Top Stories

Experts worried that US’s weak power grid has already given China the AI lead
China’s AI edge may be less about chips and more about electricity. After touring China’s AI hubs, analyst Rui Ma says energy is treated as a solved problem there, which is in stark contrast to the U.S., where Goldman warns AI demand is outrunning a decade-old grid. The implication is blunt: China can stand up data centres faster because it has power to spare.

• China’s power overbuild. According to electricity expert David Fishman, China’s reserve margin hasn’t dipped below 80%–100% nationwide, and data centres are used to “soak up oversupply.” He claims China adds more annual electricity demand than Germany’s total consumption each year.

• U.S. grid bottlenecks. Deloitte flags power constraints as the limiting factor for U.S. data-centre buildout, forcing some firms to build private plants. Households feel it too with Ohio bills reportedly rising at least $15 a month due to data centres. Goldman calls it a “critical bottleneck.”

• Different playbooks. China’s long-term, state-led planning funds capacity ahead of demand. The U.S. relies on private capital seeking 3–5 year returns, plus permitting and local opposition that slow new generation and transmission.

Why it matters: AI leadership now hinges on energy infrastructure as much as algorithms. If China can deploy compute faster by tapping abundant power (even via coal as a bridge), it could tilt the AI race. The claims are sweeping and deserve scrutiny — especially on emissions and regional realities — but the signal for executives is clear: in AI, power is product.

Low-cost ChatGPT Go rolls out in India
OpenAI has launched ChatGPT Go in India (which is their second-largest market) at Rs 399 a month (about $4.60 USD), which they call their most affordable plan yet, and it includes access to GPT-5.

• Low-cost tier for India. ChatGPT Go sits below Plus at Rs 1,999 and Pro at Rs 19,900 (roughly $23 USD and $230 USD), which suggests a push to convert heavy free users without the price shock of higher plans.

• More capability than free. The pitch is broader access at a fraction of the usual cost. Go users get GPT-5 with improved Indic language support, plus 10 times more messages, image generations and file or image uploads, and two times longer memory, according to OpenAI.

• Payments and availability. Subscriptions can be bought via UPI or major Indian payment methods, which should lift conversion.

Why it matters: Price and payments localisation often decide who wins at scale in India. This move signals OpenAI’s intent to monetise a vast user base, which could boost revenue but will also test margins and support quality. For AI leaders, the lesson is simple: tailor plans to market realities and remove payment friction if you want real adoption.

Nvidia working on new chip for China that is more powerful than the H20
Nvidia is reportedly building a new China-focused AI chip, the B30A, that would beat its current H20 while staying within export limits. It is based on the latest Blackwell architecture with a single-die design that is said to deliver roughly half the compute of the flagship B300. Samples could reach Chinese customers as soon as next month, although regulatory approval remains very uncertain.

• Specs and timing. The B30A would include high-bandwidth memory and NVLink, similar to H20, which suggests strong interconnect and memory performance. Reuters’ sources say Nvidia aims to ship samples for testing next month, but specifications are not final.

• Rules in flux. Trump has hinted he might allow scaled-down Blackwell sales in China and floated a new revenue-sharing arrangement, which means policy could shift quickly. U.S. lawmakers remain wary that even reduced chips could erode America’s AI lead.

• Market and competitors. China accounted for 13% of Nvidia revenue last year. Nvidia only just resumed H20 sales in July after an April halt, and it is also prepping a cheaper RTX6000D for inference at 1,398 GB per second which sits just under a 1.4 TB per second control threshold. Nvidia argues it must stay present so buyers do not switch to Huawei, although Chinese state media have raised security concerns about Nvidia chips.

Why it matters: If approved, a Blackwell-class B30A could keep Nvidia’s software ecosystem embedded in China, which would blunt Huawei’s momentum and protect a key revenue stream. If blocked, the gap may widen for domestic rivals. Either way, this shows AI leadership is now a policy product as much as a silicon product, so executives need contingency plans across chips, geographies and supply chains.

In Brief

• AI eroded a doctor’s ability to spot cancer after only a few months of use

• Claude can now end abusive chats

• DeepSeek V3.1 open model has just dropped

• Ex-Google exec calls law/medicine degrees “obsolete” with AI (mentions nothing about company execs)

• Warmer-sounding LLMs more likely to repeat false information and conspiracy theories

Market Trends

OpenAI staffers selling $6 Billion in stock to SoftBank
OpenAI is in talks to run a roughly $6 billion employee stock sale that would value the company at about $500 billion. If it goes ahead, this would not bring new cash into the business, which means it is mainly about liquidity for current and former staff while setting a headline valuation in a jittery market.

• Employee-led secondary at a sky-high price. CNBC reports the shares would be sold by employees at a roughly $500 billion valuation. It is a secondary transaction which means no new primary capital for OpenAI’s balance sheet.

• Investor lineup and deal status. SoftBank, Dragoneer and Thrive Capital are in discussions, according to a source, and Thrive could lead. Talks are early and details may change, and Bloomberg first reported the process.

• Product backdrop and growing pains. The move comes a week after OpenAI announced GPT-5, which it says is “smarter, faster and a lot more useful.” The rollout has been rocky, though, with some users losing access to prior models, and Sam Altman conceded they “underestimated” how much people liked GPT-4o.

Why it matters: A $500 billion tag on a secondary sale signals unbroken investor appetite for the perceived AI leader, even as product transitions annoy users. It also underlines a strategic shift toward employee liquidity as a retention tool, which is logical given hiring pressures. The caveat is obvious, since tender valuations can be fragile and the deal is not done. For leaders, the message is that market confidence in AI remains high, but cash for chips and data centres still needs primary funding.

Silicon Valley deals creating “zombie startups”
Silicon Valley’s AI consolidation is happening without classic takeovers, as Big Tech strikes multi‑billion talent and licensing deals that siphon founders and researchers and leave “zombie” shells behind. Windsurf is the latest example, where OpenAI talks collapsed and Google then hired leaders via a reported $2.4 billion licensing deal. The workaround is simple: use minority stakes and IP licences which avoid premerger reviews while hoovering up the brains.

• The new playbook. Menlo’s Matt Murphy calls it a “new playbook,” where acquirers “buy exactly what we want and leave the rest behind.” Public Citizen says firms are edging “as close as possible” to majority stakes without “tripping any alarms.”

• Fallout for startups. Touring Capital’s Samir Kumar says you “hollowed out the organization,” and employees at Windsurf were “crying.” Cognition later bought Windsurf’s IP and brand for a reported $250 million, which is far below rumours tied to OpenAI talks, and offered opt‑out buyouts.

• Exceptions and pushback. Meta put $14.3 billion into Scale AI for a 49% stake and hired CEO Alexandr Wang, then Scale cut 14% of staff, although it says it has over 1,000 employees and nearly $1 billion in revenue. Google disputes the “zombie” label at Character and Windsurf, and the FTC is probing deals like Microsoft‑Inflection and Amazon‑Adept.

Why it matters: This is consolidation by another name, which concentrates talent and compute inside the giants while leaving investors and rank‑and‑file holding the bag. Founders may get life‑changing cheques, but product roadmaps, customers and competition suffer. If you are building in AI, plan retention and secondary liquidity early, or risk becoming the next cautionary tale.

Sam Altman sees an “AI bubble” forming
OpenAI CEO Sam Altman says AI may be in a bubble, likening the hype to the dot-com era, which is a striking admission from the industry’s poster child. He also says AI is “the most important thing to happen in a very long time,” which is a useful reminder that bubbles can form around real breakthroughs.

• Bubble talk with historical echoes. Altman’s comments mirror warnings from Joe Tsai, Ray Dalio and Torsten Slok, who argues this boom looks bigger than the 1990s internet run-up. The Nasdaq lost nearly 80% after the dot-com peak, which is the spectre hovering over today.

• Not everyone buys the bubble label. Analyst Ray Wang says supply-chain fundamentals remain strong and long-term demand supports continued investment, although he concedes speculation is chasing weak stories. DeepSeek’s claim of sub-$6 million training costs fuels questions about capital intensity.

• Product and profit reality check. OpenAI says ARR could top $20 billion, yet it is still unprofitable. GPT-5’s launch drew criticism, so OpenAI restored access to GPT-4 for paying users, and Altman now says the term “AGI” is losing relevance.

Why it matters: Leadership teams should treat AI like any transformative platform, which means separate substance from sizzle. Capital is flowing, valuations are frothy, and unit economics still need proving at scale. Plan for durable ROI and user value, not just multiple expansion.

Recommended Reading

Watch: What AI can and can’t do

Earlier this year, Princeton professor Arvind Narayanan gave a talk at MIT, discussing the things AI can and can't to, and how to spot the so-called "snake oil" that companies and tech CEOs are forever trying to lob our way.

Hit reply to let us know which of these stories you found the most important or surprising! And, if you’ve stumbled across an interesting link/tweet/news story of your own, send it our way at [email protected] It might just end up in the next issue!

Thanks for reading. Stay tuned for the next AI Leadership Weekly!

Brought to you by Data Wave your AI and Data Team as a Subscription.
Work with seasoned technology leaders who have taken Startups to IPO and led large transformation programmes.