The tech world doesn’t do subtle. This week delivered the kind of market signals, geopolitical maneuvers, and infrastructure pivots that separate the temporary headlines from the structural shifts. IBM lost a quarter of its value in a single day. Xi Jinping launched a 29-nation AI alliance in Shanghai. TSMC doubled down on American soil. Chip stocks cratered. Google’s Gemini slipped again. And Elon Musk bought a power plant. Here’s what actually matters.
IBM’s $69 Billion Wake-Up Call: The SaaSpocalypse Is Real
IBM shares collapsed 25% on July 14 — the single worst day in the company’s 100+ year history — erasing $69 billion in market value. The trigger? A surprise earnings warning revealing that enterprise customers are shifting spending away from software and toward AI infrastructure: servers, storage, and memory. Software revenue grew 5%, but infrastructure revenue dropped 7%, and the forward guidance spooked everyone.
This isn’t just an IBM story. It’s the first hard data point backing the “SaaSpocalypse” thesis — the increasingly credible argument that AI is cannibalizing traditional software budgets at the enterprise level. When CFOs have to choose between another SaaS renewal and GPU capacity for internal AI workloads, the GPU is winning.
Why it matters: IBM’s crash is a canary in the coal mine for the entire SaaS sector. If enterprise IT budgets are truly zero-sum between AI capex and software opex, every software vendor with a per-seat pricing model has a structural headwind coming. The Q2 earnings season — with Alphabet, ServiceNow, and Apple reporting later this month — will tell us whether this is an IBM-specific problem or the start of something much bigger.
Xi Jinping Launches a 29-Nation AI Alliance in Shanghai
On Friday, July 17, Xi Jinping opened the World AI Conference (WAIC) in Shanghai with a full-throated pitch for a new global AI order — one led by China. He announced AI training programs and cooperation centres with BRICS, ASEAN, Latin American, and African Union nations. Twenty-nine countries signed on to a new World AI Alliance framework the day before his speech.
At the same conference, China’s Moonshot AI unveiled what it claims is the world’s largest open-weight AI model, closing the gap with US frontier labs. And DeepSeek — already capturing 30–46% of US enterprise API token traffic through developer platforms — is now seeking $1.5 billion at a $71 billion valuation, just weeks after closing $7.4 billion at roughly $50 billion. The IPO filing is expected as soon as late 2026.
The throughline is clear: China is no longer just competing on models — it’s building the diplomatic, financial, and ecosystem infrastructure to make its AI stack the default choice for the developing world.
Why it matters: The AI cold war just entered its institutional phase. The US leads on frontier capability, but China is building a coalition of nations that will train on its models, deploy on its infrastructure, and align with its governance frameworks. The battle isn’t just about who builds the best model — it’s about who defines the standards, trains the workforce, and owns the developer ecosystem for the next decade.
TSMC Bets Another $100 Billion on American Soil
TSMC announced a fresh $100 billion US investment, bringing its total American commitment to $265 billion. The plan: four additional semiconductor fabrication facilities in Arizona, for a total of 12 semiconductor and packaging plants. The news came alongside a stellar Q2: revenue up over 30% year-over-year to $40.2 billion, driven almost entirely by AI chip demand.
The investment is both an economic hedge and a geopolitical necessity. With US-China tensions escalating and Taiwan’s strategic vulnerability under constant discussion, TSMC is effectively building a second manufacturing backbone outside the island. The Trump administration framed the deal as a national security win, and it is — but it’s also a recognition that AI chip demand is so immense that even $265 billion might not be enough.
Why it matters: TSMC manufactured essentially every advanced AI chip in existence. Its decision to physically diversify production changes the risk calculus for the entire AI supply chain. But the timeline matters: these facilities won’t come online at scale until the late 2020s. In the meantime, the world’s AI compute supply chain still runs through a single island in the Taiwan Strait.
The AI Chip Rout: Markets Reprice the Hype
It was a brutal week for AI-exposed equities. SK Hynix shares plunged as memory chip demand expectations softened. SoftBank sank over 9%, tracking a broader Asian chip sell-off. Even the mighty Nvidia wasn’t immune — Apple overtook Nvidia to become the world’s most valuable company, a rotation that says as much about AI fatigue as it does about Apple’s resilience.
The sell-off wasn’t random. It followed a week of signals: IBM’s warning that AI infrastructure spending is cannibalizing software, reports that hyperscaler spending growth may be slowing, and growing political backlash against data center construction in US communities. The AI trade that drove markets for 18 months is finally being interrogated.
Why it matters: The chip rout isn’t signaling that AI demand is collapsing — TSMC’s numbers prove otherwise. It’s signaling that the market had priced in infinite growth and is now recalibrating to something closer to reality. The AI buildout continues, but the “everything goes up” phase is over. Selectivity is back.
Google’s Gemini 3.5 Pro: Delayed Again
Google DeepMind postponed Gemini 3.5 Pro once more, with Bloomberg reporting the model fell short of internal goals after the team scrapped its original foundation and rebuilt the architecture from scratch. The new target date? July 17 — today, at the time of writing. Meanwhile, four senior Google researchers left for Anthropic in recent weeks, adding a talent-retention subplot to an already difficult launch.
The delay is particularly awkward because Gemini 3.5 Pro was meant to be Google’s answer to GPT-5.6 and Anthropic’s Claude Opus 4.8. Instead, Google is watching from the sidelines as OpenAI ships broader access and Anthropic captures mindshare among developers and enterprises alike.
Why it matters: Frontier model development is not a linear process. Google’s struggles show that even with near-infinite compute and talent, building a competitive top-tier model is genuinely hard. The delay also creates a window for open-weight models — including China’s — to close the gap while Google regroups.
Elon Musk Buys a Power Company to Feed Grok
In what might be the most on-brand story of the week, Elon Musk’s xAI acquired APR Energy — a fleet of more than 1 gigawatt of mobile and diesel turbines — for roughly $1 billion. The reason? xAI’s Colossus 2 data center needs power, and the grid can’t deliver it fast enough. Reuters also reported that xAI is running nearly 60 gas turbines without permits, with pollution disproportionately affecting Black communities in the area.
The acquisition is a blunt acknowledgment of what many in the industry have been saying quietly: power, not chips, is now the binding constraint on frontier AI. Buying a turbine company is one way around the queue, but it comes with environmental and regulatory baggage that won’t stay quiet for long.
Why it matters: The AI industry’s energy problem has moved from PowerPoint slides to real-world consequences. When a leading AI lab buys a fossil fuel power company to keep training runs going, it forces a conversation about whether the pace of AI development is compatible with climate commitments — and who gets to decide.
Quick Hits: What Else Happened
Databricks raised a Coatue-led round at a $188 billion valuation — the data and AI platform continues its meteoric rise, cementing its position as one of the most valuable private companies in the world.
The White House launched an AI cybersecurity coordination group with open-source companies and critical infrastructure operators to share vulnerability intelligence. This follows Anthropic’s Claude Mythos release, which demonstrated the ability to identify and exploit zero-day vulnerabilities at machine speed.
The UK designated Microsoft, Google, Amazon, and Oracle as “Critical Third Parties” for the financial sector, giving the Bank of England and FCA direct regulatory oversight over their cloud services — a significant expansion of government authority over Big Tech infrastructure.
Helsing, Europe’s leading AI defense startup, raised $1.8 billion at an $18 billion valuation — the largest defense-tech round in European history, reflecting surging investor appetite for dual-use AI technologies amid rising geopolitical tensions.
OpenAI launched GPT-5.6 to broader access after US government review, while simultaneously facing a sanctions motion from The New York Times over alleged evidence withholding in the copyright lawsuit. The contrast between product momentum and legal headwinds is becoming OpenAI’s defining tension.
What I’m Watching
Three things on my radar for the weeks ahead:
- Q2 tech earnings. Alphabet, Apple, and ServiceNow report in the coming weeks. If IBM’s AI-cannibalization pattern shows up elsewhere, the software sector has a real reckoning ahead. If it doesn’t, IBM looks like a company-specific problem rather than a sector trend.
- Xi’s AI alliance in practice. Twenty-nine countries signed on. Now we see whether it translates into procurement decisions, training partnerships, and regulatory alignment — or whether it’s a photo-op that fades after the conference ends.
- Gemini 3.5 Pro’s actual launch. If it ships and performs, Google re-enters the frontier conversation. If it slips again, the narrative shifts from “delayed” to “in trouble” — and that’s a much harder hole to climb out of.
What are you watching? Drop a comment or reach out — I’d love to hear what’s on your radar.
— Jody