recent years has led to persistent warnings of disaster. Most often, these have turned on the kind of meltdown that hit Silicon Valley at the turn of the century, when a stampede to make money on the early internet led to massive over-investment.

It’s safe to say, though, that none of the disaster scenarios envisaged the kind of financial implosion that struck this week at SVB Financial, the parent company of Silicon Valley Bank.

As an institution that is estimated to work with half of local tech start-ups, its collapse presents an obvious threat. It led the head of Y Combinator, San Francisco’s pre-eminent accelerator for early stage tech companies, to warn on Friday that Silicon Valley’s start-ups could be facing an “extinction-level event”.

In many ways, this looks like a familiar tale in the banking world: In pursuit of higher returns, SVB failed to notice what, in hindsight, seems an obvious flaw in its risk management. Its assets soared nearly three-fold in the space of three years as capital poured into start-ups and was deposited, in turn, at the bank. SVB put much of the money into longer-maturity bonds to generate a higher return. When interest rates rose, the market value of those investments slumped, leaving the bank with losses that, on paper, stood at $15bn at the end of last year.

Rather than sell the bonds and take a hit, SVB hoped to nurse its low-yielding bond portfolio through to maturity, suffering lower net interest margins along the way. The plan might have worked. But it emerged this week that the bank’s start-up customers, facing more difficult times, had been drawing down their cash, forcing it to sell investments and take a loss. The resulting need for more capital set alarm bells ringing and led to a flight by depositors: By Friday morning, regulators had to step in and close SVB down.

Heading into the weekend, it was impossible to tell exactly how deeply this financial shock would hit tech start-ups with deposits that have now been frozen. SVB’s surplus capital at the end of last year was roughly enough to absorb its notional losses at that stage. Even after a further $1.8bn hit it reported this week, the losses still look modest in the context of a total deposit base that stood, in December, at $173bn (though $42bn flew out the door on Thursday alone.)

Yet the losses could well escalate as regulators carry out a forced sale of the bank’s assets. Even more damaging, for many start-ups, is the risk that their much-needed cash will be locked up indefinitely, leaving them unable to meet immediate commitments like staff salaries and forcing some to close their doors.

There has been no shortage of people seeking to turn this into a Silicon Valley morality tale. To some, it is another example of the tech world’s hubris, and proof that the good times blinded the tech industry to some very real risks. Why, for instance, did a public company like streaming video outfit Roku leave $487mn on deposit at what until recently barely counted as a medium-sized bank in US terms?

To others, meanwhile, the fallout from the SVB collapse is a reminder of how Silicon Valley, which usually fights hard to escape the heavy hand of government regulation, is quick to ask Washington for support when a crisis hits. Tan, the tech accelerator boss who warned of extinction, urged tech entrepreneurs to write to their local Congressional representatives calling for immediate government help.

By late Friday, the finger-pointing also had begun. The run on the bank that laid SVB low has been held up as an example of the herd-like behaviour often displayed by tech investors. A number of venture capital firms urged companies they had invested in to take their cash out of SVB after the bank said it was seeking to raise more capital. A partner at one prominent venture firm told me such withdrawals had caused a crisis that was entirely avoidable. Meanwhile, more than a dozen VC firms had banded together to promise they would stand behind SVB in future, should another institution step in to bail it out — though a number of well-known Silicon Valley firms were not part of the group.

The scramble among VCs underlined a dawning sense that, if SVB is wound up, something irreplaceable may be lost. One investor described the bank as “like a left ventricle” for Silicon Valley’s financial scene — not as visible as the VCs which supply the risk capital that has floated the modern tech industry, but vital to the sector’s smooth functioning. It was founded 40 years ago to fill the void left by big banks that often baulked at lending to start-ups. The VC firms that banded together on Friday night hope that it’s not too late revive the bank. But if it is, Silicon Valley will have lost an institution that has played an important role in its rise.

richard.waters@ft.com

The price of the most important raw material feeding the latest artificial intelligence boom is collapsing fast. That should push the technology into the mainstream much more quickly. But it also threatens the finances of some of the start-ups hoping to cash in on this boom, and could leave power in the hands of a small group.

The raw material in question is the processing power of the large language models, or LLM, that underpin services such as ChatGPT and the new chat-style responses Microsoft recently showed off for its Bing search engine.

The high computing costs from running these models have threatened to be a serious drag on their use. Only weeks ago, using the new language AI cost search engine You.com 50 per cent more than carrying out a traditional internet search, according to chief executive Richard Socher. But by late last month, thanks to competition between LLM companies OpenAI, Anthropic and Cohere, that cost gap had fallen to only about 5 per cent.

Days later, OpenAI released a new service to let developers tap directly into ChatGPT, and slashed its prices for using the technology by 90 per cent.

This is great for customers but potentially ruinous for OpenAI’s rivals. A number, including Anthropic and Inflection, have raised or are in the process of trying to raise cash to support their own LLM ambitions.

Seldom has a technology moved straight from research into mainstream use so rapidly, prompting a race to “industrialise” processes that were developed for use in lab settings. Most of the gains in performance — and reduction in costs — are coming from improvements in the underlying computing platform on which the LLMs run, as well as from honing the way the models are trained and operate.

To a certain extent, plunging hardware costs benefit all contenders. That includes access to the latest chips specifically designed to handle the demands of the new AI models such as Nvidia’s H100 graphics processing units or GPUs. Microsoft, which runs OpenAI’s models on its Azure cloud platform, is offering the same facilities — and cost benefits — to other LLM companies.

Yet large models are as much art as science. OpenAI said “a series of system-wide optimisations” in the way ChatGPT processes its responses to queries had brought costs down 90 per cent since December, enabling that dramatic price reduction for users.

Training an LLM costs tens of millions of dollars, and techniques for handling the task are changing fast. At least in the short term, that puts a premium on the relatively small number of people with experience of developing and training the models.

By the time the best techniques are widely understood and adopted, early contenders could have achieved a first-mover advantage. Scott Guthrie, head of Microsoft’s cloud and AI group, points to new services such as GitHub Copilot, which the company launched last summer to suggest coding ideas to software developers. Such services improve quickly once they are in widespread use. Speaking at a Morgan Stanley investor conference this week, he said the “signal” that comes from users of services such as this quickly becomes an important point of differentiation.

The main hope for rival LLM makers comes from selling the extra services needed to make the technology easier for developers and big corporate customers to use, as well as from the creation of more narrowly-targeted models that fit particular business needs.

When it unveiled its latest LLM this week, for instance, Israeli start-up AI21 Labs also announced a series of APIs — or programming links — for higher-level services such as text summarisation or rewriting.

Rather than broad-based models such as the one behind ChatGPT, most companies will want to use models that have been trained on the language used in particular industries like finance or healthcare, or on a company’s own data, said Ori Goshen, AI21’s co-CEO.

As he noted, large language models are in their infancy. There is still plenty of work to be done on reducing their tendency to spew falsehoods and prevent them from “hallucinating”, or generating plausible-sounding answers that have no bearing on reality. To succeed, the AI companies will have to stay on the cutting edge of research.

But it is still the case that the underlying costs of these generative AI services are tumbling. OpenAI’s price cut is a sign of how quickly the new technology is moving into mass adoption, and a warning sign that this may be a business with few producers.

richard.waters@ft.com

There is a power law at work in many mass-market internet services. A relatively small proportion of users often account for a disproportionate share of activity, whether that means posting on social networks or selling on eBay.

Finding better ways to align the economics of the service with the interests of those users is a good way to boost profits. But it also risks upsetting the delicate balance that made the services appeal to large numbers in the first place.

That risk is worth bearing in mind as social media companies search for new ways to generate revenue. Meta is the latest to join the hunt, following Twitter and Snap with the announcement this week of new subscription tiers for its Facebook and Instagram services that will cost web users $11.99 a month.

The wide range of features the companies have included in their subscription offerings highlight that this is a period of experimentation. They have yet to work out what they should or shouldn’t charge for, as they try to boost profits while at the same time protecting the overall health of their networks.

One idea is to let paying customers see less advertising, as Twitter has promised. This may have strong appeal for some, but it amounts to an admission that the ad-filled experience delivered to the majority of people is inferior — not a message that will be welcomed by the advertisers who are paying most platforms’ bills.

The existence of an “ad-lite” or even ad-free tier also reduces the incentive to improve the experience of “free” users who don’t get this relief. There is an assumption that if they are unhappy, they can always switch to a subscription.

A second common theme is building a higher level of security into subscription offerings. Meta says it verifies subscribers’ accounts and monitors them to prevent impersonation, while next month Twitter will only allow paying customers to use text messages for two-factor authentication of their accounts.

There is some logic to giving higher protection to power users, since they are most likely to have their accounts hacked or suffer impersonation. But it leaves the impression that only subscribers deserve an adequate level of security, and it again reduces the incentive to improve the experience of “free” users.

The third approach is giving some users special privileges that increase their influence — but this risks detracting from everyone else’s experience. This trade-off is not new: LinkedIn has long let subscribers send direct messages to anyone they want, a privilege not given to everyone to prevent mass solicitations.

Meta’s new algorithms will single out subscribers for special attention, making their profiles appear more prominently in search results and spreading their posts more widely. Twitter, which does something similar, says this will “reduce the visibility of scams, spam and bots” on its network — implying that the output of all its non-paying users has been assigned to the same category of unwanted dross that it hopes to expunge from its network.

Besides creating an us-and-them division within services that always prided themselves on their “democratic” nature, this approach risks eroding the quality of content most users see. Paying users aren’t inherently wiser, wittier or more virtuous than others. The idea echoes the early days of internet search, when some search engines looking for a way to make money mixed paid searches with their “organic” results.

Beyond ideas like these, there is a class of services that wouldn’t give power users an undue level of influence, but that such users would still welcome. The most obvious are analytical tools that help people monitor the reach of their posts and how others are interacting with them, and tools that enhance the user experience or improve the quality of posts. Snap subscribers, for instance, have a variety of ways to customise their experience on the service, while Twitter Blue users can edit tweets within 30 minutes of posting.

Whether any of these measures will have a meaningful effect seems questionable. Before its acquisition by Microsoft, LinkedIn generated only 17 per cent of its revenue from premium subscriptions, even though its status as a professional network put it in a strong position to charge users. There is obvious value, after all, in paying for features that help you build a professional network or generate sales leads. On mass-market consumer networks, subscriptions may bring some extra revenue at the margin — but they are unlikely to make much of a dent in social media’s heavy advertising dependence.

richard.waters@ft.com

Nvidia said it is moving faster to a “new business model” of selling AI services directly to large companies and governments, potentially putting it on a collision course with the big tech companies that are its largest customers.

The move comes as some of the leading tech companies are designing chips to handle the huge data-crunching demands of AI, reducing their need for Nvidia’s chips in the long term.

Nvidia chief executive Jensen Huang on Wednesday said the company had started to make its AI services available through the cloud platforms of groups including Google and Microsoft. Both of them are among the biggest buyers of Nvidia’s chips to train their AI models, while Google now also designs its own AI chips and Microsoft is widely expected to follow suit.

The chipmaker has said its services, which include selling access to supercomputers to train AI models and supplying its own pre-trained large AI models, will soon generate “hundreds of millions of dollars” of revenue, opening up a new software business to add to its existing chip sales.

Nvidia’s increasingly complex relationship with its largest customers comes as generative AI services such as OpenAI’s ChatGPT have opened a big new market for AI chips. In a call with analysts to discuss Nvidia’s latest earnings on Wednesday, Huang said generative systems had brought an “inflection point” after a decade of work on artificial intelligence that has led a much wider range of companies to start experimenting with the technology.

Nvidia’s share price has taken off this year on hopes that its position as the largest supplier of the GPUs used to train large AI models will make it one of the biggest beneficiaries of generative AI. The shares jumped another 8 per cent in after-market trading on Wednesday, taking the gain this year to 54 per cent, after the company issued a forecast that suggested it was recovering from a downturn in its gaming business since the peak of the coronavirus pandemic.

The company said it expected revenue this quarter to reach $6.5bn, above the $6.2bn that Wall Street had been expecting, as it starts to bounce back from a sharp downturn in the second half of last year. Sales of graphics cards used for gaming fell 46 per cent in the latest quarter, to $1.83bn, but were still ahead of most forecasts.

For the final quarter of its latest financial year, Nvidia reported pro forma earnings of 88 cents per share and revenue of $6.05bn, compared to expectations of earnings of 81 cents a share and revenue of $6bn. A year before, it reported pro forma earnings of $1.32 per share and revenue of $7.64bn.