Headlines this week - Aug 17, 2025

A look at how capital is being deployed across future opportunities

C Gavilanes

Aug 17, 2025

This week in the future:

1 - Nvidia and AMD agree to a 15% tariff on AI chip sales to China

Both companies will give the US government a 15% cut of revenue. In an unusual arrangement, both Nvidia and AMD have agreed to give the U.S. government a 15% cut of their revenue from AI chip sales to China. This deal has apparently been struck in exchange for the administration lifting the previous export ban on these components.

The deal is still seen as favorable for Nvidia. The FT’s Lex Column claims that paying the tariff is "worth it for Nvidia," as it allows the company to restart sales and shift excess inventory. The article notes that the high profit margins on these chips can easily absorb the cost of the new tax.

However, the measure is not expected to boost US chip manufacturing. This new tariff policy is unlikely to solve the U.S. chip-making dilemma, according to a WSJ analysis. The article argues that the core issue is not cost but the complex, globalized supply chain, and that tariffs alone will not be enough to reshore advanced manufacturing.

2 - Meanwhile, Beijing is raising concerns over Nvidia's chips

The Chinese government is now discouraging the use of Nvidia's H20 chips. Just as sales were set to resume, Beijing has begun urging local firms to avoid Nvidia’s H20 chips, particularly for government-related projects. While this is not an outright ban, it signals a strong official preference for domestic alternatives.

The move reflects a deep-seated fear of reliance on US suppliers. This guidance is part of a broader strategy to avoid becoming "too reliant" on American technology for critical infrastructure. While Chinese firms are keen to exploit the chips' performance, Beijing fears the potential for security vulnerabilities and geopolitical leverage.

New US tracking measures could actually justify (in part) these security concerns. Beijing's fears may have been confirmed by a Reuters exclusive this week, which revealed the US government is embedding trackers in AI chip shipments. This new capability is designed to catch illicit diversions of high-end chips to China, confirming that the hardware can be monitored remotely.

However, domestic alternatives (i.e. Huawei) do not yet look mature enough. China's push for self-sufficiency is facing technical hurdles, as highlighted this week by an FT report on top AI firm DeepSeek. The company’s attempt to train its latest model on Huawei’s chips was reportedly unsuccessful, forcing a delay and a return to Nvidia hardware for the critical training phase.

3 - Private capital's growing role in the AI build-out raises new questions

Private equity is now a major player in data center financing. Private equity and credit funds are now essential sources of capital for the $3trn AI building boom, as reported this week by the FT. These firms are stepping in to fund the massive data center projects that are too large or risky for traditional lenders.

But the debate continues on the rationality of these massive investments. An blog post by Andrew Cote this week warns of a potential decline in “marginal intelligence," where the immense cost of training ever-larger models yields diminishing returns. This raises new questions about the long-term economic viability of the current AI infrastructure investment thesis (particularly for efforts linked to training).

A social and political backlash may also be brewing. As data centers expand, they are creating a new political battleground "in people’s backyards,” according to a POLITICO report this week. Local communities are increasingly pushing back against the construction of these massive facilities, creating a new and unpredictable risk for the industry.

4 - Embryo screening is becoming a new trend for affluent tech consumers

Startups in the field are now targeting a growing affluent clientele. A WSJ report this week details Silicon Valley's "growing obsession with having smarter babies," as a new industry of startups now offers embryo screening not just for disease, but also for traits like intelligence and height, as previously reported here. The services would now be becoming increasingly popular among an affluent, tech-savvy consumer base.

5 - The race to build the first AI "SuperApp" is heating up

OpenAI's GPT-5 is seen as a key step toward a "SuperApp”. A SemiAnalysis report argues that GPT-5's new "Router" feature is designed to turn ChatGPT into a "Consumer SuperApp". This will integrate various services, setting the stage for new ad monetization models and making the app a central hub for users.

Meanwhile, xAI has started to use a freemium model to attract a mass user base. xAI is also competing for consumer attention by offering limited free access to its powerful Grok 4 model. The strategy aims to impress non-paying users enough that a percentage will convert to paid subscriptions once they hit usage limits.

This competition is now spilling over into the App Store The rivalry has escalated, with Elon Musk now accusing Apple of unfairly favoring ChatGPT in its App Store rankings. This highlights the growing importance of mobile distribution in the battle to become the dominant consumer AI platform.

6 - The Altman-Musk rivalry is now expanding to brain-computer interfaces

OpenAI is backing a new Neuralink competitor. OpenAI and its co-founder Sam Altman are preparing to back Merge Labs, a startup that will compete directly with Elon Musk's Neuralink. This move would heighten the rivalry between the two billionaires, opening a new front in the battle for technological supremacy.

7 - Perplexity looks at the browser as a key tool for consumer AI monetization

This week the company made a $34.5bn offer to buy Google Chrome. In a move that caught the industry by surprise, Perplexity has made an unsolicited $34.5bn all-cash offer for Google's Chrome browser. The WSJ reports that the AI search startup is betting that the browser, not a standalone app, will be the key to monetizing consumer AI.

The move is widely seen as a tactic to influence regulators. The FT describes the bid as part of the "AI dealmaking circus," suggesting the primary motive is to increase regulatory pressure on Google. The article notes that the bid comes as a federal court is deciding on remedies in an antitrust case against Google, which could include the forced divestiture of Chrome.

8 - The path to AI monetization in B2B remains uncertain

The current AI narrative is more focused on investment than on practical applications. An FT article this week noted that the overwhelming AI narrative is not about practical business use but about massive valuations and geopolitical competition. The piece highlights a growing frustration with the wide and persistent gap between the hype and the reality of AI's current utility.

Companies are investing heavily in AI, but with little to show for it so far. Enterprises are "pouring billions to prepare for A.I." but the technology "has yet to pay off," according to a NY Times report. The article describes an "A.I. paradox" where, despite huge investment, many companies report "no bottom-line impact" and a high failure rate for pilot projects.

Even the most successful vendors, like Palantir, have a product that is difficult to define. Despite its soaring valuation, many still struggle to understand what Palantir actually sells. A WIRED feature this week described the company as a "consulting firm that builds bespoke software," highlighting the complexity of its offering and the challenge of pinning down a single, clear value proposition.

9 - The backlash over GPT-5 reveals new "safety" issues for consumer AI

The user revolt focused on a less "empathetic" chatbot. Following the GPT-5 launch, a "massive user revolt" forced OpenAI to bring back a fan-favorite older model. The core complaint was that the new, more powerful model was less engaging and empathetic, a change that many users experienced as a personal loss.

This suggests consumer AI is fulfilling a social, not just a utilitarian, need. A Bloomberg analysis argues that this backlash reveals one of the technology's most "concerning problems": users are forming deep emotional attachments to their chatbots. This shows that many people are using the technology to satisfy a need for personal connection, not just for productivity.

And the industry is not yet prepared for these "relationship issues”. AI companies are now running a "giant social experiment" without fully understanding the consequences. The article suggests that by designing models to be empathetic, companies are encouraging a level of emotional attachment that they are not yet equipped to manage safely or responsibly.

10 - Confidence is growing that a "workable" quantum computer will arrive by 2030

Industry leaders Google and IBM are increasingly optimistic. Both Google and IBM now believe the first workable, industrial-scale quantum computer is in sight, with a target date of 2030. A recent series of technical breakthroughs has fueled a new wave of optimism that the immense engineering hurdles can be overcome.

Microsoft, though more bullish, may be further behind. Microsoft has taken a different and more ambitious approach, betting on a more exotic type of "topological qubit". While this could ultimately prove to be a more stable design, it is seen by many in the industry (including theoretical physicist Sabine Hossenfelder) as a less mature technology at this stage of the race.

Meanwhile, preparations for post-quantum cryptography are lagging. This accelerating progress is creating a new sense of urgency around cybersecurity. An IEEE Spectrum article this week warns that the world is not adopting post-quantum cryptography fast enough, creating a major risk that our current encryption standards will be vulnerable before new, quantum-resistant ones are widely deployed.