How's the AI chip market doing these days?
In our AI chip market deck, you will find everything you need to understand the market
SUMMARY
How's the AI chip market doing these days? The AI chip market is still doing very well, but the easy, broad phase is over.
The strongest signal is not just Nvidia’s scale. It is that recent evidence points in the same direction across public chip vendors, hyperscaler capex, memory suppliers, AI cloud infrastructure, and semiconductor startup funding.
Money is still rushing into AI chips, but it has become much more selective. Investors are now favoring bottlenecks such as inference, memory bandwidth, non-Nvidia capacity, custom silicon, and deployed AI infrastructure rather than generic accelerator stories.
Nvidia remains the gravity center of the market. The gap with AMD is still too large to frame the market as a simple catch-up story, especially when Nvidia’s data-center revenue is roughly thirteen times larger in the latest comparable results used in the analysis.
The hyperscaler signal is still extremely strong. Microsoft, Meta, Amazon, and Alphabet are not behaving like buyers waiting for a slowdown; their capex and cloud data suggest AI infrastructure demand is still expanding from an already huge base.
The market’s center of excitement is shifting toward inference. Training remains expensive, but inference turns compute into a recurring cost every time users query models, run agents, generate images, or call AI APIs.
Custom AI chips are now a real market, not just internal hyperscaler experiments. Broadcom, Marvell, Google’s TPUs, and hyperscaler-specific silicon all show that workload-specific chips are becoming a serious revenue pool.
The supply shortage has changed shape. The market is no longer only about GPU availability; the tighter constraints now sit in HBM, advanced packaging, power, networking, and full-system deployment.
Memory is one of the clearest hidden winners. HBM and high-capacity server memory are becoming gatekeepers for AI system scale, which makes memory suppliers more strategically important than a normal cyclical recovery story would suggest.
The Nvidia alternative story is working only in narrow forms. AMD clouds, TPUs, custom ASICs, and inference chips matter when they solve a specific buyer pain, but broad “Nvidia killer” narratives still look weak.
China is no longer a clean growth engine for Western AI chip companies. Export controls, Chinese policy signals, and possible tighter enforcement through Taiwan make that demand pool much harder to model.
The final read is that the operating market is stronger than the stock-market setup. AI chip demand is still hot, but investor expectations have become so high that even strong results can disappoint if they do not beat an already aggressive story.
This market map, featured in our AI chip market deck, highlights top companies and startups in the AI chip market
Is money still rushing into AI chips right now?
Yes. The AI chip market is still attracting serious money, but the money is much pickier than it was during the first AI infrastructure wave.
The lazy answer would be to say “Nvidia is huge, therefore the market is hot.” That is true but incomplete. What looks more interesting today is that money is still flowing across several different layers: public chip vendors, semiconductor startups, AI cloud infrastructure, memory suppliers, and hyperscaler capex. That gives us a stronger signal than one isolated stock chart.
On the public side, Nvidia’s May 2026 quarterly results were still absurdly strong: $81.6 billion of revenue, up 85% year over year, with data-center revenue at $75.2 billion, up 92%. Broadcom’s June 2026 results added a different proof point: AI semiconductor revenue hit $10.8 billion, up 143% year over year, and the company guided to $16 billion next quarter. AMD’s May 2026 results were smaller but still relevant, with data-center revenue up 57% year over year to $5.8 billion.
On the private side, semiconductor startup funding also stayed active. Crunchbase reported in June 2026 that semiconductor startups had already raised around $10.7 billion in 2026, putting the category on track to beat last year. That is not just a generic AI software funding bubble; chip startups still need much bigger checks, longer timelines, and more investor patience.
The nuance is that investors are not funding every “AI accelerator” story blindly anymore. Groq’s recent fundraising reports, TensorWave’s $350 million Series B at a $1.55 billion valuation, and Positron’s $230 million raise all point toward a narrower funding thesis: inference, alternative compute capacity, AMD-based AI clouds, and hardware that reduces Nvidia dependency.
So yes, money is still rushing in. But today it is flowing toward bottlenecks, not slogans.
If you want more recent data on this point, please see our latest AI chip market report.
Is Nvidia still basically untouchable these days?
Yes. Nvidia is still the default power center of the AI chip market, and the latest numbers make the gap look wider rather than narrower.
The obvious signal is revenue scale. Nvidia’s Q1 fiscal 2027 data-center revenue was $75.2 billion. AMD’s Q1 2026 data-center revenue was $5.8 billion. Even if AMD is growing well, Nvidia’s data-center business was roughly thirteen times larger in the latest comparable results we checked. That is too large to explain away with “AMD is catching up.”
The more important signal is reporting structure. Nvidia recently moved away from reporting gaming GPUs as a separate segment and reorganized around deployment markets like data center and edge computing. That tells us the company itself now sees AI infrastructure as the organizing center of the business, not as a fast-growing segment next to gaming.
Nvidia’s margin profile also matters. In its May 2026 results, the company reported roughly 75% gross margin while still growing revenue 85% year over year. That combination is rare. Usually, when a hardware company scales this fast, margins compress. Here, the opposite has broadly happened because customers are still capacity-constrained and Nvidia sells a system-level platform, not just a chip.
The only credible pushback is that hyperscalers are trying hard to reduce dependence. Google has TPUs, Amazon has Trainium, Microsoft has Maia, Meta has MTIA, and Broadcom is growing fast in custom accelerators. But those efforts mostly reduce Nvidia concentration in selected workloads. They do not yet remove Nvidia from the core AI infrastructure buying decision.
All things considered, Nvidia is still the market’s gravity center.
As this chart shows, and as featured in our AI chip market deck, search interest in AI chips has grown significantly
Are hyperscalers still buying chips like crazy now?
Yes. Hyperscalers are still buying aggressively, and the recent capex numbers are too large to treat as normal cloud expansion.
Microsoft’s April 2026 results gave one of the cleanest demand signals. Revenue reached $82.9 billion, Azure and other cloud services grew 40%, and Microsoft said its AI business had passed a $37 billion annual run-rate, up 123% year over year. That matters because it connects chip demand to paying cloud and enterprise workloads, not just experimental model training.
Meta’s April 2026 results were even more direct on infrastructure. The company raised its 2026 capex outlook to $125 billion to $145 billion, citing higher component pricing and data-center costs. That is a very concrete signal: Meta is not just buying more chips, it is accepting higher infrastructure costs to secure future capacity.
Amazon’s April 2026 results added the same pattern from another angle. AWS revenue grew 28% year over year to $37.6 billion, its fastest growth in several quarters, while external earnings-call summaries pointed to more than $40 billion of quarterly cash capex, mostly tied to AWS and generative AI infrastructure. Alphabet also raised its 2026 capex guidance to roughly $180 billion to $190 billion, with AI infrastructure and Google Cloud capacity as key drivers.
IDC’s infrastructure data gives useful context. AI infrastructure spending reached $318 billion for full-year 2025, more than double 2024, and Q4 alone reached $89.9 billion. TrendForce then forecast AI server shipments to grow more than 28% in 2026. These changes show the hardware cycle is still expanding from a very large base.
So, hyperscaler demand has not rolled over. The debate is whether the return on that capex will be good enough, not whether the purchasing wave is still there.
If you want more recent data on this point, please see our latest AI chip market report.
Are custom AI chips finally becoming a real market?
Yes. Custom AI chips have moved from “interesting internal projects” to a real revenue pool, especially for hyperscalers.
Broadcom is the strongest proof. In its June 2026 results, the company said AI semiconductor revenue grew 143% year over year to $10.8 billion, driven by custom AI accelerators and AI networking. It also guided AI semiconductor revenue to $16 billion next quarter, which would be more than 200% year-over-year growth. That is already bigger than many standalone semiconductor markets.
Marvell is a second useful signal because it sits closer to custom silicon, optical connectivity, and data infrastructure. Its fiscal 2026 revenue reached $8.195 billion, up 42% year over year, and management tied the growth to AI demand. Hyperscaler AI buildouts then create large spending pools around custom chips, connectivity, and cluster plumbing.
Google’s Ironwood TPU adds the product-level proof. Google has positioned Ironwood as its seventh-generation TPU for the inference era, available through Google Cloud rather than only as a hidden internal advantage. That shift matters because custom AI silicon is gradually becoming customer-facing cloud infrastructure.
The reason this is happening now is simple: once AI workloads become predictable, custom chips make economic sense. A hyperscaler does not need a perfectly general GPU for every task but, actually, the best cost, power, latency, and supply profile for its own workloads.
This chart, featured in our AI chip market deck, shows annual VC investment in AI chip startups
Is inference the part of AI chips everyone is chasing lately?
Yes. Inference is the freshest battleground in AI chips today.
Training still absorbs huge budgets, especially at the frontier. But the market’s incremental excitement is moving toward inference because model serving becomes a recurring cost every time users ask questions, generate images, run agents, or call AI APIs. Training is a project cost; inference is a usage tax.
We see this in product positioning. Google’s Ironwood TPU is explicitly framed around the “age of inference.” Groq’s recent fundraising reports emphasized its pivot and focus on inference. Positron’s $230 million round was also built around inference hardware. TensorWave’s AMD-based AI cloud is another adjacent signal: buyers want more ways to access AI compute without relying only on Nvidia-based capacity.
The demand signal also shows up in cloud results. Microsoft’s AI business run-rate reached $37 billion in April 2026, while Amazon said AWS growth accelerated to 28%. When cloud AI usage expands, the bottleneck shifts from “can we train the next model?” to “can we serve enough workloads at acceptable cost?”
This is why inference hardware is becoming more investable. A chip that lowers cost per token, improves latency, or uses less power can create value even without beating Nvidia on broad training benchmarks. That is a more specific and more realistic wedge.
So it looks like inference is where the AI chip market is becoming more economically interesting.
Are AI chips still supply-constrained now?
Yes. The AI chip shortage has not disappeared. Instead, it has moved into memory, packaging, power, and full-system capacity.
A year ago, people talked mostly about GPU availability. Today, the better market thermometer is HBM, advanced packaging, and data-center power. That is where the newest strain shows up.
SK hynix’s May 2026 results are one of the clearest memory signals. The company said first-quarter demand stayed strong despite normal seasonal weakness because AI infrastructure investment kept expanding. It also highlighted high-value products like HBM, high-capacity server DRAM, and enterprise SSDs, while operating profit nearly doubled quarter over quarter.
Gartner’s April 2026 semiconductor forecast made the memory point even sharper. It projected worldwide semiconductor revenue above $1.3 trillion in 2026, up 64%, with memory revenue tripling because of AI demand and higher memory prices. IDC separately forecast the semiconductor market at roughly $1.29 trillion in 2026, up 52.8%, led by AI infrastructure, memory, and hyperscaler capex.
TSMC’s April 2026 results confirm the same pressure at the foundry layer. Revenue rose 35.1% year over year, and net income rose 58.3%. External earnings-call summaries also emphasized that high-performance computing had crossed a very large share of TSMC’s revenue mix and that advanced-node demand remained heavily tied to AI accelerators.
The fact that more GPUs are shipping does not mean the market has normalized. HBM stacks, CoWoS-style advanced packaging, power availability, networking, and rack integration can each slow the final deployment. These constraints also protect pricing power for the suppliers that own scarce capacity.
AI chip supply is still tight. The bottleneck just looks less like a single chip shortage and more like a full infrastructure shortage.
If you want more recent data on this point, please see our latest AI chip market report.
This chart, featured in our AI chip market deck, shows how Nvidia is leading in AI chips
Is memory becoming the hidden winner in AI chips?
Yes. Memory is clearly becoming one of the most important hidden winners in the AI chip market.
This is one of the most under-discussed changes. Investors often focus on GPUs and accelerators because they are more visible. But the AI compute stack increasingly depends on high-bandwidth memory, server DRAM, enterprise SSDs, and the ability to package memory close to compute.
SK hynix’s first-quarter 2026 update showed this clearly. The company pointed to HBM and high-capacity server products as key drivers, and it said stable supply capability had become a core competitive advantage because customer demand exceeded capacity. That sentence is important: memory suppliers are no longer just cyclical commodity players in this part of the market. They are gatekeepers for AI system scale.
Gartner’s April 2026 forecast adds the order of magnitude. It expects memory revenue to triple in 2026 as AI demand and price inflation reshape semiconductor spending. That is a bigger shift than “memory is recovering from a downcycle.” It means AI is changing the mix of semiconductor profit pools.
Recent market reporting around Micron, Samsung, and SK hynix also points in the same direction: customers are reserving HBM supply far ahead, and memory capacity additions take time. Even when new fabs or lines are announced, they do not solve the near-term bottleneck quickly because HBM is complex to manufacture and package.
This makes memory one of the cleaner picks-and-shovels angles in the AI chip market today. The more models grow, the more memory bandwidth and capacity matter. Compute without memory is stranded performance.
Is the “Nvidia alternative” story actually working now?
Partly, but only when it is tied to a very specific buyer pain.
The broad “replace Nvidia” story is still weak. Nvidia’s scale, software ecosystem, networking, and systems integration remain too strong. But alternatives are becoming real in specific lanes: AMD-based AI clouds, hyperscaler custom ASICs, inference accelerators, TPUs, and networking-heavy architectures.
TensorWave is a useful fresh signal. The company reportedly raised $350 million in June 2026 at a $1.55 billion valuation, backed by AMD and Magnetar, while positioning itself as a non-Nvidia AI cloud using AMD hardware and ROCm. It also reportedly operated around 10,000 AMD GPUs across 14 megawatts and had leased 500 megawatts of future capacity. That is not just a chip benchmark story; it is a capacity story.
AMD’s own results support the same direction, but with limits. Its data-center revenue grew 57% year over year to $5.8 billion in Q1 2026, showing real traction. Still, compared with Nvidia’s $75.2 billion data-center quarter, AMD remains a second-source growth story rather than a market-control story.
Broadcom’s custom AI accelerator business is probably the strongest Nvidia-alternative signal because it does not try to win the same open GPU category. It helps hyperscalers build their own accelerators and networking for workloads they understand well. That is a more credible wedge than asking customers to rewrite everything for an unproven chip.
So the Nvidia alternative story is working, but only in precise forms. It works when the buyer wants supply diversity, lower inference cost, or workload-specific economics. It becomes much weaker when the pitch is simply “we are faster than Nvidia.”
If you want more recent data on this point, please see our latest AI chip market report.
This chart, featured in our AI chip market deck, shows annual funding in AI chip startups
Are AI chip startups still investable these days?
Yes, but only a smaller subset of AI chip startups still deserves attention.
The bar has moved. A few years ago, a startup could raise serious money with a clever accelerator architecture and a benchmark promise. Today, we need to see customer access, software maturity, capacity, energy economics, and a workload where the buyer’s pain is obvious.
The latest startup funding signals show where investors still care. Crunchbase reported roughly $10.7 billion of semiconductor startup funding already in 2026. Groq’s reported internal fundraising discussions, Positron’s $230 million Series B, TensorWave’s $350 million round, and continued attention around inference or alternative AI clouds all show capital is still available.
But the funding pattern is more disciplined than the headline suggests. The strongest rounds are tied to inference, AI cloud capacity, silicon photonics, memory-adjacent infrastructure, and non-Nvidia access. That is very different from funding every company that says “transformer chip.”
The commercial risk is still high. Chip startups face long development cycles, manufacturing risk, software friction, and brutal competition from Nvidia, AMD, Broadcom, Google, Amazon, and Microsoft. Even a technically good chip can lose if customers do not want to port workloads or trust the supply chain.
AI chip startups are still investable, but the good ones look less like science projects now. They look like infrastructure companies with a narrow wedge, serious capital access, and a clear path to deployed workloads.
Is China still a big growth engine for Western AI chip companies?
No. China is still a large market, but it is no longer a clean growth engine for Western AI chip companies.
The freshest policy signal came from the U.S. Bureau of Industry and Security in January 2026. BIS moved exports of Nvidia H200, AMD MI325X, and similar chips to China and Macau into case-by-case license review, subject to security requirements. That is less restrictive than a blanket denial, but it still makes demand uncertain.
The second signal is that China itself may not simply absorb every approved U.S. chip. Recent reporting around Nvidia H200 shipments suggested that Chinese authorities were discouraging or blocking some shipments even after U.S. clearance. That creates a strange two-sided constraint: U.S. policy limits what can go in, while Chinese policy can limit what customers are encouraged to buy.
Taiwan’s possible tightening of AI chip export enforcement adds another layer. Recent reports said Taiwan was considering stricter rules that could criminalize unauthorized shipments of AI chips or AI servers into China. Since Taiwan sits at the center of advanced chip manufacturing and AI server assembly, this would make rerouting more dangerous.
The result is that China demand cannot be modeled like normal hyperscaler demand anymore. It is too exposed to licensing rules, customs decisions, domestic substitution policy, and geopolitical retaliation.
If you want more recent data on this point, please see our latest AI chip market report.
This chart, featured in our AI chip market deck, compares the main business model options for AI accelerator chip companies
Is edge AI finally becoming the next big chip wave?
Not yet. Edge AI is real, but it is not the strongest part of the AI chip market right now.
There are real product-cycle signals in phones, PCs, cars, cameras, and industrial devices. Apple, Qualcomm, Intel, AMD, Nvidia, and many embedded-chip vendors all want more AI compute on-device. But compared with data-center AI, the recent revenue pull is weaker and less urgent.
The strongest recent market signals still come from data-center infrastructure. Nvidia’s data-center revenue, Broadcom’s AI semiconductor revenue, Microsoft’s AI run-rate, Meta’s capex raise, Amazon’s AWS acceleration, TSMC’s AI-linked foundry growth, and SK hynix’s HBM strength all point to centralized compute as the main driver today.
The reason is economic. Edge AI chips usually ride device refresh cycles, which are slower and more fragmented. Data-center AI chips are being pulled by immediate scarcity, cloud demand, model serving, and hyperscaler competition. That creates more pricing power and faster revenue conversion.
Edge AI may become a much bigger market later, especially if AI agents, local inference, privacy, and latency-sensitive applications become mainstream on devices. But currently, it is not the cleanest answer to “how is the AI chip market doing?”
For now, edge AI is a secondary opportunity. The hot center of the market is still data-center compute and everything that helps deploy it.
Is the AI chip trade getting too fragile now?
Yes. The AI chip market is operationally strong but financially fragile.
This is where we need to separate business reality from stock-market reaction. The business reality is strong: Nvidia, Broadcom, AMD, TSMC, SK hynix, Microsoft, Meta, Amazon, and Alphabet are all showing rising AI infrastructure demand in recent results. Across chips, memory, cloud, and capex, the signal is still expansion.
The financial reality is harsher. Broadcom’s June 2026 results were extremely strong by normal standards: record revenue, AI semiconductor revenue up 143%, and guidance for more than 200% year-over-year AI growth next quarter. Yet the stock still sold off because investors wanted even more certainty, cleaner guidance, or broader upside.
That tells us the market has moved into a more demanding phase. Investors are no longer asking, “Is AI chip demand real?” They are asking, “Is the next quarter even better than the already huge expectation?” That is a much more fragile setup.
Goldman Sachs’ recent comments about AI capex risk point in the same direction. The concern is not that hyperscalers are stopping. It is that spending may reach such extreme levels that investors start worrying about returns, financing, power constraints, and overbuilding.
Everything considered together, the AI chip market itself is healthy. The trade around it is fragile because expectations are now enormous.
This chart, featured in our AI chip market deck, shows how revenue is split across customer segments in the AI chip market
So, how is the AI chip market doing these days?
The AI chip market is doing very well right now, but the opportunity has become much more specific.
The evidence is strong enough to be direct. Recent results from Nvidia, Broadcom, AMD, TSMC, SK hynix, Microsoft, Meta, Amazon, and Alphabet all point in the same direction: AI infrastructure demand is still growing, hyperscalers are still spending, memory is tightening, custom ASICs are scaling, and inference is becoming the next major compute battleground.
But this is no longer a simple “AI chips go up” market. The best opportunities today sit where the system is still short: HBM, advanced packaging, AI networking, inference economics, power-aware systems, custom silicon, and alternative AI cloud capacity. The weaker opportunities are broad Nvidia-killer stories, China-dependent growth cases, and edge AI narratives without immediate revenue pull.
Bottom line: the AI chip market is still hot, but it is no longer broad and easy. Today, the money is in the constraints: memory bandwidth, packaging, networking, power, inference cost, and custom infrastructure for buyers large enough to shape their own silicon stack.
Here is the current market-temperature table.
| Check | Trend | Explanation |
|---|---|---|
| Investor appetite | Up | Money is still flowing into AI chips, with semiconductor startups already around $10.7 billion of 2026 funding according to Crunchbase. The money is more selective now, mostly favoring inference, non-Nvidia capacity, and infrastructure bottlenecks. |
| Nvidia position | Up | Nvidia remains the default AI infrastructure platform after reporting $75.2 billion of quarterly data-center revenue in May 2026. Competitors are growing, but none has closed the system-level gap yet. |
| Hyperscaler demand | Up | Microsoft, Meta, Amazon, and Alphabet all showed aggressive AI infrastructure demand in recent 2026 results. The capex numbers are large enough to confirm that buying has not rolled over. |
| Custom AI silicon | Up | Broadcom’s AI semiconductor revenue grew 143% year over year, while Marvell and Google’s TPU push confirm the custom-chip market is becoming real. This is one of the clearest recent changes in the market. |
| Inference chips | Up | Inference is now the freshest battleground because serving AI at scale turns compute into a recurring cost. Groq, Positron, Google Ironwood, TensorWave, and Broadcom all point toward this shift. |
| HBM and memory | Up | SK hynix, Gartner, and recent memory-market signals show that HBM and server memory are becoming core AI bottlenecks. This may be one of the best hidden winners in the market today. |
| Supply constraints | Up | The shortage has moved from GPUs alone into HBM, advanced packaging, power, and full-system deployment. That keeps pricing power alive across less obvious parts of the stack. |
| Nvidia alternatives | Mixed | AMD clouds, custom ASICs, TPUs, and inference accelerators are gaining traction. But the strongest alternatives are narrow wedges, not broad Nvidia replacements. |
| AI chip startups | Mixed | Funding is still active, but the bar is much higher. Startups need deployed workloads, software maturity, and clear cost advantages, not just benchmark claims. |
| China growth | Down | China remains large but too policy-dependent for Western vendors. U.S. licensing rules, Chinese policy signals, and possible Taiwan enforcement make this a risky demand pool. |
| Edge AI chips | Mixed | Edge AI is real but still secondary. The strongest recent revenue and capex signals remain in data-center AI infrastructure. |
| Stock-market setup | Mixed | The operating market is strong, but the trade is fragile. Broadcom’s selloff after very strong AI results shows expectations are now extremely hard to beat. |
OUR METHODOLOGY
This analysis tests how the AI chip market is doing today by looking across the full infrastructure stack rather than relying on one company, one stock chart, or one headline number.
The AI chip market is difficult to read from one headline number, because the answer depends on which part of the stack we look at. Nvidia’s results, hyperscaler capex, memory demand, custom silicon, inference, startup funding, China exposure, edge AI, and stock-market reactions can each point to a slightly different story.
So we broke the question into the main dimensions that shape the market today. For each one, we looked at recent signals rather than relying on intuition, broad market narratives, or one-company evidence. We prioritized fresh company results, capex guidance, startup funding, product launches, supply-chain indicators, policy changes, and market reactions.
We then aggregated those signals point by point. That matters because the AI chip market is not moving in one clean direction everywhere. Demand is still strong, Nvidia remains dominant, hyperscalers are still spending, memory is tightening, custom silicon is scaling, and inference is becoming more economically important. At the same time, China is less reliable as a Western growth engine, edge AI is still secondary, and the stock-market setup is more fragile than the operating market.
This structured aggregation is what makes the final answer clearer: the AI chip market is still hot, but the opportunity is now concentrated around the parts of the system where recent evidence shows scarcity, spending urgency, or improving economics.
Key sources used for this analysis include: Nvidia’s Q1 FY2027 financial results, Broadcom’s Q2 fiscal 2026 financial results, AMD’s Q1 2026 financial results, Microsoft’s Q3 FY2026 results, Meta’s Q1 2026 results, Amazon’s Q1 2026 results, Alphabet investor materials, TSMC’s Q1 2026 earnings release, SK hynix’s Q1 2026 business results, Gartner’s 2026 semiconductor revenue forecast, IDC’s AI infrastructure spending update, IDC’s 2026 semiconductor market forecast, TrendForce’s AI server shipment forecast, Google’s Ironwood TPU announcement, Google Cloud TPU7x documentation, Marvell’s fiscal 2026 financial results, Crunchbase on 2026 semiconductor startup funding, Positron’s Series B announcement, TechCrunch on Groq fundraising, Groq’s financing announcement, U.S. BIS on China semiconductor export licensing, and the Federal Register rule on advanced computing exports.
This chart, featured in our AI chip market deck, shows how AI accelerator chip technology has evolved over time
Related blog posts
- How strong is fundraising in the AI chip market right now?
- Which companies have raised the most funding in the AI chip market?
- Which companies are the most valued in the AI chip market?
