US stocks closed out a winning week mixed on Friday as Wall Street took stock of the US economy from a lofty, record-setting perch ahead of the Federal Reserve’s highly anticipated decision on interest rates next week.

The tech-heavy Nasdaq Composite (^IXIC) climbed around 0.5% to notch its fifth-consecutive record as Tesla (TSLA) stock hit a seven-month high. The S&P 500 (^GSPC) fell just below the flat line, while the Dow Jones Industrial Average (^DJI) fell 0.6%.

Still, the Dow gained nearly 1% for the five trading sessions through Friday, its first win in three weeks, and the S&P 500 and Nasdaq had their best showings since early August.

Investors have taken in several weeks’ worth of economic data to gain clues on the Fed’s next move. Over the last week, jobs data has shown clear signals of labor market weakness, with just over 20,000 jobs added last month and weekly initial jobless claims surging to a near four-year high.

Meanwhile, inflation remains stubborn, with consumer prices rising last month amid more signs that President Trump’s tariffs are filtering their way into the economy.

The University of Michigan’s consumer sentiment survey released Friday showed consumer sentiment slipped more than expected in September, while long-run inflation expectations jumped to 3.9%, as Americans worried over the effects of tariffs.

But investors are betting inflation is tame enough for the Fed to cut next week — and then some.

Traders are pricing in a more than 90% chance of a quarter-point cut when the Fed holds its September meeting, according to CME Group. Beyond that, around 75% are betting the central bank will cut the equivalent of three times before the end of the year.

WASHINGTON — SpaceX Chief Executive Elon Musk says he is confident that Starship can start delivering 100 tons of payload to orbit next year while reusing both stages. In an interview during the All-In Summit Sept. 9, Musk said he expected an upgraded version of the vehicle, known as version 3, to start flying next year, with both the Super Heavy booster and Starship upper stage recovered and reused. “Unless we have some very major setbacks, SpaceX will demonstrate full reusability next year, catching both the booster and the ship, and being able to deliver over 100 tons to a useful orbit,” he said. That performance is essential for SpaceX, allowing it to both place larger next-generation Starlink satellites into orbit and support the lunar lander version of Starship the company is developing for NASA’s Artemis lunar exploration campaign. Version 3 of Starship will be a “gigantic upgrade” from the current version 2, he said, including the use of third-generation Raptor engines. “Pretty much everything changes on the rocket with version 3.” He cautioned that the upgraded rocket “might have some initial teething pains because it’s such a radical redesign.” The current version 2 vehicle suffered three consecutive mission-ending failures in test flights earlier this year before a largely successful Flight 10 mission Aug. 26. Musk confirmed there is one remaining launch of version 2 of Starship, planned for later this year, before switching to version 3. He didn’t offer an estimate of when the first version 3 launch would take place. At the American Astronautical Society’s Glenn Space Technology Symposium Sept. 8, Bill Gerstenmaier, vice president of build and flight reliability at SpaceX, said that final version 2 launch will, like the previous flights, be a suborbital mission. “We’re going to try and understand how the ship flies,” he said. He said he expected the first launch of Starship version 3 would also be suborbital. “If that’s successful, then we’ll probably go orbital after that with the next v3.”

Thermal protection system testing

One of the key challenges for Starship is developing a thermal protection system for the upper stage that can survive the rigors of reentry but does not require significant maintenance between flights. “For full reusability of the ship, there’s still a lot of work that remains on the heat shield. No one has ever made a fully reusable orbital heat shield,” Musk said, citing the extensive work required on the space shuttle’s heat shield between flights. “We really are looking at fundamental physics here,” he said, “trying to figure out how do we make something that can withstand the heat, is very light, doesn’t transmit the heat to the primary structure, and the tiles stay on and don’t crack.” On Flight 10, SpaceX tested some alternative tile technologies. That contributed to the odd discoloration of the vehicle after reentry, with part of the nose section white while much of the body was rust-colored. Gerstenmaier said that on Flight 10, SpaceX installed three metallic tiles to test their performance compared to the ceramic tiles used elsewhere on the ship. “They would be simpler to manufacture and more durable than the ceramic tiles,” he said. “Turns out they’re not.” The metallic tiles “didn’t do so good,” he said, oxidizing in the upper atmosphere during reentry. The white material in the nose, he said, was from an ablative material below the tiles that SpaceX also uses on its Dragon spacecraft. “What that’s showing us is that we’re having heat essentially get into that region between the tiles, go underneath the tiles, and this ablative structure is then ablating underneath. So again, we learned that we need to seal the tiles.” One technique that shows promise, he said, is a material he called “crunch wrap” that he compared to wrapping paper that goes around each tile when it is installed on the vehicle. Tiles on the ship that used the crunch wrap had less of the white ablative material than those in other parts of the vehicle that did not use it. “What we found was that this crunch wrap technique is allowing us to essentially seal between the tiles without putting a gap filler in between the tiles,” Gerstenmaier said. For the next flight, SpaceX plans to use crunch wrap for tiles throughout the vehicle “and see if we can get better sealing and better tile performance moving forward.” That next flight, he added, will do less experimentation with thermal tiles than on Flight 10. “We’ll probably cut back a little bit. We’re going to try to go more toward the configuration we want to fly next year.”

A new, high-resolution audio format has just launched on Spotify. Spotify announced Lossless Listening on Wednesday, Sept. 10. The new feature allows users to stream music in 24-bit/44.1 kHz FLAC, the least compressed and highest resolution audio format. Simply put, the audio is richer-sounding than regular, standard audio. Due to their high resolution, lossless audio files are large in size. For the time being, Spotify users should only listen to music at this quality with wired headphones or speakers on a non-Bluetooth connection, according to a Spotify news release. Bluetooth currently doesn’t provide enough bandwidth to transmit the file sizes of lossless audio. Additionally, because the files are so large, users should be prepared for tracks to take a moment to play, rather than instantaneously, the news release states. Here’s what to know about Spotify’s new feature.

What does Lossless Listening mean?

Spotify’s Lossless Listening, or more generally lossless audio, is described as compressed files that retain all of the original recording audio information, according to audio tech brand Sonos. In contrast, lossy audio is a compressed file format that deletes some of the original audio information. “Lossless refers to when there’s no change between what you deliver to a streaming service and what the listener hears,” Jack Mason, Spotify Studios producer and engineer, said in an Instagram Reel. “Without lossless, track files might get compressed in the encoding process, which might not lend to the best quality.” Going a bit further, every digital audio file contains snapshots of a recording’s original soundwave. Stitch these snapshots together and you have a reproduction of the original recording, according to Sonos. The more snapshots, the higher the sample rate. The higher the sample rate, the higher the quality. Additionally, these digital snapshots capture a wider range between the soft and loud sounds of a recording and have a higher bit depth, which also contributes to higher quality, Sonos states. A high sample rate, in tandem with a high bit depth, results in an audio file that is nearly indistinguishable from an original recording. This is lossless audio. In terms of Spotify’s lossless audio, which is available at 24-bit/44.1 kHz FLAC, this refers to a 24-bit depth and a 44.1 kHz sample rate.

How to turn on Spotify Lossless Listening

If Spotify’s Lossless Listening is available on your device, here’s how to turn it on:

• Tap your profile icon in the top left corner of the Spotify mobile app
• Tap Settings & Privacy
• Tap Media quality
• Select where you want to enable lossless audio (available under Wi-Fi, cellular and download streaming qualities)

Lossless Listening is available on Spotify mobile, tablet and desktop apps. When Lossless Listening is enabled, an icon will appear where the track title is displayed.

When is Lossless Listening available on Spotify?

Lossless Listening is available for select Spotify Premium subscribers in Australia, Austria, Czechia, Denmark, Germany, Japan, New Zealand, the Netherlands, Portugal, Sweden, the United States and the United Kingdom, according to the news release. More markets will see the feature in October.

Do you need Spotify Premium for Lossless Listening?

Yes, Lossless Listening is only available to Spotify Premium subscribers. Spotify Premium Individual plans are $11.99 per month.

Does Lossless Listening apply to podcasts and audiobooks, too?

No, Spotify’s Lossless Listening only applies to music streamed on the app, Spotify engineer John Cieslik-Bridgen told USA TODAY.

Do other music streaming platforms offer lossless audio?

Yes, Apple Music and Amazon Music both offer lossless audio. Apple Music began offering lossless audio in 2021, while Amazon Music added it in 2019. The feature is available for paid subscribers of the streaming services.

HONG KONG (Reuters) – Global investment firms Carlyle Group and EQT, alongside regional players HongShan Capital Group and Boyu Capital, are preparing final offers for a controlling stake in Starbucks’ China operations, said five people with knowledge of the matter.

Starbucks has asked them to submit binding bids by early October, said three of the sources, who declined to be identified as the information was private.

An agreement could be reached by the end of next month, one of them added.

Starbucks had invited about 10 potential buyers to submit non-binding bids by early September, with most offering to value the China business at as much as $5 billion, Reuters reported last month.

Starbucks has recently decided to sell control of its China operations to the final buyer, said two of the sources. The size of the stake has not yet been disclosed.

The final round of bidders also includes Chinese private equity firm Primavera Capital, which is likely to team up with a co-investor, said two of the sources.

The Seattle-based coffee group is seeking to retain control of its coffee bean roasting facility in the world’s second-largest economy, said two of the sources, with one adding that it was for quality control purposes.

Terms of the deal structure, including the size of the stake being sold, remain negotiable, said one of the sources.

Starbucks has said that it would maintain a meaningful stake in the China business.

In response to a Reuters request for comment, a spokesperson for Starbucks referred to its latest quarterly earnings where it had record-breaking sales growth in its international business and the third consecutive quarter of revenue growth in China.

The spokesperson declined to comment on the ongoing sale process.

Carlyle, Primavera and HSG, formerly known as Sequoia China, all declined to comment. EQT and Boyu did not respond to a request for comment.

Goldman Sachs, which is advising Starbucks on the sale, declined to comment.

The sale comes as Starbucks faces declining market share in China – home to more than a fifth of its cafes – due to intensifying competition from local rivals.

Its market share fell sharply to 14% last year from 34% in 2019, according to Euromonitor International data.

To counter these challenges, the chain has since implemented measures such as reducing prices for select non-coffee beverages in China and accelerating the introduction of new, localised products.

Comparable-store sales in China increased 2% in the quarter ended on June 29, versus zero growth in the previous quarter.

Oracle (NYSE: ORCL) lit up after-hours trading on Sept. 9 after reporting a massive jump in its booked work tied to cloud infrastructure. The database giant turned cloud platform is benefiting directly from the artificial intelligence (AI) buildout, but in a way that provides unusually clear visibility into future revenue. Nvidia (NASDAQ: NVDA) is still the face of AI infrastructure and continues to post eye-popping results. But the paths these two companies are on are different. Oracle’s growth is increasingly anchored by multiyear contracts that companies recognize as revenue over time, while Nvidia’s revenue depends on hardware shipments that ebb and flow with product transitions and customer ordering patterns. That’s why, from here, Oracle looks like the better buy.

Oracle’s cloud business is exploding

Start with the backlog. Oracle’s remaining performance obligations (RPO) — a leading indicator of revenue tied to signed contracts–surged to $455 billion (yes, you read that figure correctly) in the quarter ended Aug. 31, up 359% year over year. Management said it signed “four multibillion-dollar contracts with three different customers” in the quarter, and expects RPO to exceed half a trillion dollars in the coming months. Cloud revenue rose 28% and infrastructure-as-a-service (IaaS) revenue jumped 55%. This isn’t a one-off headline. RPO was $138 billion just last quarter, so the step-function increase reflects a wave of very large, multi-year deals landing at once. That’s the kind of demand AI leaders want to see — and it turns into revenue progressively over time, which typically smooths results compared to one-time hardware shipments. Oracle also raised the bar on its cloud infrastructure outlook. CEO Safra Catz previewed a plan to grow Oracle Cloud Infrastructure (OCI) revenue 77% this fiscal year to $18 billion and then scale it to $32 billion, $73 billion, $114 billion, and $144 billion over the subsequent four years–much of which is already embedded in RPO. The company highlighted blistering multicloud momentum as well: “MultiCloud database revenue from Amazon, Google and Microsoft grew at the incredible rate of 1,529% in Q1,” with 37 more data centers slated for delivery to hyperscaler partners (71 in total). Oracle even declared another $0.50 quarterly dividend, underscoring confidence and cash generation.

Nvidia’s AI engine is phenomenal, but more cyclical

Nvidia’s latest results remain exceptional: In the quarter ended July 27, revenue rose 56% year over year to $46.7 billion, with data-center revenue up 56% to $41.1 billion. Blackwell revenue grew 17% sequentially, and the company guided next quarter’s revenue to about $54 billion. None of that is weak. But look under the hood, and you see dynamics that can swing. Sequential compute revenue dipped 1% because of a $4.0 billion reduction in sales of H20 products, and there were no H20 sales to China in the quarter. Inventory climbed to $15.0 billion to support the next product ramp, and purchase commitments reached $45.8 billion as Nvidia lines up capacity for future cycles. This is what a world-class hardware franchise looks like — powerful, but still subject to product transitions, export rules, and hyperscaler ordering patterns. That’s the key contrast. Oracle’s growth is increasingly contract-based and recognized over time, with RPO providing a multi-year line of sight. Nvidia’s growth, while extraordinary, is inherently linked to hardware cycles and customers’ deployment timing. For portfolio construction, those differences matter. Pulling it together, Oracle offers investors a clearer runway tied to contractual obligations, accelerating multicloud distribution with the biggest platforms in tech, and a growing dividend — all while still being earlier in its AI cloud build-out than Nvidia is in AI silicon. Nvidia will likely continue to compound value, but the path can be choppier as architectures evolve and regional rules shift. For investors choosing one AI leader to buy today, Oracle’s visibility and mix of growth and durability make it the better buy.

Microsoft will pay to use Anthropic’s AI in Office 365 apps, The Information reports, citing two sources. The move means that Anthropic’s tech will help power new features in Word, Excel, Outlook, and PowerPoint alongside OpenAI’s, marking the end of Microsoft’s previous reliance solely on the ChatGPT maker for its productivity suite.

Microsoft’s move to diversify its AI partnerships comes amid a growing rift with OpenAI, which has pursued its own infrastructure projects as well as a potential LinkedIn competitor.

Microsoft’s Anthropic deal also comes as the company negotiates a new deal with OpenAI to secure access to its AI models after a pending for-profit restructuring. But The Information says the move isn’t a negotiating tactic. Leaders at Microsoft believe Anthropic’s latest models — Claude Sonnet 4, specifically — perform better than OpenAI’s in certain functions, like creating aesthetically pleasing PowerPoint presentations.

This isn’t the first time Microsoft has branched out, though. While OpenAI is the default model, Microsoft offers other models like xAI’s Grok and Anthropic’s Claude through GitHub Copilot. Microsoft is also trying to set itself up for self-reliance. The company recently introduced its first two in-house models: MAI-Voice-1 and MAI-1-preview.

Meanwhile, OpenAI is similarly seeking to step out from under Microsoft’s influence. Last week, OpenAI launched a jobs platform to take on Microsoft’s LinkedIn, and The Financial Times reported that OpenAI is set to begin mass production on its first AI chips in partnership with Broadcom in 2026. That means it will be able to potentially run training and inference on hardware it controls, rather than being dependent on Microsoft’s Azure setup.

“As we’ve said, OpenAI will continue to be our partner on frontier models and we remain committed to our long-term partnership,” Microsoft spokesperson Michael Collins told TechCrunch.

TechCrunch has reached out to Anthropic for comment.

Fidelity has launched the Fidelity Digital Interest Token (FDIT), an on-chain money market fund backed by mostly U.S. Treasuries, mainly from Ondo Finance’s Short-Term U.S. Treasuries Fund (OUSG) as the only anchor and largest investor. According to Ondo, OUSG makes up more than 99% of FDIT assets at this time. The launch is yet another milestone in the continued efforts by major asset managers to tokenize traditional financial products.

Fidelity now joins BlackRock, Franklin Templeton, and WisdomTree in offering tokenized money market or Treasury funds, demonstrating the rapid transition of real-world assets from experimentation to mainstream finance.

Ondo’s role in the tokenization momentum

Ondo Finance has been at the forefront of this change, having launched OUSG in January 2023, as one of the first tokens with on-chain access to U.S. Treasuries. Anyone can buy or invest in OUSG at any time. And this is now designated as the flagship tokenized product of Ondo, boasting a total value locked (TVL) of more than $730 million, which provides instant subscriptions and redemptions, daily interest accruals, and support across multiple blockchains, including Ethereum, Solana, Ripple, and Polygon. Included in the OUSG portfolio of product offerings, Ondo has already introduced other tokenized Treasury instruments, specifically BlackRock’s USD Institutional Digital Liquidity Fund (BUIDL) — which was itself an innovator in allowing instant redemptions to USDC through Circle — Franklin Templeton/BENJI and WisdomTree/WTGXX. Now, with Fidelity’s FDIT added, Ondo is growing its network of top-tier asset manager partnerships, which will facilitate liquidity and maturity of the ecosystem. On Sept. 3, Ondo Finance and the Ondo Foundation launched Ondo Global Market, a platform bringing over 100 tokenized U.S. stocks and ETFs to Ethereum. The launch provides non-U.S. investors with 24/7 on-chain access to U.S. equities and ETFs, furthering Ondo’s mission to bridge traditional finance and blockchain.

The bigger picture

The launch illustrates the gradual tokenization impacts capital markets and, by placing money-market funds onto blockchain rails, managers can offer faster settlement, 24/7 access, and potential interoperability with the extended digital-asset ecosystem again. “Fidelity’s entry marks another major milestone in the institutional embrace of tokenization,” Ondo said, adding that this partnership solidifies Ondo’s role as a bridge between traditional finance and decentralized solutions. As the tokenized finance industry develops, Ondo says it will continue to grow its infrastructure and intends to be at the center of building the “open and global” next generations of financial systems.

Rightly or wrongly, Amazon (NASDAQ: AMZN) has been viewed as something of a laggard in artificial intelligence (AI) among big tech companies since the launch of ChatGPT.

Microsoft, which started investing in OpenAI in 2019, helped engineer the transformative moment when ChatGPT debuted, and Alphabet had been waiting for it with its own large language model (LLM) ready to go. Even Meta Platforms unveiled its own LLM and chat platform, Meta AI, shortly after.

Amazon, on the other hand, seemed unprepared for the shift to AI. The company has since launched Amazon Bedrock, a managed service on Amazon Web Services (AWS) that provides access to AI models.

It has also launched its own LLM, Nova, which has received much less attention than LLMs from its peers. And it has built AI chips for training (called Trainium) and inference (Inferentia).

However, Amazon’s cloud growth has also lagged behind both Microsoft and Alphabet in recent years, which seems to be another sign that those two companies are outperforming it in AI.

Amazon’s most consequential move in AI may have been investing $8 billion in Anthropic and forming a strategic partnership with the AI start-up behind the Claude chatbot. And that move looks even smarter after Anthropic’s valuation ballooned to $183 billion.

Anthropic strikes gold

In a funding round that closed on Tuesday, Anthropic’s valuation tripled to $183 billion, just six months after a funding round that valued the company at $61.5 billion.

Anthropic raised $13 billion in a round led by Iconiq Capital, making it clear that it’s the closest challenger to OpenAI, which is now in talks for a valuation as high as $500 billion.

The surge in Anthropic’s valuation could also be a big win for Amazon because Anthropic is likely to spend at least some of that $13 billion on AI infrastructure supplied by AWS. It’s even set to be the primary user of a new Amazon data center in Indiana.

According to analyst research, Anthropic could spend as much as $5 billion on AWS next year, and that number could grow if it receives more funding, or if its exceptional growth rate continues, since the company said that its run-rate revenue improved from $1 billion at the beginning of the year to $5 billion by August 2025, implying an annual growth rate of roughly 700%. Anthropic has differentiated itself from OpenAI with a focus on “safe” AI, and Claude is differentiated from ChatGPT in a number of meaningful ways.

It has also committed to using Amazon’s Trainium and Inferentia chips, creating a market for those products that could help it build momentum. The tripling in Anthropic’s valuation also makes Amazon’s stake in the company that much more valuable.

Amazon’s AI future

For Amazon, the Anthropic funding round is more than a win for the company. It also offers evidence that it’s still early in the AI boom, since a tripling in valuation in six months shows that investors are eager to throw money at the new technology; and given the partnership between the two companies, a win for Anthropic is a win for Amazon.

While Amazon may have gotten a slow start in AI, it’s still the largest cloud infrastructure platform in the world, and it has other outlets for AI, including its e-commerce business, its logistics network, and an advertising business that will help it benefit from the technology.

It’s still early days for AI, and there’s plenty of room for Amazon to make gains, especially when its close partner is flush with a $13 billion infusion. Its bet on Anthropic is already paying off in more ways than one, and that partnership should help it grab other AI spending at AWS.

Snap is breaking itself apart and rebuilding from within. In a new annual company letter, CEO Evan Spiegel just announced the company is restructuring around small “startup squads” of 10 to 15 people to better compete against larger competitors.

The move comes as the 5,000-person company faces mounting pressure. Advertising revenue growth flatlined at 4% in the second quarter, and North American daily active users declined 2% to 98 million, a troubling sign in Snap’s most important market.

Spiegel does highlight one bright spot: Snapchat+ subscriptions now generate over $700 million in annual recurring revenue from more than 15 million paying subscribers, making direct revenue “one of Snap’s fastest-growing opportunities.”

Snap is also doubling down on Specs, building its own AR glasses that Spiegel envisions will replace smartphones entirely. He calls them a “a once-in-a-generation transformation towards human-centered computing.” (Meta and Google see the same future, partnering with Ray-Ban and Warby Parker, respectively.)

Spiegel acknowledges the current stock price “reflects doubt” but writes that there’s “startup-style return potential” at Snap’s roughly $12 billion valuation. Left unsaid: that number is down 90% from September 2021, when Snap’s market cap topped $116 billion during the height of social media mania.

Oil clawed back some of last week’s loss after OPEC+ agreed to raise production at a modest rate and against a backdrop of doubts about how many of its members can keep lifting output. The hike marks the reversal of cuts that were set to remain in place until the end of 2026 —following the rapid return of a previous tranche of idled barrels over recent months — as the alliance seeks to reclaim market share. It signals the group has the confidence to push its bold oil market strategy a little bit further. Brent climbed above $66 a barrel after losing almost 4% last week, when it became apparent that an output hike was on the way. Prices also pushed higher after Bloomberg News reported that the European Union is exploring new sanctions on Russian banks and energy companies as part of its latest measures to end the war in Ukraine, a move it’s hoping to coordinate with the US. The Organization of the Petroleum Exporting Countries and its partners will add 137,000 barrels a day in October, smaller than the increments scheduled for the previous two months. The actual volume is likely to be lower than announced, as some members of the group face pressure to forgo their share of increases to compensate for previous hikes, while others lack spare capacity. Early last month, the International Energy Agency predicted the surplus would reach a record next year, which Goldman Sachs Group Inc. forecasts will push Brent to the low-$50s. The global benchmark is down more than 10% this year, with President Donald Trump’s trade tariffs also weighing on the energy demand outlook. OPEC+ said on Sunday that restarting the remainder of the 1.66 million barrels of cuts would be contingent on “evolving market conditions,” and increases could be reversed. The group’s faster-than-expected return of idled barrels over recent months stunned sections of the oil market, but prices have held up relatively well following an initial slump in April. “The market had expected a bigger unwind,” FGE NexantECA Chairman Emeritus Fereidun Fesharaki said on Bloomberg Television, adding that oil could still fall below $60 a barrel into the end of the year and the start of 2026. “Until you actually see inventories building up, then there will be no impact.” Saudi Arabia’s Crown Prince Mohammed bin Salman is visiting Washington in November to meet with Trump, indicating there could also be political considerations behind the supply decision. The US president has repeatedly called for lower fuel prices as he seeks to tame inflation. China’s stockpiling of roughly 200,000 barrels a day in recent months has helped to support demand, Frederic Lasserre, global head of research and analysis at Gunvor Group, said at the Asia Pacific Petroleum Conference in Singapore on Monday. Still, the country might not be able to absorb all of the impending market surplus, he added.