Amazon opens talks for a $7bn AI startup, the White House lets Britain test US models again, and 3 more AI moves
Five verified developments from October 11 to 12, 2026: Amazon opens talks to buy the AI startup Decart for about $7bn, the White House restores Britain's pre-release access to American models for safety testing, Broadcom's chief executive says many AI investors are betting blind, the IMF credits AI investment with holding global growth near 3%, and Tencent borrows about $5bn mainly to buy chips.
Between October 11 at 12:00 UTC and October 12 at 12:00 UTC, Amazon opened talks to buy an Israeli AI startup for about $7bn, the White House gave Britain back its pre-release look at American models, the head of Broadcom told a Singapore audience that many AI investors have no idea what they own, the IMF built its growth forecast around the AI boom, and Tencent told bond investors what its next $5bn is for.
In brief
- Markets: Broadcom's Hock Tan said many AI investors "have no clue what they are investing in" and end up "almost like they are in a casino", while Intel's Lip-Bu Tan told the same Singapore panel that hardware stays in short supply for another two to three years. 5
- Economy: IMF managing director Kristalina Georgieva said AI investment is holding global growth at around 3% and called the labour-market shift ahead a "slow-moving tsunami". IMF research puts 40% of jobs at risk of being augmented, downgraded or eliminated. 6
- Funding: Tencent told fixed-income investors that its planned roughly $5bn bond is mainly for AI capital spending, with the largest share of that going to chips, and said the sale could price as early as Tuesday. 7
Five developments
1. Amazon opened talks to buy the startup at the centre of the chip question
The Wall Street Journal reported on Sunday that Amazon is negotiating to acquire Decart for around $7bn, citing people familiar with the matter. Bloomberg relayed the report at 20:46 UTC and Reuters carried it at 21:16 UTC. Neither company has commented, and the talks could still collapse. 18
Decart was founded in 2023 by three Israeli engineers: brothers Dean and Orian Leitersdorf and Moshe Shalev. Its core software moves AI workloads between chips built by different manufacturers and extracts more computing from hardware a customer already owns. The company also makes generative video: Lucy edits live footage in real time, and Oasis produces simulated environments. 2
The buyer has changed twice. Nvidia pursued Decart first, then Anthropic became the preferred bidder in August and opened advanced talks at around $6bn. Anthropic completed due diligence in September and withdrew, for reasons that were never made public. Amazon arrives with a relationship already in place: AWS is a customer, and in 2025 it gave Decart's founders early access to Trainium3, Amazon's in-house accelerator built to compete with Nvidia's. 2
At $7bn the deal would rank among Amazon's largest. It paid about $6.5bn for the MGM film studio in 2022 and has announced roughly $11bn for the satellite operator Globalstar this year. The reported figure sits well above Decart's last disclosed valuation: a $300mn round in May led by Radical Ventures at about $4bn, with Nvidia participating, and about $450mn raised in total. CTech noted that the startup's commercial scale remains far smaller than the price being discussed. 2
What a buyer would be paying for is a capability aimed squarely at the industry's most expensive problem. Getting more use out of each accelerator, and being able to move work off Nvidia's processors, matters more as data-centre budgets grow. Whether the talks close, and at what price against a $4bn May valuation, is open. 2
2. The White House let Britain back into the room before release
The Telegraph reported on Sunday evening in London that the White House had reinstated Britain's access to the newest AI models for safety testing, after lobbying by UK ministers. The change reverses an order last month from the Office of the National Cyber Director that told Anthropic, OpenAI and other developers not to share unreleased models with the AI Security Institute, Britain's state evaluator. That body was set up after the 2023 AI safety summit and tests the most advanced models for capability and for the risk of misuse. 34
The order had a visible effect while it stood: Anthropic restricted UK officials from reviewing its Claude Mythos 5.1 model. The Telegraph reports that decision has been reversed, and that the institute will again receive pre-release access to models from Anthropic, Google DeepMind and OpenAI. The reversal followed lobbying by the UK's AI minister, Kanishka Narayan. 3
Reuters confirms the substance of the report and adds that spokespeople for President Donald Trump, OpenAI and Anthropic did not immediately respond to its requests for comment. Reuters also notes that the administration's new AI task force said on Friday that developers must report security incidents with their models. 4
Narayan speaks on Monday, and the Telegraph reports he will warn that technology companies cannot be left to regulate themselves. His announced measures include changes to the Cyber Security and Resilience Bill that would strengthen the duty on companies working on critical national infrastructure to report security breaches involving AI, wider government use of frontier models to defend Whitehall departments, and more National Cyber Security Centre support and more resources for officials investigating incidents. On Sunday it was reported that the heads of MI6, MI5 and GCHQ had warned ministers that AI had empowered hostile actors. 3
Both halves of this story rest on unnamed officials. The Telegraph attributes the access reversal to its own reporting and to lobbying it describes without naming the officials involved, and the White House, OpenAI and Anthropic have not confirmed the arrangement on the record. Narayan's speech was still ahead when this issue closed, so its content is what the newspaper says it will contain. 34
3. Two chip chief executives described the people funding the boom
Hock Tan, chief executive of Broadcom, told an audience in Singapore on Monday that the investors behind the AI boom often cannot evaluate it. Broadcom designs components used throughout the data-centre buildout, which gives its chief executive an unusually close view of the money moving through it. 5
"A lot of investors have no clue what they are investing in," Tan said. They "set very extreme expectations only to find out that it's almost like they are in a casino." Asked about OpenAI's annualised revenue approaching $50bn in September, he allowed that it points to a viable business model and stopped there: "For now it's still losing money because they're spending a lot on training. So frankly, it doesn't tell anything at this point except that there is a lot of potential revenue growth out of providing frontier generative AI models." 5
His sharpest criticism was aimed at enterprise customers rather than investors. "When you are a kid you go to a video arcade, you play. Your parents give you $10 of tokens, you finish it in 10 minutes and you haven't learned anything," he said. "That's what happening in enterprises today." Tan named capital, rather than demand, as the next constraint on the industry's growth, and cited the Australian data-centre operator Firmus Grid's withdrawal of its listing last week as a sign that investors are questioning valuations built on future revenue. 5
Intel's chief executive, Lip-Bu Tan, joined him on the panel and addressed supply instead, saying hardware will remain in short supply for the next two to three years and that memory chips could be among the first parts of the market to lose steam. Before the two of them took the stage, DBS Group's chief executive, Tan Su Shan, said the market is repricing the risks attached to AI financing. All of this is conference commentary from named executives, and none of it is a published finding; what it establishes is where three people who sell into the buildout currently place the risk. 5
4. The IMF folded the AI boom into its growth number
Kristalina Georgieva, the International Monetary Fund's managing director, opened the fund's week-long annual meetings in Bangkok on Monday by crediting AI investment with holding the world economy together. "That shift into higher gear is what is keeping the world economy growing at around 3%, despite the fact we are being dragged down by the energy supply shock," she told the conference. "AI, love it, hate it. It's already here." The fund expects about 18,000 attendees, including finance officials from its 191 member countries. 6
Two numbers size the claim. The IMF forecast global growth of 3% for 2026 in July, and publishes an updated outlook on Tuesday. Georgieva said the AI investment boom is already delivering a "demand shock, an investment shock" that could add half a percentage point to a full point to growth in advanced Asian economies, and 0.2 to 0.8 of a point in emerging Asian ones. Her list of the most important AI supply-chain producers — the United States, China and India — with the hardware suppliers sitting in Asia is what those ranges describe. 6
On employment she was blunter. IMF research indicates that about 40% of jobs will be either augmented, downgraded or eliminated by AI, and as many as 60% in advanced economies. That prospect, she said, carries consequences for labour markets, taxation and education as income shifts from labour toward capital. An initial study cited by the fund found a tenth of US jobs already require AI skills, with routine and repetitive roles shrinking while low-paid work grows in construction and in data centres. 6
Two other officials at the same event read the jobs picture differently. Singapore's president, Tharman Shanmugaratnam, said the first rung of employment — the job a graduate takes immediately after finishing study — is already seeing less demand, and warned that AI could widen the divergence between developing and advanced economies. World Bank president Ajay Banga said he was not convinced that entry-level jobs will be destroyed and that most of the disruption will land in the advanced economies that build large models, adding that the World Bank is focused on "small AI" that does not require large computing budgets for farmers, clinics and teachers. 6
5. Tencent is borrowing against its next round of chips
Tencent told fixed-income investors on Monday that its planned bond offering of about $5bn will help finance AI-related capital spending, and said the sale could price as early as Tuesday. Bloomberg reported the call at 07:43 UTC. The offering splits into $1.5bn to $2bn across dollar tranches and 20bn to 25bn yuan, roughly $3bn to $3.7bn, of offshore yuan notes. Tencent said it has no current plans to issue shares or convertible bonds. 7
The stated use is general corporate purposes, with AI capital spending the main component and chip purchases the largest part of that, according to people familiar with the call. The spending baseline is steep. Tencent raised capital expenditure on AI projects and computing by 176% to 52.8bn yuan in the June quarter, which pushed free cash flow to an outflow of 13.8bn yuan. Management told investors the level is likely to stay elevated through the rest of 2026 and into 2027, which at the June-quarter rate implies annual spending slightly above 200bn yuan. 7
The company is the latest to fund AI through capital markets rather than cash flow. SoftBank raised $11.1bn in one of the largest corporate junk-bond offerings on record, Alibaba took a record $10.2bn in a follow-on offering, and the Chinese model maker Z.AI has raised $9.6bn across its listing, share placements and convertible bonds. Tencent did not respond to a request for comment. Whether the bond prices on Tuesday, and at what spread, will show how much of that appetite is still available to a Chinese issuer buying chips. 7
The read-through
Every one of today's items is about money or about who may look inside a model before it ships. The White House returned Britain's pre-release access to American frontier models. Amazon is negotiating for a startup whose value lies in running workloads across other companies' chips. Broadcom's chief executive says the people funding the boom cannot evaluate what they own, and Intel's says the hardware shortage lasts years. The IMF has put AI investment inside a 3% global growth forecast and counted the jobs it will move. Tencent is going to the bond market to pay for chips. 23567
These are five decisions taken by five separate actors, and nothing coordinates them. Two relationships do appear in the facts. Access and scrutiny moved in opposite directions over the weekend: a state evaluator regained its pre-release seat while the supplier of the boom described a market that does not test what it buys. And the financing calendar is filling rather than emptying, with Tencent's bond due to price as early as Tuesday and the IMF's updated forecast due the same day. Those are the two dates on which figures in this issue can be revised. 3567
Where each figure in this issue rests. The Amazon talks come from people familiar with them, relayed by Bloomberg and CTech from the Journal's reporting; no price, term or structure has been confirmed by either company, and CTech's account of the two earlier bids comes from its own reporting and the Journal's. The access reversal rests on Telegraph reporting attributed to unnamed officials, with Reuters repeating the headline claim and obtaining no comment from the White House or the two labs. The Broadcom and Intel remarks are conference comments by named executives, accurately quoted but not findings. The IMF figures are the fund's own research as relayed by its managing director, with the jobs number stated as a share of all jobs and the growth contribution given as a range. Tencent's numbers come from people familiar with an investor call, not from a filing, and the June-quarter capex and cash-flow figures are the company's reported results. 123567
Watch next
- Whether Amazon signs, and at what price. A completed deal near $7bn would nearly double Decart's May valuation of about $4bn; a signed agreement would also put software that spreads workloads across chips inside the company building Trainium, which changes what AWS can promise customers who want to buy less Nvidia silicon. 2
- What Narayan's speech actually contains. A duty on critical-infrastructure firms to report AI-related breaches is a legal obligation, and the difference between what a minister announces and what reaches the statute book is where this item's impact lives. 3
- Whether capital becomes the binding constraint, as Hock Tan said. The test is visible in the next data-centre IPO after Firmus Grid's withdrawal and in the order books for debt issued against future AI revenue. 5
- What the IMF's updated outlook says on Tuesday. It will either hold the 3% global figure that Georgieva's remarks imply or move it, and it will show how much of that number the fund attributes to AI investment specifically. 6
- What Tencent's bond prices at. The spread will say what bond investors will charge a Chinese technology company borrowing to buy chips while it reports a free-cash outflow. 7
References
- 1Bloomberg — Amazon Is in Talks to Buy AI Startup Decart, WSJ Reports
finance.yahoo.com
- 2
- 3
- 4
- 5
- 6
- 7
- 8
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.