Mark Zuckerberg started Meta‘s earnings call by talking about artificial intelligence. Then he moved onto the metaverse, touting his company’s headsets, glasses and operating system. He spent almost the entirety of his opening remarks focused on the many ways Meta loses money.

Investors weren’t into it. Meta shares tumbled as much as 19% in extended trading on Wednesday, wiping out more than $200 billion in market cap. The drop came despite Meta reporting better-than-expected profit and revenue for the first quarter.

Zuckerberg appeared ready for the sell-off.

“I think it’s worth calling that out, that we’ve historically seen a lot of volatility in our stock during this phase of our product playbook where we’re investing in scaling a new product but aren’t yet monetizing it,” Zuckerberg said. He cited past efforts like short-video service Reels, Stories and the transition to mobile.

Meta generates 98% of its revenue from digital advertising. But to the extent Zuckerberg talked about ads, he was looking to the future and the ways the company could potentially turn its current investments into ad dollars. In discussing Meta’s effort to build a “leading AI,” he said, “There are several ways to build a massive business here including scaling business messaging, introducing ads or paid content into AI interactions.”

He spent time talking about Meta Llama 3, the company’s newest large language model, and the recent rollout of Meta AI, the company’s answer to OpenAI’s ChatGPT. 

Zuckerberg then moved onto potential opportunities for expansion within the mixed reality headset market, like a headset for work or fitness. Meta opened up access to the operating system that powers its Quest headsets on Monday, which Zuckerberg said will help the mixed reality ecosystem grow faster.

He also talked up Meta’s AR glasses, which he called “the ideal device for an AI assistant because you can let them see what you see and hear what you hear.”

In the meantime, Meta’s Reality Labs unit, which houses the company’s hardware and software for development of the nascent metaverse, continues to bleed cash. Reality Labs reported sales of $440 million for the first quarter and $3.85 billion in losses. The division’s cumulative losses since the end of 2020 have topped $45 billion.

Zuckerberg has bought himself some time.

Meta’s stock price almost tripled last year and, as of Wednesday’s close, was up 40% in 2024. It reached a record $527.34 in early April.

After a brutal 2022, during which the company lost about two-thirds of its value, Zuckerberg appears to have regained the confidence of Wall Street.

The driver for the rally has been a cost-cutting plan that the Meta CEO put in place early last year, when he told investors that 2023 would be the “year of efficiency.” The company slashed headcount and eliminated unnecessary projects in an effort to become a “stronger and more nimble organization.”

Zuckerberg said Wednesday that Meta will continue to operate efficiently, but that shifting existing resources to investments in AI will “grow our investment envelope meaningfully.”

Capital expenditures for 2024 are anticipated to be in the $35 billion to $40 billion range, an increase from a prior forecast of $30 billion to $37 billion “as we continue to accelerate our infrastructure investments to support our artificial intelligence (AI) roadmap,” Meta said.

Zuckerberg said he expects to see a “multiyear investment cycle” before Meta’s AI products will scale into profitable services, but noted that the company has a “strong track record” in that department.

Meta finance chief Susan Li echoed Zuckerberg’s remarks, saying the company needs to develop advanced models and scale products before they will drive meaningful revenue.

“While there is tremendous long-term potential, we’re just much earlier on the return curve,” Li said.

Even before the call began, investors were trimming their holdings. That’s because Meta issued a light revenue forecast for the second quarter, overshadowing the first-quarter beat.

As the stock plunge intensified, Zuckerberg told investors that if they’re willing to come along for the ride, they may well be rewarded.

“Historically, investing to build these new scaled experiences in our apps has been a very good long-term investment for us and for investors who stuck with us and the initial signs are quite positive here too,” Zuckerberg said. “But building a leading AI will also be a larger undertaking than the other experiences we’ve added to our apps and this is likely going to take several years.”

Nvidia-backed artificial intelligence firm Synthesia on Thursday unveiled a new wave of AI-generated digital avatars that can convey human emotions using a user’s text inputs.

The company said its “Expressive Avatars” can blur the lines between the virtual world and real characters. It aims to eliminate cameras, microphones, actors, lengthy edits and other costs from the professional video production process. Synthesia has a studio in London, where actors read scripts in front of a green screen to train the system.

In one demonstration, the company showed three lines of text being inserted into its platform — “I am happy. I am sad. I am frustrated” — after which the AI-generated actor in the video responded by reading the text in the tone of each corresponding emotion.

The company’s technology is used by more than 55,000 businesses, including half of the Fortune 100, to make digital avatars for corporate presentations and training videos, according to Synthesia.

Founded in 2017, Synthesia raised $90 million from investors last year at a valuation of around $1 billion, making it one of Britain’s more recent AI “unicorn” firms. Accel, Kleiner Perkins, GV, FirstMark Capital and MMC are also shareholders.

The company addressed concerns over how its videos might be used to create fake news content, saying publishers must sign up as enterprise customers to make synthetic avatars. Content made with its technology is vetted by moderators.

Synthesia doesn’t publicly disclose pricing for its enterprise customers.

The company also requires all of its new clients to undergo a thorough “Know Your Customer” process similar to that used by the banking industry, which helps prevent bad actors from creating fake company profiles to spread misinformation.

Synthesia said it’s already preparing for the upcoming global elections and has implemented a range of controls to ensure its platform isn’t abused by hostile actors seeking to manipulate the outcome of various votes.

The company is also a part of the Coalition for Content Provenance and Authenticity — an organization of AI companies that aims to implement content credentials and digital “watermarking” of AI-generated content to ensure viewers know that what they are looking at is made by artificial intelligence and not by a human.

South Korean memory chipmaker SK Hynix on Thursday reported a net profit of 1.92 trillion South Korean won ($1.39 billion) in the first quarter, reversing a loss of 2.58 trillion won logged in the same period a year ago.

This was the first positive income recorded since the third quarter of 2022, LSEG data showed. SK Hynix posted net losses for five consecutive quarters from a slump in the memory chip market.

Revenue in the first quarter stood at 12.43 trillion won, a 144% increase from a year ago. This was the highest revenue logged since second quarter 2022, according to LSEG data.

SK Hynix attributed the strong performance to an “increase in the sales of AI server products backed by its leadership in AI memory technology including high-bandwidth memory” as well as efforts to drive profitability.

SK Hynix is the world’s second-largest memory chipmaker after Samsung Electronics and supplies high-bandwidth memory chips catering to AI chipsets for companies like Nvidia.

The explosive demand for AI chipsets boosted the high-end memory chip market, hugely benefiting players like SK Hynix and Samsung Electronics.

Large language models such as ChatGPT – which caused AI adoption to skyrocket – require a lot of high-performance memory chips as such chips allow these models to remember details from past conversations and user preferences in order to generate humanlike responses.

To meet AI memory demand, the firm said it plans to increase supply of HBM3E – the latest generation of high-bandwidth memory for AI. SK Hynix said it will also introduce 32GB Double Data Rate 5 products this year to strengthen its leadership in the high-capacity server DRAM market.

“We will continue to work towards improving our financial results by providing the industry’s best performing products at a right time and maintaining the profitability-first commitment,” said Chief Financial Officer Kim Woohyun.

The firm projects the overall memory market to grow steadily in the coming months amid rising demand for AI memory, while the conventional DRAM market starts recovering from the second half of 2024.

Pandemic-induced demand for consumer electronics led companies to stockpile memory chips. But macroeconomic uncertainties such as inflation caused consumers to cut back on purchases of such consumer goods, driving down demand and prices for memory chips.

To address the excess inventories, companies like SK Hynix cut production of its memory chips.

SK Hynix shares slid more than 4% on Thursday morning, though in the last one year, they have jumped more than 100%.

The firm on Wednesday said it plans to build a new fab in South Korea, with an estimated completion date set for November 2025, to increase production of the next-generation DRAM including HBM to capture the proliferating demand for AI chips.

Total investment would amount to more than 20 trillion won in the long term, SK Hynix said.

SK Hynix is also partnering with TSMC, the world’s largest contract chip manufacturer, to build high-bandwidth memory 4 chips and next-generation packaging technology. Mass production of the HBM4 chips is expected to start from 2026.

SK Hynix will leverage on TSMC’s leading-edge processes, according to an April 19 statement.