By Krystal Hu

Oct 29 (Reuters) - (Artificial Intelligencer is published every Wednesday. Think your friend or colleague should know about us? Forward this newsletter to them. They can also subscribe here.)

When my colleague Greg Bensinger broke the story on Amazon AMZN.O preparing to lay off 30,000 corporate employees this week, it sent shockwaves through Silicon Valley and beyond. It’s one of the largest tech layoffs since ChatGPT burst onto the scene, and a stark reminder of how the AI era is reshaping even the most powerful companies.

Amazon said the cuts are part of an effort to make the company “leaner and less bureaucratic” as it retools for the AI wave . The idea is to increase efficiency before investing heavily in new AI initiatives — a mindset that echoes the revenue-per-employee obsession venture capitalists use when judging AI startups.

CEO Andy Jassy had already hinted at this back in June, warning that the company had grown “too bureaucratic”, while making clear that as AI automates more repetitive and routine work, further job reductions were inevitable.

But beneath the surface, Amazon’s AI challenge runs deeper than workforce automation. The company is contending with tariffs, weaker consumer spending and a cloud business that investors fear could lose its edge due to its AI offerings.

It’s a reminder that the AI transformation isn’t just about replacing jobs; it’s about rewriting how the business makes money. AI-related layoffs may not be widespread yet, but the tension between efficiency and innovation is becoming the defining corporate theme of this new era.

In this week’s edition, we’re charting how corporations plan to pour more money into AI in the next few years. I also got an insider look at the year-long negotiations that led to the announcement this week of OpenAI’s restructuring, one of the most significant business reorganizations in tech history. Scroll on.

Email me or follow me on LinkedIn to share any thoughts.

OUR LATEST REPORTING IN TECH & AI:

• Altman touts trillion-dollar AI vision as OpenAI restructures to chase scale

• Nvidia will build AI supercomputers for US Energy Department, wants to get back into China

• US startup Substrate announces chipmaking tool that it says will rival ASML

• US nuclear generation to grow 27% post 2035, as data centers fuel power demand

• Nvidia's $1 billion stake sends Nokia to decade high on AI hope

• OpenAI to offer ChatGPT Go free for a year in India

INSIDE THE DEAL THAT FREED OPENAI’S AMBITION:

It took over 15 months, an army of bankers and lawyers, and countless walkthroughs with regulators to pull off one of the largest and most complex restructurings in tech history.

From the time they began discussing OpenAI’s transition to the moment they announced the deal on Tuesday, its valuation had soared fivefold — from less than $100 billion to $500 billion.

People close to the talks told me the meteoric rise in valuation was key to getting the greenlights from regulators and Microsoft MSFT.O to make sure they feel the win-win.

The new deal had to satisfy almost every power center with a stake in OpenAI — its for-profit arm, its non-profit parent, its once-biggest investor Microsoft, and even the attorneys general of California and Delaware. The goal: reform OpenAI’s convoluted “capped-profit” model to set it free to raise capital and move fast, while avoiding legal and regulatory risks by evolving beyond its non-profit roots.

For Sam Altman, the stakes were high. As ChatGPT’s costs and popularity exploded, OpenAI needed the kind of capital — and freedom — that its initial structure could not support. “They were stuck asking Microsoft for permission on every raise, every acquisition, every partnership,” one person involved in the talks told me. “That was unsustainable.”

Over the past year, each side built out its own advisory army. Goldman Sachs and veteran banker Michael Klein advised OpenAI’s non-profit parent, while Morgan Stanley represented Microsoft. In a rare move, even the Delaware Attorney General hired independent advisers to examine the proposals line by line.

Regulators were eventually convinced by the math, as OpenAI made a major concession following my story last year that revealed its plans to remove non-profit control — a move that drew strong opposition from those who argued the company was violating its charitable mission.

Now the non-profit will keep roughly a third stake in OpenAI, plus warrants to receive additional shares tied to “extraordinary outcomes” — like achieving AGI — while maintaining control over board appointments to the for-profit. To regulators, the numbers looked better than the previous structure, where the non-profit’s upside was whatever remained after capped investor returns. Now, it immediately becomes one of the best-funded non-profits in history, at least on paper, tied to OpenAI’s valuation.

Then came the hardest negotiation: Microsoft. Once entitled to 49% of OpenAI’s future profits through its $13 billion investment, the giant will now hold about 27% of OpenAI’s equity and roughly 20% of its revenue, turning its $13 billion investment into a $135 billion stake. It keeps exclusive API access through 2032, even in a post-AGI world — but gives up the right to approve OpenAI’s major strategic decisions, including acquisitions or a future IPO.

In short, OpenAI traded some economics for freedom. It can now raise capital, buy companies, and partner with infrastructure players— Nvidia and AMD— without Redmond’s sign-off.

The overhaul clears the runway for OpenAI’s next chapter: a public listing and a new world of financing. And for Sam Altman, the dealmaker who’s already rewritten Silicon Valley’s playbook, the pace and the stakes are only getting higher from here.

As one insider put it, “This was the most complicated deal any of us have ever touched — the one that finally lets OpenAI act like a $500 billion company.”

CHART OF THE WEEK:

Corporations’ willingness to spend on AI tools shows no sign of slowing down — after all, who doesn’t want more efficiency and better margins? The chart, based on Gartner forecasts, shows that AI’s share of corporate software spending is growing fast. As global enterprise software spending heads toward $2 trillion by 2029, about $76 billion of that is expected to go to generative AI.

Big companies are already signing up for broader deployments, betting that AI can boost productivity — from speeding up call-center ticket resolutions to helping engineers write code faster. The results have been uneven so far, but AI firms that can turn pilot projects into long-term, high-value contracts stand to claim the biggest slice of that expanding software pie.

AI RESEARCH TO READ

By Kenrick Cai, Technology Correspondent

Chinese AI firm DeepSeek has published research suggesting it can dramatically lower the cost of AI by converting text into images, a method that tests the adage that "a picture is worth a thousand words.” In a research paper this month, the lab presented a new AI model called DeepSeek-OCR that accurately converts large chunks of text into images for processing. The AI can then reconstruct the text from the images.

It turns out that at the current capabilities of the technology, a picture is worth about seven or eight words. By compressing that amount of text into an image, researchers found that the AI could still retain 97% accuracy. The accuracy rate quickly fell off a cliff once they attempted to store more words in a single image. But even at this starting point, AI systems can process longer documents more cheaply and with less computing power compared to the traditional method of storing words as text data.

The research addresses one of the biggest hurdles in AI: the cost and computing power required to process vast amounts of information. Even large language models from well-capitalized companies like Google and OpenAI struggle with processing very long texts due to the massive amounts of computing power needed to perform the task. DeepSeek’s latest research suggests that the path forward doesn’t necessarily have to be ever-larger LLMs.

(Reporting by Krystal Hu; Editing by Lisa Shumaker)

((krystal.hu@thomsonreuters.com, +1 917-691-1815))