Peter Thiel, once one of Donald Trump’s major financial backers in the tech industry, said Thursday that even though he’s not providing money to the Republican presumptive nominee’s campaign this time around, he’d vote for him over President Joe Biden.

“If you hold a gun to my head, I’ll vote for Trump,” Thiel said in an interview on stage at the Aspen Ideas Festival. “I’m not going to give any money to his super PAC.”

Thiel donated $1.25 million to Trump’s campaign in 2016 at a time when the vast majority of tech money was going to Democratic rival Hillary Clinton. Thiel, best known for an early bet on Facebook and for co-founding Palantir, also spoke at the Republican National Convention that year and, after the election, helped organize a meeting between Trump and top execs at Amazon, Apple, Google, Tesla and several other giant tech companies.

However, Thiel later soured on Trump and said last year that he wouldn’t be funding any politician in the 2024 presidential campaign. That’s after he spent $32 million on Republican candidates in the 2022 midterm elections with mixed results.

In Ohio, Trump’s pick, Republican J.D. Vance, defended a GOP-held seat against Democratic Rep. Tim Ryan. But in Arizona, Republican Blake Masters failed in his bid to unseat Democratic Sen. Mark Kelly.

WATCH: Tech for Trump

For years, Apple avoided using the acronym AI when talking about its products. Not anymore.

The boom in generative artificial intelligence, spawned in late 2022 by OpenAI, has been the biggest story in the tech industry of late, lifting chipmaker Nvidia to a $3 trillion market cap and causing a major shifting of priorities at Microsoft, Google and Amazon, which are all racing to add the technology into their core services.

Investors and customers now want to see what the iPhone maker has in store.

New AI features are coming at Apple’s Worldwide Developers Conference (WWDC), which takes place on Monday at Apple’s campus in Cupertino, California. Apple CEO Tim Cook has teased “big plans,” a change of approach for a company that doesn’t like to talk about products before they’re released.

WWDC isn’t typically a major investor attraction. On the first day, the company announces annual updates to its iOS, iPadOS, WatchOS and MacOS software in what’s usually a two-hour videotaped keynote launch event emceed by Cook. This year, the presentation will be screened at Apple’s headquarters. App developers then get a week of parties and virtual workshops where they learn about the new Apple software.

Apple fans get a preview of the software coming to iPhones. Developers can get to work updating their apps. New hardware products, if they appear at all, are not the showcase.

But this year, everyone will be listening for the most hyped acronym in tech.

With more than 1 billion iPhones in use, Wall Street wants to hear what AI features are going to make the iPhone more competitive against Android rivals and how the company can justify its investment in developing its own chips.

Investors have rewarded companies that show a clear AI strategy and vision. Nvidia, the primary maker of AI processors, has seen its stock price triple in the past year. Microsoft, which is aggressively incorporating OpenAI into its products, is up 28% over the past year. Apple is only up 9% over that same period, and has seen the other two companies surpass it in market cap.

“This is the most important event for Cook and Cupertino in over a decade,” Dan Ives, an analyst at Wedbush, told CNBC. “The AI strategy is the missing piece in the growth puzzle for Apple and this event needs to be a showstopper and not a shrug-the-shoulders event.”

Taking the stage will be executives including software chief Craig Federighi, who will likely address the real-life uses of Apple’s AI, whether it should be run locally or in massive cloud clusters and what should be built into the operating system versus distributed in an app.

Privacy is also a key issue, and attendees will likely want to know how Apple can deploy the data-hungry technology without compromising user privacy, a centerpiece of the company’s marketing for over half a decade.

“At WWDC, we expect Apple to unveil its long-term vision around its implementation of generative AI throughout its diverse ecosystem of personal devices,” wrote Gil Luria, an analyst at D.A. Davidson, in a note this week. “We believe that the impact of generative AI to Apple’s business is one of the most profound in all of technology, and unlike much of the innovation in AI that’s impacting the developer or enterprise, Apple has a clear opportunity to reach billions of consumer devices with generative AI functionality.”

Last month, OpenAI revealed a voice mode for its AI software called ChatGPT-4o.

In a short demo, OpenAI researchers held an iPhone and spoke directly to the bot inside the ChatGPT app, which was able to do impressions, speak fluidly and even sing. The conversation was snappy, the bot gave advice and the voice sounded like a human. Further demos at the live event showed the bot singing, teaching trigonometry, translating and telling jokes.

Apple users and pundits immediately understood that OpenAI had demoed a preview of what Apple’s Siri could be in the future. Apple’s voice assistant debuted in 2011 and since has gained a reputation for not being useful. It’s rigid, only able to answer a small proportion of well-defined queries, partially because it’s based on older machine learning techniques.

Apple could team up with OpenAI to upgrade Siri next week. It’s been discussing licensing chatbot technology from other companies, too, including Google and Cohere, according to a report from The New York Times.

Apple declined to comment on an OpenAI partnership.

One possibility is that Apple’s new Siri won’t compete directly with fully featured chatbots, but will improve its current features, and toss off queries that can only be answered by a chatbot to a partner. It’s close to how Apple’s Spotlight search and Siri work now. Apple’s system tries to answer the question, but if it can’t, it turns to Google. That agreement is part of a deal worth $18 billion per year to Apple.

Apple might also shy away from a full-throated embrace of an OpenAI partnership or chatbot. One reason is that a malfunctioning chatbot could generate embarrassing headlines, and could undermine the company’s emphasis on user privacy and personal control of user data.

“Data security will be a key advantage for the company and we expect them to spend time talking about their privacy efforts during the WWDC as well,” Citi analyst Atif Malik said in a recent note.

OpenAI’s technology is based on web scraping, and ChatGPT user interactions are used to improve the model itself, a technique that could violate some of Apple’s privacy principles.

Large language models like OpenAI’s still have problems with inaccuracies or “hallucinations,” like when Google’s search AI said last month that President Barack Obama was the first Muslim president. OpenAI CEO Sam Altman recently found himself in the middle of a thorny societal debate about deepfakes and deception when he denied accusations from actress Scarlett Johansson that OpenAI’s voice mode had ripped off her voice. It’s the kind of conflict that Apple executives prefer to avoid.

Outside of Apple, AI has become reliant on big server farms using powerful Nvidia processors paired with terabytes of memory to crunch numbers.

Apple, by contrast, wants its AI features to run on iPhones, and iPads, and Macs, which operate on battery power. Cook has highlighted Apple’s own chips as superior for running AI models.

“We believe in the transformative power and promise of AI, and we believe we have advantages that will differentiate us in this new era, including Apple’s unique combination of seamless hardware, software, and services integration, groundbreaking Apple Silicon with our industry-leading neural engines, and our unwavering focus on privacy,” Cook told investors in May on an earnings call.

Samik Chatterjee, an analyst at JPMorgan, wrote in a note this month that, “We expect Apple’s presentation at WWDC keynote to be focused on the features and the on-device capabilities as well as the GenAI models being run on-device to enable those features.”

In April, Apple published research about AI models it calls “efficient language models” that would be able to run on a phone. Microsoft is also publishing research on the same concept. One of Apple’s “OpenELM” models has 1.1 billion parameters, or weights — far smaller than OpenAI’s 2020 GPT-3 model which has 175 billion parameters, and smaller even than the 70 billion parameters in one version of Meta’s Llama, which is one of the most widely used language models.

In the paper, Apple’s researchers benchmarked the model on a MacBook Pro laptop running Apple’s M2 Max chip, showing that these efficient models don’t necessarily need to connect to the cloud. That can improve response speed, and provide a layer of privacy, because sensitive questions could be answered on the device itself, rather than being sent back to Apple servers.

Some of the features built into Apple’s software could include providing users a summary of their missed text messages, image generation for new emojis, code completing in the company’s development software Xcode, or drafting email responses, according to Bloomberg.

Apple could also decide to load up its M2 Ultra chips in its data centers to process AI queries that need more horsepower, Bloomberg reported.

WWDC won’t strictly be about AI.

The company has more than 2.2 billion devices in use, and customers want improved software and new apps.

One potential upgrade could be Apple’s adoption of RCS, an improvement to the older system of text messaging known as SMS. Apple’s messages app diverts texts between iPhones to its own iMessage system, which displays conversations as blue bubbles. When an iPhone texts an Android phone, the bubble is green. Many features such as typing notifications aren’t available.

Google led development of RCS, adding encryption and other features to text messaging. Late last year Apple confirmed that it would add support for RCS alongside iMessage. The debut of iOS 18 would be the logical time to show its work.

The conference will also be the first anniversary of Apple’s reveal of the Vision Pro, its virtual and augmented reality headset, which was released in the U.S. in February. Apple could announce its expansion to more countries, including China and the U.K.

Apple said in its WWDC announcement that the Vision Pro would be in the spotlight. Vision Pro is currently on the first version of its operating system, and core features, such as its Persona videoconferencing simulation, are still in beta.

For users with a Vision Pro, Apple will offer some of its virtual sessions at the event in a 3D environment.

Former President Donald Trump’s campaign debuted an official account on TikTok on Saturday night, the social media platform facing a potential ban in the U.S.

“It’s my honor,” Trump said in his first TikTok post under the handle “@realdonaldtrump,” followed by a montage of him waving to crowds at a Saturday Ultimate Fighting Championship show. That post received 1.5 million likes within 10 hours of going online.

Trump’s TikTok rollout came as his own social media company, Trump Media, took a financial tumble in the wake of the historic verdict that convicted the former president on 34 felony counts in his Manhattan hush money trial.

Trump Media, the parent company of Truth Social trading under the DJT ticker, was down over 5% at market close on Friday, the day after Trump’s conviction, with shares priced at $49. Immediately following Trump’s conviction on Thursday, the stock was down roughly 15% in extended trading hours.

Trump launched Truth Social in early 2022 as an alternative, “non-woke” social media platform after he was banned from sites like Twitter and Facebook following the Jan. 6 Capitol riots. Since then, Trump Media has gone public and the former president now holds a 65% stake in the company.

Truth Social did not immediately respond to CNBC’s request for comment about Trump’s move to TikTok.

Trump is several months later than his Democratic opponent, President Joe Biden, whose reelection campaign launched on TikTok in February. But the presumptive Republican presidential nominee already had over 2 million followers on Sunday, outpacing the Biden campaign’s near 340,000.

The disparity in those follower counts is typical for social media accounts directly attached to a specific candidate, which generally tend to outperform accounts associated with a campaign. For example, as of Sunday, Trump’s direct Truth Social account, “@realDonaldTrump” had over 7 million followers, while his campaign account, “@TeamTrump” had 427,000.

“We refuse to cede any ground to Biden and the Democrats,” Trump campaign spokesperson Karoline Leavitt said in a statement to NBC News on Sunday. “We will get President Trump’s winning message to every voter possible. He has already gained significant ground with young voters and this is another way to reach them.”

Both candidates joined TikTok despite previously vocalizing national security concerns about the app.

In April, Biden signed into law a foreign aid package with a clause to force TikTok’s Chinese parent company, ByteDance, to sell the app or else the platform would face a national ban in the U.S.

During his administration, Trump also said he would try to ban TikTok, though he has since flipped that stance. Still, he told CNBC’s “Squawk Box” in a March interview that he believes TikTok could threaten U.S. national security.

Nvidia reported fourth fiscal quarter earnings that beat Wall Street’s forecast for earnings and sales, and said revenue during the current quarter would be better than expected, even against elevated expectations for massive growth.

Nvidia shares rose about 10% in extended trading.

Here’s what the company reported compared with what Wall Street was expecting for the quarter ending in January, based on a survey of analysts by LSEG, formerly known as Refinitiv:

• Earnings per share: $5.16 adjusted vs. $4.64 expected
• Revenue: $22.10 billion vs. $20.62 billion expected

Nvidia said it expected $24.0 billion in sales in the current quarter. Analysts polled by LSEG were looking for $5.00 per share on $22.17 billion in sales. 

Nvidia has been the primary beneficiary of the recent technology industry obsession with large artificial intelligence models, which are developed on the company’s pricey graphics processors for servers.

Nvidia CEO Jensen Huang addressed investor fears that the company may not be able to keep up this growth or level of sales for the whole year on a call with analysts.

“Fundamentally, the conditions are excellent for continued growth” in 2025 and beyond, Huang told analysts. He says demand for the company’s GPUs will remain high due to generative AI and an industry-wide shift away from central processors to the accelerators that Nvidia makes.

Nvidia reported $12.29 billion in net income during the quarter, or $4.93 per share, up 769% versus last year’s $1.41 billion or 57 cents per share. 

Nvidia’s total revenue rose 265% from a year ago, based on strong sales for AI chips for servers, particularly the company’s “Hopper” chips such as the H100, it said.

“Strong demand was driven by enterprise software and consumer internet applications, and multiple industry verticals including automotive, financial services and health care,” the company said in commentary provided to investors.

Those sales are reported in the company’s Data Center business, which now comprises the majority of Nvidia’s revenue. Data center sales were up 409% to $18.40 billion. Over half the company’s data center sales went to large cloud providers.

Nvidia said its data center revenue was hurt by recent U.S. restrictions on exporting advanced AI semiconductors to China.

“We understood what the restrictions are, reconfigured our products in a way that is not software hackable in any way, and that took some time so we reset our product offering to China,” Huang said. “Now we’re sampling to customers in China.”

Nvidia Chief Financial Officer Colette Kress said that while the company had improved supply of its AI GPUs, it still expected them to be in short supply, especially the next-generation chip, called B100, expected to ship later this year.

“We are delighted that supply of Hopper architecture products is improving,” Kress said on a call with analysts. “Demand for Hopper remains very strong. We can expect our next-generation products to be supply constrained as demand far exceeds supply.”

“Whenever we have new products, as you know, it ramps from zero to a very large number and you can’t do that overnight,” Huang said.

The company’s gaming business, which includes graphics cards for laptops and PCs, was merely up 56% year over year to $2.87 billion. Graphics cards for gaming used to be Nvidia’s primary business before its AI chips started taking off, and some of Nvidia’s graphics cards can be used for AI.

Nvidia’s smaller businesses did not show the same meteoric growth. Its automotive business declined 4% to $281 million in sales, and its OEM and other business, which includes crypto chips, rose 7% to $90 million. Nvidia’s business making graphics hardware for professional applications rose 105% to $463 million.

Former Meta operating chief Sheryl Sandberg is leaving the company’s board of directors.

“With a heart filled with gratitude and a mind filled with memories, I let the Meta board know that I will not stand for reelection this May,” Sandberg wrote in a Facebook post on Wednesday.

Sandberg, 54, joined Facebook in 2008 as Mark Zuckerberg’s top deputy after spending about seven years at Google. In 2012, she became a board member at the company. During her tenure, Facebook rose from a highflying startup to become one of the most valuable companies in the world, topping a $1 trillion market cap at its peak in 2021.

Sandberg announced her departure from Meta in mid-2022, following multiple controversies that dogged the company and sullied its reputation among users, lawmakers and investors. Most notably, Facebook was central to the spread of disinformation ahead of the 2016 election and during the early days of the Covid pandemic in 2020. The company has also been in the subject of antitrust investigations and was scrutinized in Sandberg’s waning days for its insufficient efforts to combat hate on its platform.

When Sandberg stepped down as Meta COO in June 2022, she was replaced by Javier Olivan, who had been serving as Meta’s chief growth officer.

Since leaving Meta, Sandberg has dedicated much of her time on her LeanIn.org nonprofit, which focuses on empowering women tin the workplace, and related projects.

“I wanted my new chapter to be able to really make a difference,” Sandberg told CNBC Make It in August. “We’ve been in development on this since I was at Meta, but being able to have the time to put into [this launch] and to really be … a bigger part of this has meant a lot to me.”

Shortly after Sandberg’s post, Zuckerberg responded with a short reply.

“Thank you Sheryl for the extraordinary contributions you have made to our company and community over the years,” Zuckerberg wrote. “Your dedication and guidance have been instrumental in driving our success and I am grateful for your unwavering commitment to me and Meta over the years. I look forward to this next chapter together!”

Meta technology chief Adam Bosworth wrote, “Amazing run Sheryl, thank you so much for everything you did for all of us and also for me personally.”

Meta’s board consists of Zuckerberg, who serves as chairman, as well as former PayPal Executive Vice President Peggy Alford, venture capitalist Marc Andreessen, Dropbox CEO Drew Houston, former McKinsey & Company senior partner Nancy Killefer, former U.S. deputy secretary of the treasury Robert M. Kimmitt, DoorDash CEO Tony Xu and Tracey T. Travis, a former CFO at Estée Lauder.

Here’s the full text of Sandberg’s post:

With a heart filled with gratitude and a mind filled with memories, I let the Meta board know that I will not stand for reelection this May. After I left my role as COO, I remained on the board to help ensure a successful transition. Under Mark’s leadership, Javi Olivan, Justin Osofsky, Nicola Mendelsohn, and their teams have proven beyond a doubt that the Meta business is strong and well-positioned for the future, so this feels like the right time to step away. Going forward, I will serve as an advisor to the company, and I will always be there to help the Meta teams.

Serving as Facebook’s – and then Meta’s – COO for 14 ½ years and a board member for 12 years has been the opportunity of a lifetime. I will always be grateful to Mark for believing in me and for his partnership and friendship; he is that truly once-in-a-generation visionary leader and he is equally amazing as a friend who stays by your side through the good times and the bad. I will always be grateful to my colleagues and teammates at Meta for all the years of working side by side and all they taught me. And I am particularly grateful to my fellow Meta board members for their lasting friendships, the guidance they provided me for so many years, and their stewardship of products that mean so much to people all over the world.

WATCH: Three buys and a bail

Nvidia on Monday unveiled the H200, a graphics processing unit designed for training and deploying the kinds of artificial intelligence models that are powering the generative AI boom.

The new GPU is an upgrade from the H100, the chip OpenAI used to train its most advanced large language model, GPT-4. Big companies, startups and government agencies are all vying for a limited supply of the chips.

H100 chips cost between $25,000 and $40,000, according to an estimate from Raymond James, and thousands of them working together are needed to create the biggest models in a process called “training.”

Excitement over Nvidia’s AI GPUs has supercharged the company’s stock, which is up more than 230% so far in 2023. Nvidia expects around $16 billion of revenue for its fiscal third quarter, up 170% from a year ago.

The key improvement with the H200 is that it includes 141GB of next-generation “HBM3” memory that will help the chip perform “inference,” or using a large model after it’s trained to generate text, images or predictions.

Nvidia said the H200 will generate output nearly twice as fast as the H100. That’s based on a test using Meta’s Llama 2 LLM.

The H200, which is expected to ship in the second quarter of 2024, will compete with AMD’s MI300X GPU. AMD’s chip, similar to the H200, has additional memory over its predecessors, which helps fit big models on the hardware to run inference.

Nvidia said the H200 will be compatible with the H100, meaning that AI companies who are already training with the prior model won’t need to change their server systems or software to use the new version.

Nvidia says it will be available in four-GPU or eight-GPU server configurations on the company’s HGX complete systems, as well as in a chip called GH200, which pairs the H200 GPU with an Arm-based processor.

However, the H200 may not hold the crown of the fastest Nvidia AI chip for long.

While companies like Nvidia offer many different configurations of their chips, new semiconductors often take a big step forward about every two years, when manufacturers move to a different architecture that unlocks more significant performance gains than adding memory or other smaller optimizations. Both the H100 and H200 are based on Nvidia’s Hopper architecture.

In October, Nvidia told investors that it would move from a two-year architecture cadence to a one-year release pattern due to high demand for its GPUs. The company displayed a slide suggesting it will announce and release its B100 chip, based on the forthcoming Blackwell architecture, in 2024.

WATCH: We’re a big believer in the AI trend going into next year

Don’t miss these stories from CNBC PRO:

Semiconductor technology company Arm reported its first post-initial public offering earnings on Wednesday that beat Wall Street expectations for sales and showed that the company’s lucrative licensing business doubled in size over the past year.

Arm shares fell over 7% in extended trading after the company’s revenue guidance was short of expectations.

Here’s how the semiconductor licensing company did versus consensus expectations by LSEG, formerly known as Refinitiv, for Arm’s second fiscal quarter ending Sept. 30:

• Earnings per share: 36 cents, adjusted
• Revenue: $806 million vs. $744.3 million expected

Arm said it was expecting earnings per share between 21 cents and 28 cents on sales of between $720 million and $800 million in the current quarter. That’s a little lighter than what Wall Street was looking for, which was 27 cents per share on revenue between $730 million and $805 million.

Arm reported a net loss of $110 million, or 11 cents per share. The company said the loss was due to more than $500 million in one-time share-based compensation triggered by the recent IPO, and that share-based compensation would land between $150 million and $250 million in future quarters.

Total revenue was up 28% on an annual basis during the quarter.

Arm’s intellectual property is in nearly every smartphone, many PCs and other miscellaneous chips. Arm says more than 7.1 billion Arm-based chips were shipped during the quarter.

It makes money through royalties, or when chipmakers pay Arm for access to build Arm-compatible chips, typically a small percentage of the final chip price. It also sells licenses to more complete chip designs, saving chipmakers time and effort, which are recorded as licensing revenue.

Arm royalty revenue was $418 million, a 5% decline from the same period last year. But Arm licensing sales were $388 million, up 106% from the same period last year. It’s a sign that Arm may be able to sell increasing amounts of technology to its current customers, which is a key metric watched by analysts.

Arm attributed licensing sales to multiple long-term agreements with technology companies, suggesting the segment’s growth could continue in future quarters, but warned that the broader economy could affect future licensing growth.

Arm went public in an IPO in September. Before that, it was owned by SoftBank, which reached a deal to sell the firm to Nvidia before the transaction was scuttled by regulators in 2022. It was founded in 1990 to develop technology for low-power chips.

Arm said that firms including Google, Meta and Nvidia were developing artificial intelligence-capable chips with its technology.

Don’t miss these stories from CNBC PRO:

FTX founder Sam Bankman-Fried told jurors in his criminal trial on Friday that he didn’t commit fraud, and that he thought the crypto exchange’s outside expenditures, like paying for the naming rights at a sports arena, came out of company profits.

Bankman-Fried addressed the New York courtroom a day after U.S. District Judge Lewis Kaplan sent jurors home early to consider whether some aspects of the defendant’s planned testimony, related to legal advice he got while running FTX, would be admissible in court.

On Friday morning, defense attorney Mark Cohen asked Bankman-Fried if he defrauded anyone.

“No, I did not,” Bankman-Fried responded.

Cohen followed by asking if he took customer funds, to which Bankman-Fried said “no.”

Bankman-Fried, 31, faces seven criminal counts, including wire fraud, securities fraud and money laundering, that could land him in prison for life if he’s convicted. Bankman-Fried, the son of two Stanford legal scholars, has pleaded not guilty in the case.

Prior to the defendant’s appearance on the stand, the four-week trial was highlighted by the testimony of multiple members of FTX’s top leadership team as well as the people who ran sister hedge fund Alameda Research. They all singled out Bankman-Fried as the mastermind of a scheme to use FTX customer money to fund everything from venture investments and a high-priced condo in the Bahamas to covering Alameda’s crypto losses.

Prosecutors walked former leaders of Bankman-Fried’s businesses through specific actions taken by their boss that resulted in clients losing billions of dollars last year. Several of the witnesses, including Bankman-Fried’s ex-girlfriend Caroline Ellison, who ran Alameda, have pleaded guilty to multiple charges and are cooperating with the government.

The judge’s decision to send the jury home on Thursday allowed Bankman-Fried and his defense team to audition their best legal material for Judge Kaplan.

On Friday, Bankman-Fried acknowledged that one of his biggest mistakes was not having a risk management team or chief regulatory officer. That led to “significant oversights,” he said.

Cohen walked Bankman-Fried through his background and how he got into crypto. The defendant said he studied physics at the Massachusetts Institute of Technology and graduated in 2014. He then worked as a trader on the international desk at Jane Street for over three years, managing tens of billions of dollars a day in trading. That’s where he learned the fundamentals of things like arbitrage trading.

In the fall of 2017, Bankman-Fried founded Alameda Research.

“This was when crypto was starting to become publicly visible for the first time,” Bankman-Fried testified.

He said people were excited about it, watching bitcoin, which had jumped from $1,000 to $10,000 in a two-month period. Banks and brokers weren’t involved yet and it seemed like there would probably be big demand for an arbitrage provider, he said.

“I had absolutely no idea” how cryptocurrencies worked, Bankman-Fried said. “I just knew they were things you could trade.”

The first Alameda office was in an Airbnb in Berkeley, California, he said. It was listed as a two bedroom but they used the couch in the living room as a third bed and also used the attic.

He started FTX in 2019. Trading volume grew substantially on FTX from a few million dollars a day to tens of millions of dollars that year to hundreds of millions of dollars in 2020. By 2022, that number was up to $10 billion to $15 billion of dollars per day in trading volume, he said.

Bankman-Fried said Alameda was permitted to borrow from FTX, but his understanding was that the money was coming from margin trades, collateral from other margin trades or assets earning interest on the platform.

At FTX, there were no general restrictions on what could be done with funds that were borrowed as long as the company believed assets were greater than liabilities, Bankman-Fried testified.

In 2020, a routine liquidation gone wrong led to some of the special borrowing permissions at Alameda, he said. The risk engine was sagging under the weight of growth. A liquidation that should have been in the thousands of dollars was in the trillions of dollars. Alameda was suddenly underwater because of closing the position.

The incident exposed a larger concern, that the potential of an erroneous liquidation of Alameda could be disastrous for users.

Bankman-Fried said he talked to FTX’s engineering director Nishad Singh and co-founder Gary Wang, both of whom testified earlier on behalf of the prosecution. They suggested creating an alert, which would prompt the user to deposit more collateral, or a delay, Bankman-Fried said. They later implemented a feature like that, he said, adding that he learned it was the “allow negative” feature.

Bankman-Fried testified that he wasn’t aware of the amount Alameda was borrowing or its theoretical max. As long as the net asset value was positive on the exchange and the scale of borrowing was reasonable, increasing the line of credit so Alameda could keep filling orders was fine, he said. Bankman-Fried added that he now believes what Singh and Wang did was increase the line of credit.

Convincing the jury will be a tall order for Bankman-Fried after a mountain of damning evidence was presented by the government.

Prosecutors entered corroborating materials, including encrypted Signal messages and other internal documents that appear to show Bankman-Fried orchestrating the spending of FTX customer money.

The defense’s case, which consists of Bankman-Fried’s testimony along with that of two witnesses who took the stand Thursday morning, hinges largely on whether the jury believes the defendant didn’t intend to commit fraud.

On Thursday, under questioning led by Cohen, Bankman-Fried appeared to place much of the criminal blame on FTX’s chief regulatory officer, Dan Friedberg, as well as outside counsel Fenwick & West, which advised the crypto exchange. Bankman-Fried spoke about Friedberg’s active involvement in everything from the companywide auto-deletion policy on messaging apps like Signal, to the creation of Alameda’s North Dimension bank account, where billions of dollars worth of FTX customer money was funneled.

The former FTX chief also said that the hundreds of millions of dollars in personal loans to himself and other founders of the platform were structured through promissory notes drafted by his in-house legal team and discussed in concert with his general counsel and Friedberg. Having the blessing of his legal counsel was something that Bankman-Fried said he “took comfort in.”

In afternoon testimony, Bankman-Fried was asked about FTX’s marketing and promotions.

He said there were 15 people on the marketing team, and noted that he got more involved with it as time progressed. In particular, he discussed the naming rights in 2021 for the basketball arena in Miami, which was to be a 19-year deal for $135 million.

Bankman-Fried said the sponsorship of FTX Arena would deliver returns for the company and create wide brand awareness because even he, as an “average level sports fan,” could name dozens of stadiums. He said the investment would be about $10 million a year, or 1% of revenue. The company had been deciding among a few different stadiums, including the homes to the NFL’s New Orleans Saints and Kansas City Chiefs, Bankman-Fried said.

A crucial part of his testimony came when Bankman-Fried said he thought the stadium deal funding was coming from revenue from the exchange and returns from venture investments, as opposed to customer money.

Similarly, Bankman-Fried testified that he believed the lavish Bahamas properties were being paid for with FTX operating cash that came from revenue and venture investments. He said having available property to rent was a necessary incentive if the company wanted to poach developers from Facebook and Google.

As for the venture investments, Bankman-Fried said he thought that money was coming from Alameda’s operating profits and third-party lending desks. Alameda’s venture arm was renamed Clifton Bay Investments, which Bankman-Fried said was a first step in building a dedicated venture brand.

When asked about loans he took from the business, Bankman-Fried said they were to pay for venture investments and political donations. He said that, as the primary owner of Alameda, he thought he had a few billion dollars in arbitrage profit from the past few years and there was no reason he couldn’t borrow from it. He said the loans, except for the most recent one prior to the firm’s bankruptcy filing, were all documented through promissory notes.

Bankman-Fried said he never directed Singh or former FTX executive Ryan Salame to make political donations. Salame pleaded guilty in September to federal campaign finance and money-transmitting crimes, admitting that from fall 2021 to November 2022, he steered tens of millions of dollars of political contributions to both Democrats and Republicans in his own name when the money actually came from Alameda.

Bankman-Fried, who allegedly used FTX customer funds to help finance over $100 million in political giving during the 2022 midterms, testified that he talked to politicians about pandemic prevention and crypto regulation. He said he had a vested interested in crypto policy even though FTX’s U.S. operation was relatively small, because the company was seeking to offer crypto futures products in the U.S.

Bankman-Fried then discussed his public persona. He said he hadn’t intended to be the public face of the company because he’s “naturally introverted.” But a few interviews went well, and it snowballed from there. He said he was the only person at the company that the press sought.

He wore T-shirts and shorts because they were comfortable and said he let his hair grow out because he was busy and lazy.

Bankman-Fried was photographed at the 2022 Super Bowl in Los Angeles with Katy Perry. He told the jury, which was previously presented with the photo by the prosecution, that he thought it was natural to go to the game because he was in town for meetings and the company had a commercial running.

“I thought maybe it would be interesting,” he said.

The afternoon testimony largely focused on Bankman-Fried’s repeated and unsuccessful request to Ellison that she hedge Alameda’s risk. Bankman-Fried said in late 2021, he had talked to Ellison about putting on trades to protect against the risk of market moves since Alameda had been leveraged long, meaning they would lose money if market went down.

Ellison said she would look into it, which Bankman-Fried said he “interpreted” as her being “far less enthusiastic about it.” Over the course of 2022, Bankman-Fried said every two months he would check in to see if Alameda had hedged, and each time he was told not yet, but Ellison would say she was planning to do so in the near future.

Specifically, Bankman-Fried said he had talked with Ellison and Ramnik Arora, who had been the head of product at FTX, about putting a $2 billion hedge on the company’s investment in Genesis Digital Assets, a bitcoin miner. He told the jury that the hedge was never made.

There was also more detail on how Bankman-Fried was told about FTX’s $8 billion liability. According to the defendant, in October 2022, developers built a Google database that included financial data. That’s where Bankman-Fried noticed the negative $8 billion balance, which he said he was “very surprised” to see.

Bankman-Fried testified that he’d started taking a more comprehensive look at total assets. He said he was happy to pledge everything he had as backup for the liabilities.

Cohen then brought the jury through the summer months of 2022, a time when Alameda’s lenders, specifically Genesis, BlockFi, Celsius and Voyager, all had direct conversations with Bankman-Fried about the need for emergency capital. In the end, only BlockFi and Voyager received funds from Alameda and Bankman-Fried.

In late 2021 and early 2022, Bankman-Fried said he wanted FTX revenue to be above $1 billion because it was a round number. He asked company executives if there were ways to reach that mark. Singh said he’d dealt with it by staking the company’s investment in crypto token Serum, a way of putting the coins to work. That would add another $50 million in revenue. Bankman-Fried testified that he was “a little surprised” they found that additional money, but it got him to $1 billion.

— CNBC’s Dawn Giel contributed to this report

WATCH: Sam Bankman-Fried testifying in his criminal case